{"topic":{"id":15591,"title":"兴趣笔记","slug":"topic","categoryId":11,"createdAt":"2026-06-08 17:09:23.107979","bumpedAt":"2026-08-21T15:24:11.515Z","views":1899,"postsCount":44,"highestPostNumber":44,"likeCount":8,"closed":false,"pinned":false,"pinnedGlobally":false,"pinnedAt":null,"pinnedUntil":null,"isBanner":false,"excerpt":"我觉得有意思的学习内容以前我一直认为自己对知识的渴望是纯粹的，认知、建模是我所有的意义。应试的学习让我感到不屑，却又因这份借口而无需过多进行实践现在看来，对学习进步的需求无非来自三种追求：获得知识本身优越感逃避劣等的恐惧也许大部份人都有这些追求，只是不同的人和学习的内容，三种占比不同吧我曾经自以为的纯粹知识渴望，也不过很大程度是对劣等感的逃避，为了没有胆量承认的虚荣然而也许内心并不纯粹，追求俗不可耐，这样的自己才是真实的…","deletedAt":null,"isAnonymous":false,"category":{"id":11,"name":"深入交流","slug":"distilled","color":"c01c28"},"author":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif"},"tags":[]},"posts":[{"id":72255,"postNumber":1,"raw":"我觉得有意思的学习内容 :innocent_cat:\n\n以前我一直认为自己对知识的渴望是纯粹的，认知、建模是我所有的意义。应试的学习让我感到不屑，却又因这份借口而无需过多进行实践\n\n现在看来，对学习进步的需求无非来自三种追求：\n- 获得知识本身\n- 优越感\n- 逃避劣等的恐惧\n\n也许大部份人都有这些追求，只是不同的人和学习的内容，三种占比不同吧\n\n我曾经自以为的纯粹知识渴望，也不过很大程度是对劣等感的逃避，为了没有胆量承认的虚荣\n\n然而也许内心并不纯粹，追求俗不可耐，这样的自己才是真实的，而非想象中某种完美的幻影\n\n无论出发点如何，如果还有兴趣能让我感到幸福，那便足够了吧 :slightly_smiling_face:","cooked":"

我觉得有意思的学习内容 $\":innocent_cat:\"$

以前我一直认为自己对知识的渴望是纯粹的，认知、建模是我所有的意义。应试的学习让我感到不屑，却又因这份借口而无需过多进行实践

现在看来，对学习进步的需求无非来自三种追求：

获得知识本身
优越感
逃避劣等的恐惧

也许大部份人都有这些追求，只是不同的人和学习的内容，三种占比不同吧

我曾经自以为的纯粹知识渴望，也不过很大程度是对劣等感的逃避，为了没有胆量承认的虚荣

然而也许内心并不纯粹，追求俗不可耐，这样的自己才是真实的，而非想象中某种完美的幻影

无论出发点如何，如果还有兴趣能让我感到幸福，那便足够了吧 🙂

\n","createdAt":"2026-06-08 17:09:23.558407","updatedAt":"2026-06-08 18:28:44.591789","replyToPostNumber":null,"replyCount":0,"likeCount":6,"wordCount":258,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"wave_gif","count":1,"users":[{"id":5351,"username":"Xrystal-line","name":"羽流弦ノ迹"}],"hasReacted":false},{"reaction":"👍","count":1,"users":[{"id":7444,"username":"周末","name":"😀"}],"hasReacted":false},{"reaction":"❤️","count":6,"users":[{"id":2902,"username":"1313","name":"ᐕ)ﾉ"},{"id":5351,"username":"Xrystal-line","name":"羽流弦ノ迹"},{"id":5905,"username":"PipaQinse233","name":"PipaQinse233"},{"id":5940,"username":"water","name":null},{"id":7444,"username":"周末","name":"😀"},{"id":8040,"username":"dvm","name":"Dvm"}],"hasReacted":false}],"wiki":false,"version":2,"publicVersion":2,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72256,"postNumber":2,"raw":"# SplitArgs\n\n- 把字符串根据空格/引号分割成多个 token\n- 无引号 token 按照字面字符处理\n- 单引号 token 仅 unescape `''` -> `'`\n- 双引号 token 按照 YAML 双引号字符串 unescape\n\n## FSM\n![splitargs|690x343](/uploads/image/f6/e6/f6e6497bfa2b37038c5169dcf08b41904db763a3e02e8eda6c133b5bdd1f03da.avif)\n\n## Go 实现\n这样用闭包按照我的理解应该不会有性能问题？可惜因为需要 unescape 无法做到 zero-allocation\n```go\nconst (\n\tsplitArgsLookForToken int = iota\n\tsplitArgsInToken\n\tsplitArgsInSingleQuotes\n\tsplitArgsInDoubleQuotes\n\tsplitArgsSingleQuoteEscape\n\tsplitArgsDoubleQuoteEscape\n\tsplitArgsEndSuccess\n\tsplitArgsEndFail\n)\n\nfunc SplitArgs(s string) (result []string, err error) {\n\tvar sb strings.Builder\n\tvar current rune\n\tvar currentSize int\n\tvar errMsg string\n\tpeeked := false\n\teoi := false\n\tidx := 0\n\n\tstate := splitArgsLookForToken\n\n\tpeek := func() {\n\t\tif peeked {\n\t\t\treturn\n\t\t}\n\t\tpeeked = true\n\t\tif idx >= len(s) {\n\t\t\teoi = true\n\t\t} else {\n\t\t\tcurrent, currentSize = utf8.DecodeRuneInString(s[idx:])\n\t\t}\n\t}\n\n\tpeekHex := func(size int) error {\n\t\tif idx+size > len(s) {\n\t\t\treturn errors.New(\"early EOI at index \" + strconv.Itoa(idx))\n\t\t}\n\n\t\tr := rune(0)\n\n\t\tfor i := range size {\n\t\t\tc := s[idx+i]\n\t\t\tvar v rune\n\t\t\tswitch {\n\t\t\tcase c >= '0' && c <= '9':\n\t\t\t\tv = rune(c - '0')\n\t\t\tcase c >= 'a' && c <= 'f':\n\t\t\t\tv = rune(c - 'a' + 10)\n\t\t\tcase c >= 'A' && c <= 'F':\n\t\t\t\tv = rune(c - 'A' + 10)\n\t\t\tdefault:\n\t\t\t\treturn errors.New(\"bad hex digit at index \" + strconv.Itoa(idx+i))\n\t\t\t}\n\t\t\tr = (r << 4) | v\n\t\t}\n\n\t\tcurrent = r\n\t\tcurrentSize = size\n\t\treturn nil\n\t}\n\n\t// Must only be called after calling peek()\n\tpop := func() {\n\t\tidx += currentSize\n\t\tpeeked = false\n\t}\n\n\tbegin := func() {\n\t\tsb.Reset()\n\t}\n\n\tend := func() {\n\t\tresult = append(result, sb.String())\n\t}\n\n\t// Must only be called after calling peek() and before calling pop()\n\twrite := func() {\n\t\tsb.WriteRune(current)\n\t}\n\n\tfor {\n\t\tswitch state {\n\t\tcase splitArgsEndFail:\n\t\t\terr = errors.New(\"SplitArgs: \" + errMsg)\n\t\t\tfallthrough\n\t\tcase splitArgsEndSuccess:\n\t\t\treturn\n\t\tcase splitArgsLookForToken:\n\t\t\tpeek()\n\t\t\tif eoi {\n\t\t\t\tstate = splitArgsEndSuccess\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tswitch current {\n\t\t\tcase '\\t', '\\n', '\\v', '\\f', '\\r', ' ':\n\t\t\t\tpop()\n\t\t\t\tcontinue\n\t\t\tcase '\\'':\n\t\t\t\tpop()\n\t\t\t\tbegin()\n\t\t\t\tstate = splitArgsInSingleQuotes\n\t\t\t\tcontinue\n\t\t\tcase '\"':\n\t\t\t\tpop()\n\t\t\t\tbegin()\n\t\t\t\tstate = splitArgsInDoubleQuotes\n\t\t\t\tcontinue\n\t\t\tdefault:\n\t\t\t\tbegin()\n\t\t\t\tstate = splitArgsInToken\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase splitArgsInToken:\n\t\t\tpeek()\n\n\t\t\tif eoi {\n\t\t\t\tend()\n\t\t\t\tstate = splitArgsEndSuccess\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tswitch current {\n\t\t\tcase '\\t', '\\n', '\\v', '\\f', '\\r', ' ':\n\t\t\t\tpop()\n\t\t\t\tend()\n\t\t\t\tstate = splitArgsLookForToken\n\t\t\t\tcontinue\n\t\t\tdefault:\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase splitArgsInSingleQuotes:\n\t\t\tpeek()\n\n\t\t\tif eoi {\n\t\t\t\terrMsg = \"early EOI at index \" + strconv.Itoa(idx)\n\t\t\t\tstate = splitArgsEndFail\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tswitch current {\n\t\t\tcase '\\'':\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsSingleQuoteEscape\n\t\t\t\tcontinue\n\t\t\tdefault:\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase splitArgsInDoubleQuotes:\n\t\t\tpeek()\n\n\t\t\tif eoi {\n\t\t\t\terrMsg = \"early EOI at index \" + strconv.Itoa(idx)\n\t\t\t\tstate = splitArgsEndFail\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tswitch current {\n\t\t\tcase '\"':\n\t\t\t\tend()\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsLookForToken\n\t\t\t\tcontinue\n\t\t\tcase '\\\\':\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsDoubleQuoteEscape\n\t\t\t\tcontinue\n\t\t\tdefault:\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase splitArgsSingleQuoteEscape:\n\t\t\tpeek()\n\n\t\t\tif !eoi && current == '\\'' {\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsInSingleQuotes\n\t\t\t\tcontinue\n\t\t\t} else {\n\t\t\t\tend()\n\t\t\t\tstate = splitArgsLookForToken\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase splitArgsDoubleQuoteEscape:\n\t\t\tpeek()\n\n\t\t\tif eoi {\n\t\t\t\terrMsg = \"early EOI at index \" + strconv.Itoa(idx)\n\t\t\t\tstate = splitArgsEndFail\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tswitch current {\n\t\t\tcase '\"', '\\\\', '/', ' ', '\\x09':\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsInDoubleQuotes\n\t\t\t\tcontinue\n\t\t\tcase '0', 'a', 'b', 't', 'n', 'v', 'f', 'r', 'e', 'N', '_', 'L', 'P':\n\t\t\t\tswitch current {\n\t\t\t\tcase '0':\n\t\t\t\t\tcurrent = '\\x00'\n\t\t\t\tcase 'a':\n\t\t\t\t\tcurrent = '\\a'\n\t\t\t\tcase 'b':\n\t\t\t\t\tcurrent = '\\b'\n\t\t\t\tcase 't':\n\t\t\t\t\tcurrent = '\\t'\n\t\t\t\tcase 'n':\n\t\t\t\t\tcurrent = '\\n'\n\t\t\t\tcase 'v':\n\t\t\t\t\tcurrent = '\\v'\n\t\t\t\tcase 'f':\n\t\t\t\t\tcurrent = '\\f'\n\t\t\t\tcase 'r':\n\t\t\t\t\tcurrent = '\\r'\n\t\t\t\tcase 'e':\n\t\t\t\t\tcurrent = '\\x1b'\n\t\t\t\tcase 'N':\n\t\t\t\t\tcurrent = '\\x85'\n\t\t\t\tcase '_':\n\t\t\t\t\tcurrent = '\\xa0'\n\t\t\t\tcase 'L':\n\t\t\t\t\tcurrent = '\\u2028'\n\t\t\t\tcase 'P':\n\t\t\t\t\tcurrent = '\\u2029'\n\t\t\t\t}\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsInDoubleQuotes\n\t\t\t\tcontinue\n\t\t\tcase 'x', 'u', 'U':\n\t\t\t\tvar size int\n\t\t\t\tswitch current {\n\t\t\t\tcase 'x':\n\t\t\t\t\tsize = 2\n\t\t\t\tcase 'u':\n\t\t\t\t\tsize = 4\n\t\t\t\tcase 'U':\n\t\t\t\t\tsize = 8\n\t\t\t\t}\n\t\t\t\tpop()\n\t\t\t\tif err := peekHex(size); err != nil {\n\t\t\t\t\terrMsg = err.Error()\n\t\t\t\t\tstate = splitArgsEndFail\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsInDoubleQuotes\n\t\t\t\tcontinue\n\t\t\tdefault:\n\t\t\t\terrMsg = \"bad escape sequence at index \" + strconv.Itoa(idx)\n\t\t\t\tstate = splitArgsEndFail\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tdefault:\n\t\t\terrMsg = \"internal error\"\n\t\t\tstate = splitArgsEndFail\n\t\t\tcontinue\n\t\t}\n\t}\n}\n```\n\n感觉接口设计有点问题，似乎应该把 unescape 和 split 分开？ :thinking:\n\n~~才发现状态名写错了应该是 unescape~~\n\n## 方法论\n写这种复杂状态机之前，先用表格大致设计出状态和迁移会顺利很多\n\n![image|689x102](/uploads/image/b7/bf/b7bf0d630673b0678a464cd8242817b9b0e621d99b90f638a80f47f624c496fc.avif)\n\n\n我第一次写 IP Prefix trie 的时候完全梦到哪写到哪，哪里有问题就多加一个分支，最后变得非常莫名其妙 :grinning_face_with_sweat:","cooked":"

SplitArgs

把字符串根据空格/引号分割成多个 token
无引号 token 按照字面字符处理
单引号 token 仅 unescape '' -> '
双引号 token 按照 YAML 双引号字符串 unescape

FSM

$\"splitargs\"$

Go 实现

这样用闭包按照我的理解应该不会有性能问题？可惜因为需要 unescape 无法做到 zero-allocation

const (\n\tsplitArgsLookForToken int = iota\n\tsplitArgsInToken\n\tsplitArgsInSingleQuotes\n\tsplitArgsInDoubleQuotes\n\tsplitArgsSingleQuoteEscape\n\tsplitArgsDoubleQuoteEscape\n\tsplitArgsEndSuccess\n\tsplitArgsEndFail\n)\n\nfunc SplitArgs(s string) (result []string, err error) {\n\tvar sb strings.Builder\n\tvar current rune\n\tvar currentSize int\n\tvar errMsg string\n\tpeeked := false\n\teoi := false\n\tidx := 0\n\n\tstate := splitArgsLookForToken\n\n\tpeek := func() {\n\t\tif peeked {\n\t\t\treturn\n\t\t}\n\t\tpeeked = true\n\t\tif idx >= len(s) {\n\t\t\teoi = true\n\t\t} else {\n\t\t\tcurrent, currentSize = utf8.DecodeRuneInString(s[idx:])\n\t\t}\n\t}\n\n\tpeekHex := func(size int) error {\n\t\tif idx+size > len(s) {\n\t\t\treturn errors.New(\"early EOI at index \" + strconv.Itoa(idx))\n\t\t}\n\n\t\tr := rune(0)\n\n\t\tfor i := range size {\n\t\t\tc := s[idx+i]\n\t\t\tvar v rune\n\t\t\tswitch {\n\t\t\tcase c >= '0' && c <= '9':\n\t\t\t\tv = rune(c - '0')\n\t\t\tcase c >= 'a' && c <= 'f':\n\t\t\t\tv = rune(c - 'a' + 10)\n\t\t\tcase c >= 'A' && c <= 'F':\n\t\t\t\tv = rune(c - 'A' + 10)\n\t\t\tdefault:\n\t\t\t\treturn errors.New(\"bad hex digit at index \" + strconv.Itoa(idx+i))\n\t\t\t}\n\t\t\tr = (r << 4) | v\n\t\t}\n\n\t\tcurrent = r\n\t\tcurrentSize = size\n\t\treturn nil\n\t}\n\n\t// Must only be called after calling peek()\n\tpop := func() {\n\t\tidx += currentSize\n\t\tpeeked = false\n\t}\n\n\tbegin := func() {\n\t\tsb.Reset()\n\t}\n\n\tend := func() {\n\t\tresult = append(result, sb.String())\n\t}\n\n\t// Must only be called after calling peek() and before calling pop()\n\twrite := func() {\n\t\tsb.WriteRune(current)\n\t}\n\n\tfor {\n\t\tswitch state {\n\t\tcase splitArgsEndFail:\n\t\t\terr = errors.New(\"SplitArgs: \" + errMsg)\n\t\t\tfallthrough\n\t\tcase splitArgsEndSuccess:\n\t\t\treturn\n\t\tcase splitArgsLookForToken:\n\t\t\tpeek()\n\t\t\tif eoi {\n\t\t\t\tstate = splitArgsEndSuccess\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tswitch current {\n\t\t\tcase '\\t', '\\n', '\\v', '\\f', '\\r', ' ':\n\t\t\t\tpop()\n\t\t\t\tcontinue\n\t\t\tcase '\\'':\n\t\t\t\tpop()\n\t\t\t\tbegin()\n\t\t\t\tstate = splitArgsInSingleQuotes\n\t\t\t\tcontinue\n\t\t\tcase '\"':\n\t\t\t\tpop()\n\t\t\t\tbegin()\n\t\t\t\tstate = splitArgsInDoubleQuotes\n\t\t\t\tcontinue\n\t\t\tdefault:\n\t\t\t\tbegin()\n\t\t\t\tstate = splitArgsInToken\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase splitArgsInToken:\n\t\t\tpeek()\n\n\t\t\tif eoi {\n\t\t\t\tend()\n\t\t\t\tstate = splitArgsEndSuccess\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tswitch current {\n\t\t\tcase '\\t', '\\n', '\\v', '\\f', '\\r', ' ':\n\t\t\t\tpop()\n\t\t\t\tend()\n\t\t\t\tstate = splitArgsLookForToken\n\t\t\t\tcontinue\n\t\t\tdefault:\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase splitArgsInSingleQuotes:\n\t\t\tpeek()\n\n\t\t\tif eoi {\n\t\t\t\terrMsg = \"early EOI at index \" + strconv.Itoa(idx)\n\t\t\t\tstate = splitArgsEndFail\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tswitch current {\n\t\t\tcase '\\'':\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsSingleQuoteEscape\n\t\t\t\tcontinue\n\t\t\tdefault:\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase splitArgsInDoubleQuotes:\n\t\t\tpeek()\n\n\t\t\tif eoi {\n\t\t\t\terrMsg = \"early EOI at index \" + strconv.Itoa(idx)\n\t\t\t\tstate = splitArgsEndFail\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tswitch current {\n\t\t\tcase '\"':\n\t\t\t\tend()\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsLookForToken\n\t\t\t\tcontinue\n\t\t\tcase '\\\\':\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsDoubleQuoteEscape\n\t\t\t\tcontinue\n\t\t\tdefault:\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase splitArgsSingleQuoteEscape:\n\t\t\tpeek()\n\n\t\t\tif !eoi && current == '\\'' {\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsInSingleQuotes\n\t\t\t\tcontinue\n\t\t\t} else {\n\t\t\t\tend()\n\t\t\t\tstate = splitArgsLookForToken\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase splitArgsDoubleQuoteEscape:\n\t\t\tpeek()\n\n\t\t\tif eoi {\n\t\t\t\terrMsg = \"early EOI at index \" + strconv.Itoa(idx)\n\t\t\t\tstate = splitArgsEndFail\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tswitch current {\n\t\t\tcase '\"', '\\\\', '/', ' ', '\\x09':\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsInDoubleQuotes\n\t\t\t\tcontinue\n\t\t\tcase '0', 'a', 'b', 't', 'n', 'v', 'f', 'r', 'e', 'N', '_', 'L', 'P':\n\t\t\t\tswitch current {\n\t\t\t\tcase '0':\n\t\t\t\t\tcurrent = '\\x00'\n\t\t\t\tcase 'a':\n\t\t\t\t\tcurrent = '\\a'\n\t\t\t\tcase 'b':\n\t\t\t\t\tcurrent = '\\b'\n\t\t\t\tcase 't':\n\t\t\t\t\tcurrent = '\\t'\n\t\t\t\tcase 'n':\n\t\t\t\t\tcurrent = '\\n'\n\t\t\t\tcase 'v':\n\t\t\t\t\tcurrent = '\\v'\n\t\t\t\tcase 'f':\n\t\t\t\t\tcurrent = '\\f'\n\t\t\t\tcase 'r':\n\t\t\t\t\tcurrent = '\\r'\n\t\t\t\tcase 'e':\n\t\t\t\t\tcurrent = '\\x1b'\n\t\t\t\tcase 'N':\n\t\t\t\t\tcurrent = '\\x85'\n\t\t\t\tcase '_':\n\t\t\t\t\tcurrent = '\\xa0'\n\t\t\t\tcase 'L':\n\t\t\t\t\tcurrent = '\\u2028'\n\t\t\t\tcase 'P':\n\t\t\t\t\tcurrent = '\\u2029'\n\t\t\t\t}\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsInDoubleQuotes\n\t\t\t\tcontinue\n\t\t\tcase 'x', 'u', 'U':\n\t\t\t\tvar size int\n\t\t\t\tswitch current {\n\t\t\t\tcase 'x':\n\t\t\t\t\tsize = 2\n\t\t\t\tcase 'u':\n\t\t\t\t\tsize = 4\n\t\t\t\tcase 'U':\n\t\t\t\t\tsize = 8\n\t\t\t\t}\n\t\t\t\tpop()\n\t\t\t\tif err := peekHex(size); err != nil {\n\t\t\t\t\terrMsg = err.Error()\n\t\t\t\t\tstate = splitArgsEndFail\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\t\twrite()\n\t\t\t\tpop()\n\t\t\t\tstate = splitArgsInDoubleQuotes\n\t\t\t\tcontinue\n\t\t\tdefault:\n\t\t\t\terrMsg = \"bad escape sequence at index \" + strconv.Itoa(idx)\n\t\t\t\tstate = splitArgsEndFail\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tdefault:\n\t\t\terrMsg = \"internal error\"\n\t\t\tstate = splitArgsEndFail\n\t\t\tcontinue\n\t\t}\n\t}\n}\n

感觉接口设计有点问题，似乎应该把 unescape 和 split 分开？ 🤔

~~才发现状态名写错了应该是 unescape~~

方法论

写这种复杂状态机之前，先用表格大致设计出状态和迁移会顺利很多

$\"image\"$

我第一次写 IP Prefix trie 的时候完全梦到哪写到哪，哪里有问题就多加一个分支，最后变得非常莫名其妙 😅

\n","createdAt":"2026-06-08 17:22:36.155132","updatedAt":"2026-06-08 17:38:58.16724","replyToPostNumber":null,"replyCount":0,"likeCount":3,"wordCount":854,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"❤️","count":3,"users":[{"id":2902,"username":"1313","name":"ᐕ)ﾉ"},{"id":5940,"username":"water","name":null},{"id":8040,"username":"dvm","name":"Dvm"}],"hasReacted":false}],"wiki":false,"version":3,"publicVersion":3,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72272,"postNumber":3,"raw":"每过一两个月就觉得以前写的东西依托构使，恨不得狠狠 refactor 但是又没精力有没有懂的","cooked":"

每过一两个月就觉得以前写的东西依托构使，恨不得狠狠 refactor 但是又没精力有没有懂的

\n","createdAt":"2026-06-09 08:15:44.247144","updatedAt":"2026-06-09 08:15:44.247144","replyToPostNumber":null,"replyCount":0,"likeCount":1,"wordCount":37,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"plus1","count":1,"users":[{"id":2902,"username":"1313","name":"ᐕ)ﾉ"}],"hasReacted":false},{"reaction":"❤️","count":1,"users":[{"id":2902,"username":"1313","name":"ᐕ)ﾉ"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72412,"postNumber":4,"raw":"# Ring Buffer Read/Write\n---\n行为：\n- 写入/读取异步安全\n- 读取时，如果有效数据不足，仅读取有效数据并报错\n- 写入时，如果有效空间不足，仅写入有效空间并报错\n- 保留 1 字节区分完全空/完全满\n\n### 结构体\n```go\ntype Ring struct {\n\tbuffer []byte\n\n\twp atomic.Int64 // write pointer\n\t_ [64 - 24 - 8]byte\n\n\trp atomic.Int64 // read pointer\n\t_ [64 - 8]byte\n}\n```\n\n### 读取\n```go\nfunc (ring *Ring) Read(p []byte) (n int, err error) {\n\twp := int(ring.wp.Load()) // Might be stale but it's ok\n\trp := int(ring.rp.Load())\n\n\tdefer func() {\n\t\tring.rp.Store(int64(rp))\n\t}()\n\n\t// data size\n\tn = wp - rp\n\tif n < 0 {\n\t\tn += len(ring.buffer)\n\t}\n\n\tif n < len(p) {\n\t\terr = io.EOF // B, D\n\t} else {\n\t\tn = len(p) // A, C\n\t}\n\n\tif rem := n - copy(p[:n], ring.buffer[rp:]); rem > 0 {\n\t\tcopy(p[n-rem:], ring.buffer[:rem]) // C, D\n\t\trp = rem\n\t} else {\n\t\trp += n // A, B\n\t}\n\n\treturn\n}\n```\n状态变量：\n- EOF: `len(p)` 大于有效数据\n- Wrap: 读取跨越切片结尾\n\n| | Not EOF | EOF |\n| :---: | :---: | :---: |\n| **Not Wrap** | A |B |\n| **Wrap** | C|D |\n\n#### A\n```plaintext\n0 1 2 3 4 5 6 7\nx x o o #[x x]x\n ^ ^\n wp rp\n\nRead(p), len(p): 2\n\nn = 2 - 5 + 8 = 5 > 2: n = 2\nrem = 2 - copy(p[:2], buf[5:]) = 0\nrp += 2 = 7\n```\n\n#### B\n```plaintext\n0 1 2 3 4 5 6 7\no #[x x x]o o o\n ^ ^\n rp wp\n\nRead(p), len(p): 4\n\nn = 5 - 2 = 3 < 4: EOF\nrem = 3 - copy(p[:3], buf[2:]) = 0\nrp += 3 = 5\n```\n\n#### C\n```plaintext\n0 1 2 3 4 5 6 7\nx]x o o #[x x x\n ^ ^\n wp rp\n\nRead(p), len(p): 4\n\nn = 2 - 5 + 8 = 5 > 4: n = 4\nrem = 4 - copy(p[:4], buf[5:]) = 1\ncopy(p[4-1:], buf[:1])\nrp = 1\n```\n\n#### D\n```plaintext\n0 1 2 3 4 5 6 7\nx]x o o #[x x x\n ^ ^\n wp rp\n\nRead(p), len(p): 6\n\nn = 2 - 5 + 8 = 5 < 6: EOF\nrem = 5 - copy(p[:5], buf[5:]) = 2\ncopy(p[5-2:], buf[:2])\nrp = 2\n```\n\n### 写入\n```go\nfunc (ring *Ring) Write(p []byte) (n int, err error) {\n\trp := int(ring.rp.Load()) // Might be stale but it's ok\n\twp := int(ring.wp.Load())\n\n\tdefer func() {\n\t\tring.wp.Store(int64(wp))\n\t}()\n\n\t// free space\n\tn = rp - wp - 1\n\tif n < 0 {\n\t\tn += len(ring.buffer)\n\t}\n\n\tif n < len(p) {\n\t\terr = ErrBufferFull // B, D\n\t} else {\n\t\tn = len(p) // A, C\n\t}\n\n\tif rem := n - copy(ring.buffer[wp:], p[:n]); rem > 0 {\n\t\tcopy(ring.buffer[:rem], p[n-rem:]) // C, D\n\t\twp = rem\n\t} else {\n\t\twp += n // A, B\n\t}\n\n\treturn\n}\n```\n状态变量：\n- Full: `len(p)` 大于有效空间\n- Wrap: 写入跨越切片结尾\n\n| | Not Full | Full |\n| :---: | :---: | :---: |\n| **Not Wrap** | A |B |\n| **Wrap** | C|D |\n\n#### A\n```plaintext\n0 1 2 3 4 5 6 7\nx x[o o]# x x x\n ^ ^\n wp rp\n\nWrite(p), len(p): 2\n\nn = 5 - 2 - 1 = 2 >= 2: n = 2\nrem = 2 - copy(buf[2:], p[:2]) = 0\nwp += 2 = 4\n```\n\n#### B\n```plaintext\n0 1 2 3 4 5 6 7\nx x[o o]# x x x\n ^ ^\n wp rp\n\nWrite(p), len(p): 4\n\nn = 5 - 2 - 1 = 2 < 4: Full\nrem = 2 - copy(buf[2:], p[:2]) = 0\nwp += 2 = 4\n```\n\n#### C\n```plaintext\n0 1 2 3 4 5 6 7\no #[x x x]o o o\n ^ ^\n rp wp\n\nWrite(p), len(p): 4\n\nn = 2 - 5 - 1 + 8 = 4 >= 4: n = 4\nrem = 4 - copy(p[:4], buf[5:]) = 1\ncopy(buf[:1], p[4-1:])\nwp = 1\n```\n\n#### D\n```plaintext\n0 1 2 3 4 5 6 7\no #[x x x]o o o\n ^ ^\n rp wp\n\nWrite(p), len(p): 6\n\nn = 2 - 5 - 1 + 8 = 4 < 6: Full\nrem = 4 - copy(p[:4], buf[5:]) = 1\ncopy(buf[:1], p[4-1:])\nwp = 1\n```","cooked":"

Ring Buffer Read/Write

行为：

写入/读取异步安全
读取时，如果有效数据不足，仅读取有效数据并报错
写入时，如果有效空间不足，仅写入有效空间并报错
保留 1 字节区分完全空/完全满

结构体

type Ring struct {\n\tbuffer []byte\n\n\twp atomic.Int64 // write pointer\n\t_  [64 - 24 - 8]byte\n\n\trp atomic.Int64 // read pointer\n\t_  [64 - 8]byte\n}\n

读取

func (ring *Ring) Read(p []byte) (n int, err error) {\n\twp := int(ring.wp.Load()) // Might be stale but it's ok\n\trp := int(ring.rp.Load())\n\n\tdefer func() {\n\t\tring.rp.Store(int64(rp))\n\t}()\n\n\t// data size\n\tn = wp - rp\n\tif n < 0 {\n\t\tn += len(ring.buffer)\n\t}\n\n\tif n < len(p) {\n\t\terr = io.EOF // B, D\n\t} else {\n\t\tn = len(p) // A, C\n\t}\n\n\tif rem := n - copy(p[:n], ring.buffer[rp:]); rem > 0 {\n\t\tcopy(p[n-rem:], ring.buffer[:rem]) // C, D\n\t\trp = rem\n\t} else {\n\t\trp += n // A, B\n\t}\n\n\treturn\n}\n

状态变量：

EOF: len(p) 大于有效数据
Wrap: 读取跨越切片结尾

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

	Not EOF	EOF
Not Wrap	A	B
Wrap	C	D

A

0 1 2 3 4 5 6 7\nx x o o #[x x]x\n    ^     ^\n    wp    rp\n\nRead(p), len(p): 2\n\nn = 2 - 5 + 8 = 5 > 2: n = 2\nrem = 2 - copy(p[:2], buf[5:]) = 0\nrp += 2 = 7\n

B

0 1 2 3 4 5 6 7\no #[x x x]o o o\n    ^     ^\n    rp    wp\n\nRead(p), len(p): 4\n\nn = 5 - 2 = 3 < 4: EOF\nrem = 3 - copy(p[:3], buf[2:]) = 0\nrp += 3 = 5\n

C

0 1 2 3 4 5 6 7\nx]x o o #[x x x\n    ^     ^\n    wp    rp\n\nRead(p), len(p): 4\n\nn = 2 - 5 + 8 = 5 > 4: n = 4\nrem = 4 - copy(p[:4], buf[5:]) = 1\ncopy(p[4-1:], buf[:1])\nrp = 1\n

D

0 1 2 3 4 5 6 7\nx]x o o #[x x x\n    ^     ^\n    wp    rp\n\nRead(p), len(p): 6\n\nn = 2 - 5 + 8 = 5 < 6: EOF\nrem = 5 - copy(p[:5], buf[5:]) = 2\ncopy(p[5-2:], buf[:2])\nrp = 2\n

写入

func (ring *Ring) Write(p []byte) (n int, err error) {\n\trp := int(ring.rp.Load()) // Might be stale but it's ok\n\twp := int(ring.wp.Load())\n\n\tdefer func() {\n\t\tring.wp.Store(int64(wp))\n\t}()\n\n\t// free space\n\tn = rp - wp - 1\n\tif n < 0 {\n\t\tn += len(ring.buffer)\n\t}\n\n\tif n < len(p) {\n\t\terr = ErrBufferFull // B, D\n\t} else {\n\t\tn = len(p) // A, C\n\t}\n\n\tif rem := n - copy(ring.buffer[wp:], p[:n]); rem > 0 {\n\t\tcopy(ring.buffer[:rem], p[n-rem:]) // C, D\n\t\twp = rem\n\t} else {\n\t\twp += n // A, B\n\t}\n\n\treturn\n}\n

状态变量：

Full: len(p) 大于有效空间
Wrap: 写入跨越切片结尾

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

	Not Full	Full
Not Wrap	A	B
Wrap	C	D

A

0 1 2 3 4 5 6 7\nx x[o o]# x x x\n    ^     ^\n    wp    rp\n\nWrite(p), len(p): 2\n\nn = 5 - 2 - 1 = 2 >= 2: n = 2\nrem = 2 - copy(buf[2:], p[:2]) = 0\nwp += 2 = 4\n

B

0 1 2 3 4 5 6 7\nx x[o o]# x x x\n    ^     ^\n    wp    rp\n\nWrite(p), len(p): 4\n\nn = 5 - 2 - 1 = 2 < 4: Full\nrem = 2 - copy(buf[2:], p[:2]) = 0\nwp += 2 = 4\n

C

0 1 2 3 4 5 6 7\no #[x x x]o o o\n    ^     ^\n    rp    wp\n\nWrite(p), len(p): 4\n\nn = 2 - 5 - 1 + 8 = 4 >= 4: n = 4\nrem = 4 - copy(p[:4], buf[5:]) = 1\ncopy(buf[:1], p[4-1:])\nwp = 1\n

D

0 1 2 3 4 5 6 7\no #[x x x]o o o\n    ^     ^\n    rp    wp\n\nWrite(p), len(p): 6\n\nn = 2 - 5 - 1 + 8 = 4 < 6: Full\nrem = 4 - copy(p[:4], buf[5:]) = 1\ncopy(buf[:1], p[4-1:])\nwp = 1\n

\n","createdAt":"2026-06-23 15:12:39.082249","updatedAt":"2026-06-23 15:12:39.082249","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":830,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72423,"postNumber":5,"raw":"# ID Counter\n---\n```go\ntype IDCounter interface {\n\tGetID() uint64\n}\n```\n\n行为：\n- 异步安全\n- 有限时间内 ID 唯一\n\n---\n#### SimpleIDCounter\n```go\ntype SimpleIDCounter struct{ atomic.Uint64 }\n\nfunc (c *SimpleIDCounter) GetID() uint64 {\n\treturn c.Add(1)\n}\n```\n\n异步安全，`uint64` 可生成 `1 << 64` 个唯一 ID，高并发下由于锁竞争延迟线性提高\n\n---\n#### RandomShardedIDCounter\n```go\ntype shard struct {\n\tvalue atomic.Uint64\n\t_ [64 - 8]byte\n}\n\ntype RandomShardedIDCounter struct {\n\tshards []shard\n\tmask int\n\tshifts int\n}\n\nfunc NewRandomShardedIDCounter(n int) *RandomShardedIDCounter {\n\tif n <= 0 {\n\t\tn = 1\n\t}\n\n\tnumShards := 1\n\tshifts := 1\n\n\tfor numShards < n {\n\t\tnumShards <<= 1\n\t\tshifts++\n\t}\n\n\treturn &RandomShardedIDCounter{\n\t\tshards: make([]shard, numShards),\n\t\tmask: numShards - 1,\n\t\tshifts: shifts,\n\t}\n}\n\nfunc (c *RandomShardedIDCounter) GetID() uint64 {\n\tidx := rand.Int() & c.mask\n\n\tval := c.shards[idx].value.Add(1)\n\n\treturn (val << uint64(c.shifts)) | uint64(idx)\n}\n```\n异步安全，用 `shardIdx` 作为 ID 低位，每个 shard 可连续生成 `1 << (64 - bitWidth(shardIdx))` 个唯一 ID\n\n用随机 `shardIdx` 均分请求到各 shard 缓解锁竞争\n\n这里 `rand` 是 `math/rand/v2`，因为 `v1` 的全局函数是同步的\n\n---\n#### 跑分\n\n![benchplot|690x431](/uploads/image/64/8c/648c302645bf89e9a77cebd687243850dfba46220af229175ee1e3acf682f8dd.avif)\n\nSharded `2 <= GOMAXPROCS <= 4` 突然很慢是怎么回事有没有懂的？ :melt:\n`pprof` 我还不太能用明白","cooked":"

ID Counter

type IDCounter interface {\n\tGetID() uint64\n}\n

行为：

异步安全
有限时间内 ID 唯一

SimpleIDCounter

type SimpleIDCounter struct{ atomic.Uint64 }\n\nfunc (c *SimpleIDCounter) GetID() uint64 {\n\treturn c.Add(1)\n}\n

异步安全，uint64 可生成 1 << 64 个唯一 ID，高并发下由于锁竞争延迟线性提高

RandomShardedIDCounter

type shard struct {\n\tvalue atomic.Uint64\n\t_     [64 - 8]byte\n}\n\ntype RandomShardedIDCounter struct {\n\tshards []shard\n\tmask   int\n\tshifts int\n}\n\nfunc NewRandomShardedIDCounter(n int) *RandomShardedIDCounter {\n\tif n <= 0 {\n\t\tn = 1\n\t}\n\n\tnumShards := 1\n\tshifts := 1\n\n\tfor numShards < n {\n\t\tnumShards <<= 1\n\t\tshifts++\n\t}\n\n\treturn &RandomShardedIDCounter{\n\t\tshards: make([]shard, numShards),\n\t\tmask:   numShards - 1,\n\t\tshifts: shifts,\n\t}\n}\n\nfunc (c *RandomShardedIDCounter) GetID() uint64 {\n\tidx := rand.Int() & c.mask\n\n\tval := c.shards[idx].value.Add(1)\n\n\treturn (val << uint64(c.shifts)) | uint64(idx)\n}\n

异步安全，用 shardIdx 作为 ID 低位，每个 shard 可连续生成 1 << (64 - bitWidth(shardIdx)) 个唯一 ID

用随机 shardIdx 均分请求到各 shard 缓解锁竞争

这里 rand 是 math/rand/v2，因为 v1 的全局函数是同步的

跑分

$\"benchplot\"$

Sharded 2 <= GOMAXPROCS <= 4 突然很慢是怎么回事有没有懂的？ 🫠
\npprof 我还不太能用明白

\n","createdAt":"2026-06-26 14:14:51.103626","updatedAt":"2026-06-26 14:20:48.753541","replyToPostNumber":null,"replyCount":1,"likeCount":0,"wordCount":269,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":2,"publicVersion":2,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72424,"postNumber":6,"raw":"```plaintext\n Total: 2.23s 2.33s (flat, cum) 98.73%\n 60 . . \t\tmask: numShards - 1, \n 61 . . \t\tshifts: shifts, \n 62 . . \t} \n 63 . . } \n 64 . . \n 65 10ms 10ms func (c *RandomShardedIDCounter) GetID() uint64 { \n 66 60ms 160ms \tidx := rand.Int() & c.mask \n 67 . . \n 68 60ms 60ms \tval := c.shards[idx].value.Add(1) \n 69 . . \n 70 2.10s 2.10s \treturn (val << uint64(c.shifts)) | uint64(idx) \n 71 . . } \n```\n何意味","cooked":"

  Total:       2.23s      2.33s (flat, cum) 98.73%\n     60            .          .           \t\tmask:   numShards - 1, \n     61            .          .           \t\tshifts: shifts, \n     62            .          .           \t} \n     63            .          .           } \n     64            .          .            \n     65         10ms       10ms           func (c *RandomShardedIDCounter) GetID() uint64 { \n     66         60ms      160ms           \tidx := rand.Int() & c.mask \n     67            .          .            \n     68         60ms       60ms           \tval := c.shards[idx].value.Add(1) \n     69            .          .            \n     70        2.10s      2.10s           \treturn (val << uint64(c.shifts)) | uint64(idx) \n     71            .          .           } \n

何意味

\n","createdAt":"2026-06-26 14:29:49.219966","updatedAt":"2026-06-26 14:29:49.219966","replyToPostNumber":5,"replyCount":0,"likeCount":0,"wordCount":76,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72562,"postNumber":7,"raw":"

\r\n签名的 Anon 酱老是被 GC 掉，在这里放一个也许就不会了吧\r\n\r\nBtw 上传 webp 似乎有问题，转换后的 avif 图片无法正确显示","cooked":"

\n签名的 Anon 酱老是被 GC 掉，在这里放一个也许就不会了吧\n

Btw 上传 webp 似乎有问题，转换后的 avif 图片无法正确显示

\n","createdAt":"2026-07-13T15:33:17.724Z","updatedAt":"2026-07-13T15:37:39.259Z","replyToPostNumber":null,"replyCount":1,"likeCount":0,"wordCount":51,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"cat_rabbit","count":1,"users":[{"id":30,"username":"一只玉米人","name":"玉米🌽"}],"hasReacted":false},{"reaction":"❤️","count":1,"users":[{"id":30,"username":"一只玉米人","name":"玉米🌽"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72563,"postNumber":8,"raw":"我这边可以正常显示 :heart_eyes:","cooked":"

我这边可以正常显示 😍

\n","createdAt":"2026-07-13T15:56:17.079Z","updatedAt":"2026-07-13T15:56:17.079Z","replyToPostNumber":7,"replyCount":1,"likeCount":0,"wordCount":10,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"plus1","count":1,"users":[{"id":4668,"username":"IO","name":null}],"hasReacted":false},{"reaction":"wave_gif","count":1,"users":[{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":30,"username":"一只玉米人","name":"玉米🌽","avatarUrl":"/uploads/blob/14/96/14961aeaa46ccbb75a61eec0c27b26e74fb5d1c00ccdcbbe898ca9ea37087dd6.svg","admin":false,"moderator":true,"signatureRaw":"*年月擦身过，暂且问，会更好吗？*","signatureCooked":"

年月擦身过，暂且问，会更好吗？

\n"}},{"id":72564,"postNumber":9,"raw":"# Log file change watcher\r\n## 接口\r\n```go\r\ntype Change uint8\r\n\r\nconst (\r\n\tChangeCreated Change = iota\r\n\tChangeDeleted\r\n\tChangeMoved\r\n\tChangeModified\r\n\tChangeTruncated\r\n)\r\n\r\ntype Watcher interface {\r\n\tBlockUntilChange(ctx context.Context) (Change, error)\r\n}\r\n```\r\n\r\n## Polling 实现\r\n```go\r\npackage tail\r\n\r\nimport (\r\n\t\"context\"\r\n\t\"os\"\r\n\t\"time\"\r\n)\r\n\r\nvar _ Watcher = (*PollWatcher)(nil)\r\n\r\ntype PollWatcher struct {\r\n\tinterval time.Duration\r\n\tfilepath string\r\n\tfi os.FileInfo\r\n\tsize int64\r\n\tmodified time.Time\r\n}\r\n\r\nfunc NewPollWatcher(path string, interval time.Duration) (*PollWatcher, error) {\r\n\tfi, err := os.Stat(path)\r\n\tif err != nil {\r\n\t\tif os.IsNotExist(err) {\r\n\t\t\tfi = nil\r\n\t\t} else {\r\n\t\t\treturn nil, err\r\n\t\t}\r\n\t}\r\n\r\n\tpw := &PollWatcher{\r\n\t\tfilepath: path,\r\n\t\tinterval: interval,\r\n\t\tfi: fi,\r\n\t}\r\n\r\n\tif fi != nil {\r\n\t\tpw.size = fi.Size()\r\n\t\tpw.modified = fi.ModTime()\r\n\t}\r\n\r\n\treturn pw, nil\r\n}\r\n\r\nfunc (pw *PollWatcher) BlockUntilChange(ctx context.Context) (Change, error) {\r\n\tticker := time.NewTicker(pw.interval)\r\n\tdefer ticker.Stop()\r\n\r\n\tfor {\r\n\t\tselect {\r\n\t\tcase <-ctx.Done():\r\n\t\t\treturn 0, ctx.Err()\r\n\t\tcase <-ticker.C:\r\n\t\t\tfi, err := os.Stat(pw.filepath)\r\n\t\t\tif err != nil {\r\n\t\t\t\tif os.IsNotExist(err) {\r\n\t\t\t\t\tif pw.fi != nil {\r\n\t\t\t\t\t\tpw.fi = nil\r\n\t\t\t\t\t\treturn ChangeDeleted, nil\r\n\t\t\t\t\t}\r\n\t\t\t\t\tcontinue\r\n\t\t\t\t}\r\n\t\t\t\treturn 0, err\r\n\t\t\t}\r\n\r\n\t\t\tlastSize := pw.size\r\n\t\t\tlastModified := pw.modified\r\n\t\t\tlastFi := pw.fi\r\n\t\t\tpw.size = fi.Size()\r\n\t\t\tpw.modified = fi.ModTime()\r\n\t\t\tpw.fi = fi\r\n\r\n\t\t\tif lastFi == nil {\r\n\t\t\t\treturn ChangeCreated, nil\r\n\t\t\t}\r\n\r\n\t\t\tif !os.SameFile(lastFi, fi) {\r\n\t\t\t\treturn ChangeMoved, nil\r\n\t\t\t}\r\n\r\n\t\t\tif pw.size < lastSize {\r\n\t\t\t\treturn ChangeTruncated, nil\r\n\t\t\t}\r\n\r\n\t\t\tif pw.size > lastSize || !pw.modified.Equal(lastModified) {\r\n\t\t\t\treturn ChangeModified, nil\r\n\t\t\t}\r\n\t\t}\r\n\t}\r\n}\r\n```\r\n| 情况 | 判定 |\r\n|----------|-----------|\r\n| **正常 append** | `pw.size > lastSize` -> `ChangeModified` |\r\n| **Log rotate (重命名 + 新建)** | `!os.SameFile(lastFi, fi)` -> `ChangeMoved` |\r\n| **Log rotate (复制 + 截短)** | `pw.size < lastSize` -> `ChangeTruncated` |\r\n| **新建** | `lastFi == nil` -> `ChangeCreated` |\r\n| **删除** | `pw.fi != nil` + `os.IsNotExist` -> `ChangeDeleted` |","cooked":"

Log file change watcher

接口

type Change uint8\n\nconst (\n\tChangeCreated Change = iota\n\tChangeDeleted\n\tChangeMoved\n\tChangeModified\n\tChangeTruncated\n)\n\ntype Watcher interface {\n\tBlockUntilChange(ctx context.Context) (Change, error)\n}\n

Polling 实现

package tail\n\nimport (\n\t\"context\"\n\t\"os\"\n\t\"time\"\n)\n\nvar _ Watcher = (*PollWatcher)(nil)\n\ntype PollWatcher struct {\n\tinterval time.Duration\n\tfilepath string\n\tfi       os.FileInfo\n\tsize     int64\n\tmodified time.Time\n}\n\nfunc NewPollWatcher(path string, interval time.Duration) (*PollWatcher, error) {\n\tfi, err := os.Stat(path)\n\tif err != nil {\n\t\tif os.IsNotExist(err) {\n\t\t\tfi = nil\n\t\t} else {\n\t\t\treturn nil, err\n\t\t}\n\t}\n\n\tpw := &PollWatcher{\n\t\tfilepath: path,\n\t\tinterval: interval,\n\t\tfi:       fi,\n\t}\n\n\tif fi != nil {\n\t\tpw.size = fi.Size()\n\t\tpw.modified = fi.ModTime()\n\t}\n\n\treturn pw, nil\n}\n\nfunc (pw *PollWatcher) BlockUntilChange(ctx context.Context) (Change, error) {\n\tticker := time.NewTicker(pw.interval)\n\tdefer ticker.Stop()\n\n\tfor {\n\t\tselect {\n\t\tcase <-ctx.Done():\n\t\t\treturn 0, ctx.Err()\n\t\tcase <-ticker.C:\n\t\t\tfi, err := os.Stat(pw.filepath)\n\t\t\tif err != nil {\n\t\t\t\tif os.IsNotExist(err) {\n\t\t\t\t\tif pw.fi != nil {\n\t\t\t\t\t\tpw.fi = nil\n\t\t\t\t\t\treturn ChangeDeleted, nil\n\t\t\t\t\t}\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\treturn 0, err\n\t\t\t}\n\n\t\t\tlastSize := pw.size\n\t\t\tlastModified := pw.modified\n\t\t\tlastFi := pw.fi\n\t\t\tpw.size = fi.Size()\n\t\t\tpw.modified = fi.ModTime()\n\t\t\tpw.fi = fi\n\n\t\t\tif lastFi == nil {\n\t\t\t\treturn ChangeCreated, nil\n\t\t\t}\n\n\t\t\tif !os.SameFile(lastFi, fi) {\n\t\t\t\treturn ChangeMoved, nil\n\t\t\t}\n\n\t\t\tif pw.size < lastSize {\n\t\t\t\treturn ChangeTruncated, nil\n\t\t\t}\n\n\t\t\tif pw.size > lastSize || !pw.modified.Equal(lastModified) {\n\t\t\t\treturn ChangeModified, nil\n\t\t\t}\n\t\t}\n\t}\n}\n

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

情况	判定
正常 append	`pw.size > lastSize` -> `ChangeModified`
Log rotate (重命名 + 新建)	`!os.SameFile(lastFi, fi)` -> `ChangeMoved`
Log rotate (复制 + 截短)	`pw.size < lastSize` -> `ChangeTruncated`
新建	`lastFi == nil` -> `ChangeCreated`
删除	`pw.fi != nil` + `os.IsNotExist` -> `ChangeDeleted`

\n","createdAt":"2026-07-13T16:02:14.770Z","updatedAt":"2026-07-13T17:52:06.722Z","replyToPostNumber":null,"replyCount":2,"likeCount":0,"wordCount":316,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":3,"publicVersion":3,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72565,"postNumber":10,"raw":"![anon.webp|375x500](/uploads/image/ad/c8/adc89f2248f451f288966812fb69ec6cffcaa801066524f9d129da91a0e613f2.avif)\r\n刚才那是 png，这个能显示吗？","cooked":"

$\"anon.webp\"$
\n刚才那是 png，这个能显示吗？

\n","createdAt":"2026-07-13T16:03:03.857Z","updatedAt":"2026-07-13T16:03:03.857Z","replyToPostNumber":8,"replyCount":1,"likeCount":0,"wordCount":13,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72566,"postNumber":11,"raw":"[quote=\"stateless_trash_can, post:9, topic:15591\"]\r\n**Log rotate (复制 + 截短)**\r\n`pw.size < lastSize` -> `ChangeTruncated`\r\n[/quote]\r\n\r\n这里如果在一次迭代期间，截短并写入超过原来文件大小的内容，会误判为 `ChangeModified`，暂时想不到怎么办 :melt: （不过这种情况对于日志来说不太可能？）\r\n\r\n后果是丢失一部分行","cooked":"

\n\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\tstateless_trash_can:\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t

\n
Log rotate (复制 + 截短)
\npw.size < lastSize -> ChangeTruncated
\n

\n\t\t\t\t\t\t

这里如果在一次迭代期间，截短并写入超过原来文件大小的内容，会误判为 ChangeModified，暂时想不到怎么办 🫠 （不过这种情况对于日志来说不太可能？）

后果是丢失一部分行

\n","createdAt":"2026-07-13T16:16:20.934Z","updatedAt":"2026-07-13T17:44:51.943Z","replyToPostNumber":9,"replyCount":0,"likeCount":0,"wordCount":88,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":2,"publicVersion":2,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72568,"postNumber":12,"raw":"## FSNotify 实现\r\n```go\r\npackage tail\r\n\r\nimport (\r\n\t\"context\"\r\n\t\"errors\"\r\n\t\"fmt\"\r\n\t\"os\"\r\n\t\"path/filepath\"\r\n\r\n\t\"github.com/fsnotify/fsnotify\"\r\n)\r\n\r\nvar _ Watcher = (*FSNotifyWatcher)(nil)\r\n\r\ntype FSNotifyWatcher struct {\r\n\twatcher *fsnotify.Watcher\r\n\tfilepath string\r\n\tsize int64\r\n\twasPresent bool\r\n}\r\n\r\nfunc NewFSNotifyWatcher(path string) (*FSNotifyWatcher, error) {\r\n\twatcher, err := fsnotify.NewWatcher()\r\n\tif err != nil {\r\n\t\treturn nil, err\r\n\t}\r\n\r\n\tcleanPath := filepath.Clean(path)\r\n\tparentDir := filepath.Dir(cleanPath)\r\n\r\n\t// Watch parent and filter by Event.Name\r\n\terr = watcher.Add(parentDir)\r\n\tif err != nil {\r\n\t\twatcher.Close()\r\n\t\treturn nil, err\r\n\t}\r\n\r\n\t// Get initial size and presence\r\n\tvar size int64\r\n\twasPresent := false\r\n\tfi, err := os.Stat(path)\r\n\tif err == nil {\r\n\t\tsize = fi.Size()\r\n\t\twasPresent = true\r\n\t}\r\n\r\n\treturn &FSNotifyWatcher{\r\n\t\twatcher: watcher,\r\n\t\tfilepath: cleanPath,\r\n\t\tsize: size,\r\n\t\twasPresent: wasPresent,\r\n\t}, nil\r\n}\r\n\r\nfunc (fw *FSNotifyWatcher) Close() error {\r\n\treturn fw.watcher.Close()\r\n}\r\n\r\nfunc (fw *FSNotifyWatcher) BlockUntilChange(ctx context.Context) (Change, error) {\r\n\tfor {\r\n\t\tselect {\r\n\t\tcase <-ctx.Done():\r\n\t\t\treturn 0, ctx.Err()\r\n\r\n\t\tcase err, ok := <-fw.watcher.Errors:\r\n\t\t\tif !ok {\r\n\t\t\t\treturn 0, errors.New(\"watcher channel closed\")\r\n\t\t\t}\r\n\t\t\treturn 0, err\r\n\r\n\t\tcase event, ok := <-fw.watcher.Events:\r\n\t\t\tif !ok {\r\n\t\t\t\treturn 0, errors.New(\"watcher channel closed\")\r\n\t\t\t}\r\n\r\n\t\t\tif filepath.Clean(event.Name) != fw.filepath {\r\n\t\t\t\tcontinue\r\n\t\t\t}\r\n\r\n\t\t\tswitch {\r\n\t\t\tcase event.Has(fsnotify.Create):\r\n\t\t\t\tif fw.wasPresent {\r\n\t\t\t\t\tif fi, err := os.Stat(fw.filepath); err == nil {\r\n\t\t\t\t\t\tfw.size = fi.Size()\r\n\t\t\t\t\t}\r\n\t\t\t\t\tcontinue\r\n\t\t\t\t}\r\n\t\t\t\tfw.wasPresent = true\r\n\t\t\t\treturn ChangeCreated, nil\r\n\r\n\t\t\tcase event.Has(fsnotify.Write):\r\n\t\t\t\tif fi, err := os.Stat(fw.filepath); err == nil {\r\n\t\t\t\t\tlastSize := fw.size\r\n\t\t\t\t\tfw.size = fi.Size()\r\n\r\n\t\t\t\t\tif fw.size < lastSize {\r\n\t\t\t\t\t\treturn ChangeTruncated, nil\r\n\t\t\t\t\t} else {\r\n\t\t\t\t\t\treturn ChangeModified, nil\r\n\t\t\t\t\t}\r\n\t\t\t\t} else {\r\n\t\t\t\t\treturn 0, fmt.Errorf(\"os.Stat failed: %w\", err)\r\n\t\t\t\t}\r\n\r\n\t\t\tcase event.Has(fsnotify.Remove):\r\n\t\t\t\tfw.wasPresent = false\r\n\t\t\t\treturn ChangeDeleted, nil\r\n\r\n\t\t\tcase event.Has(fsnotify.Rename):\r\n\t\t\t\treturn ChangeMoved, nil\r\n\t\t\t}\r\n\t\t}\r\n\t}\r\n}\r\n```","cooked":"

FSNotify 实现

package tail\n\nimport (\n\t\"context\"\n\t\"errors\"\n\t\"fmt\"\n\t\"os\"\n\t\"path/filepath\"\n\n\t\"github.com/fsnotify/fsnotify\"\n)\n\nvar _ Watcher = (*FSNotifyWatcher)(nil)\n\ntype FSNotifyWatcher struct {\n\twatcher    *fsnotify.Watcher\n\tfilepath   string\n\tsize       int64\n\twasPresent bool\n}\n\nfunc NewFSNotifyWatcher(path string) (*FSNotifyWatcher, error) {\n\twatcher, err := fsnotify.NewWatcher()\n\tif err != nil {\n\t\treturn nil, err\n\t}\n\n\tcleanPath := filepath.Clean(path)\n\tparentDir := filepath.Dir(cleanPath)\n\n\t// Watch parent and filter by Event.Name\n\terr = watcher.Add(parentDir)\n\tif err != nil {\n\t\twatcher.Close()\n\t\treturn nil, err\n\t}\n\n\t// Get initial size and presence\n\tvar size int64\n\twasPresent := false\n\tfi, err := os.Stat(path)\n\tif err == nil {\n\t\tsize = fi.Size()\n\t\twasPresent = true\n\t}\n\n\treturn &FSNotifyWatcher{\n\t\twatcher:    watcher,\n\t\tfilepath:   cleanPath,\n\t\tsize:       size,\n\t\twasPresent: wasPresent,\n\t}, nil\n}\n\nfunc (fw *FSNotifyWatcher) Close() error {\n\treturn fw.watcher.Close()\n}\n\nfunc (fw *FSNotifyWatcher) BlockUntilChange(ctx context.Context) (Change, error) {\n\tfor {\n\t\tselect {\n\t\tcase <-ctx.Done():\n\t\t\treturn 0, ctx.Err()\n\n\t\tcase err, ok := <-fw.watcher.Errors:\n\t\t\tif !ok {\n\t\t\t\treturn 0, errors.New(\"watcher channel closed\")\n\t\t\t}\n\t\t\treturn 0, err\n\n\t\tcase event, ok := <-fw.watcher.Events:\n\t\t\tif !ok {\n\t\t\t\treturn 0, errors.New(\"watcher channel closed\")\n\t\t\t}\n\n\t\t\tif filepath.Clean(event.Name) != fw.filepath {\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\tswitch {\n\t\t\tcase event.Has(fsnotify.Create):\n\t\t\t\tif fw.wasPresent {\n\t\t\t\t\tif fi, err := os.Stat(fw.filepath); err == nil {\n\t\t\t\t\t\tfw.size = fi.Size()\n\t\t\t\t\t}\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\tfw.wasPresent = true\n\t\t\t\treturn ChangeCreated, nil\n\n\t\t\tcase event.Has(fsnotify.Write):\n\t\t\t\tif fi, err := os.Stat(fw.filepath); err == nil {\n\t\t\t\t\tlastSize := fw.size\n\t\t\t\t\tfw.size = fi.Size()\n\n\t\t\t\t\tif fw.size < lastSize {\n\t\t\t\t\t\treturn ChangeTruncated, nil\n\t\t\t\t\t} else {\n\t\t\t\t\t\treturn ChangeModified, nil\n\t\t\t\t\t}\n\t\t\t\t} else {\n\t\t\t\t\treturn 0, fmt.Errorf(\"os.Stat failed: %w\", err)\n\t\t\t\t}\n\n\t\t\tcase event.Has(fsnotify.Remove):\n\t\t\t\tfw.wasPresent = false\n\t\t\t\treturn ChangeDeleted, nil\n\n\t\t\tcase event.Has(fsnotify.Rename):\n\t\t\t\treturn ChangeMoved, nil\n\t\t\t}\n\t\t}\n\t}\n}\n

\n","createdAt":"2026-07-13T17:51:29.141Z","updatedAt":"2026-07-13T18:56:07.677Z","replyToPostNumber":9,"replyCount":0,"likeCount":0,"wordCount":272,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":2,"publicVersion":2,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72569,"postNumber":13,"raw":"能正常显示。所有的图片除了svg之外，不管是什么格式，都会被[ImageMagick](https://imagemagick.org/)转换为AVIF格式。如果没法正常显示，可能是浏览器不支持： https://caniuse.com/avif\r\n\r\n其实普遍的做法是根据不同的浏览器UserAgent，服务器发送支持的格式。\r\n\r\n另外AVIF虽然压缩率比较高，浏览器广泛支持，但也存在一些问题：\r\n* 压缩/转码成avif较慢。\r\n* 不支持progrssive decoding，导致没法在网络较慢时像jpg一样一行一行地加载，只能全部下载后才能加载。\r\n\r\n比较理想的是jpeg-xl，支持progressive decoding, encoding较快。\r\n但是目前仅支持最新版Chrome/Firefox并需要启用experimental setting：\r\n* Chrome： \r\n访问 chrome://flags/\r\nenable-jxl-image-format \r\n* Firefox： \r\n访问 about:config\r\nimage.jxl.enabled","cooked":"

能正常显示。所有的图片除了svg之外，不管是什么格式，都会被ImageMagick转换为AVIF格式。如果没法正常显示，可能是浏览器不支持： AVIF image format | Can I use... Support tables for HTML5, CSS3, etc

其实普遍的做法是根据不同的浏览器UserAgent，服务器发送支持的格式。

另外AVIF虽然压缩率比较高，浏览器广泛支持，但也存在一些问题：

压缩/转码成avif较慢。
不支持progrssive decoding，导致没法在网络较慢时像jpg一样一行一行地加载，只能全部下载后才能加载。

比较理想的是jpeg-xl，支持progressive decoding, encoding较快。
\n但是目前仅支持最新版Chrome/Firefox并需要启用experimental setting：

Chrome：
\n访问 chrome://flags/
\nenable-jxl-image-format
Firefox：
\n访问 about:config
\nimage.jxl.enabled

\n","createdAt":"2026-07-14T01:19:10.979Z","updatedAt":"2026-07-14T01:19:49.516Z","replyToPostNumber":10,"replyCount":3,"likeCount":0,"wordCount":209,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"👍","count":1,"users":[{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":4668,"username":"IO","name":null,"avatarUrl":"/uploads/image/86/f2/86f267c0aaa73835a999b7cc3a7e7a701605574eda9a2c2a56f5ddbbba64d1a9.avif","admin":true,"moderator":false,"signatureRaw":"","signatureCooked":null}},{"id":72570,"postNumber":14,"raw":"现在可以了，之前疑似没加载出来\r\n\r\n不过之前确实上传 webp 无法立即显示，但是 png 可以\r\n\r\n现在无法复现了\r\n\r\nEdit: 编辑器内上传图片无法立即加载，手动在新标签页访问图片路径后就可以在编辑器内显示了，不知道是不是网络问题","cooked":"

现在可以了，之前疑似没加载出来

不过之前确实上传 webp 无法立即显示，但是 png 可以

现在无法复现了

Edit: 编辑器内上传图片无法立即加载，手动在新标签页访问图片路径后就可以在编辑器内显示了，不知道是不是网络问题

\n","createdAt":"2026-07-14T04:16:00.504Z","updatedAt":"2026-07-14T04:31:33.691Z","replyToPostNumber":13,"replyCount":0,"likeCount":0,"wordCount":95,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":2,"publicVersion":2,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72571,"postNumber":15,"raw":"## Tail\r\n```go\r\npackage tail\r\n\r\nimport (\r\n\t\"bufio\"\r\n\t\"bytes\"\r\n\t\"context\"\r\n\t\"errors\"\r\n\t\"fmt\"\r\n\t\"io\"\r\n\t\"os\"\r\n\t\"sync\"\r\n\t\"time\"\r\n)\r\n\r\nconst (\r\n\tpollInterval = 250 * time.Millisecond\r\n)\r\n\r\ntype fakeReader struct{}\r\n\r\nfunc (*fakeReader) Read(p []byte) (n int, err error) {\r\n\treturn 0, nil\r\n}\r\n\r\ntype Config struct {\r\n\t// Seek to before tailing.\r\n\tSeek SeekInfo\r\n\r\n\t// Provide appended data as the file grows.\r\n\tFollow bool\r\n\r\n\t// Keep trying to open a file if it is inaccessible.\r\n\t// Ignored when Follow is false.\r\n\tRetry bool\r\n\r\n\t// Use PollWatcher for file changes.\r\n\t// FSNotifyWatcher is used if false.\r\n\tPoll bool\r\n\r\n\t// Disable seeking.\r\n\t// Enable this for named pipes (mkfifo).\r\n\tNoSeek bool\r\n\r\n\t// Line delimiter.\r\n\tDelimiter byte\r\n}\r\n\r\nfunc DefaultConfig() Config {\r\n\treturn Config{\r\n\t\tDelimiter: '\\n',\r\n\t}\r\n}\r\n\r\n// SeekInfo\r\ntype SeekInfo struct {\r\n\tOffset int64\r\n\tWhence int\r\n}\r\n\r\ntype Tail struct {\r\n\tcfg Config\r\n\tfilename string\r\n\tfile *os.File\r\n\twatcher Watcher\r\n\r\n\treader *bufio.Reader\r\n\r\n\tfirstOpen bool\r\n\r\n\tmu sync.Mutex\r\n\tctx context.Context\r\n\tcancel context.CancelFunc\r\n}\r\n\r\nfunc New(ctx context.Context, filename string, cfg Config) (*Tail, error) {\r\n\r\n\tvar watcher Watcher\r\n\tvar err error\r\n\tif cfg.Poll {\r\n\t\twatcher, err = NewPollWatcher(filename, pollInterval)\r\n\t\tif err != nil {\r\n\t\t\treturn nil, fmt.Errorf(\"failed to create PollWatcher: %w\", err)\r\n\t\t}\r\n\t} else {\r\n\t\twatcher, err = NewFSNotifyWatcher(filename)\r\n\t\tif err != nil {\r\n\t\t\treturn nil, fmt.Errorf(\"failed to create FSNotifyWatcher: %w\", err)\r\n\t\t}\r\n\t}\r\n\r\n\treturn newTailWithWatcher(ctx, filename, cfg, watcher), nil\r\n}\r\n\r\nfunc newTailWithWatcher(ctx context.Context, filename string, cfg Config, watcher Watcher) *Tail {\r\n\tctx, cancel := context.WithCancel(ctx)\r\n\r\n\treturn &Tail{\r\n\t\tcfg: cfg,\r\n\t\tfilename: filename,\r\n\t\twatcher: watcher,\r\n\t\treader: bufio.NewReader(&fakeReader{}),\r\n\t\tfirstOpen: true,\r\n\t\tctx: ctx,\r\n\t\tcancel: cancel,\r\n\t}\r\n}\r\n\r\nvar (\r\n\tErrBufferTooShort = errors.New(\"buffer is too short to read complete line\")\r\n)\r\n\r\nfunc (t *Tail) ReadLine(b []byte) (n int, err error) {\r\n\tt.mu.Lock()\r\n\tdefer t.mu.Unlock()\r\n\r\n\tfor {\r\n\t\t// Try to read a complete line already in the buffer\r\n\t\tif t.file != nil {\r\n\t\t\tn, found, err := t.tryReadLine(b)\r\n\t\t\tif found || err != nil {\r\n\t\t\t\treturn n, err\r\n\t\t\t}\r\n\t\t\t// No complete line in buffer\r\n\t\t\tif !t.cfg.Follow {\r\n\t\t\t\treturn 0, io.EOF\r\n\t\t\t}\r\n\t\t}\r\n\r\n\t\t// Eagerly open file\r\n\t\tif t.file == nil && !t.cfg.Follow {\r\n\t\t\tif err := t.openFile(); err == nil {\r\n\t\t\t\tcontinue\r\n\t\t\t}\r\n\t\t}\r\n\r\n\t\t// Wait for next file change\r\n\t\tch, bErr := t.watcher.BlockUntilChange(t.ctx)\r\n\t\tif bErr != nil {\r\n\t\t\treturn 0, bErr\r\n\t\t}\r\n\r\n\t\tfileWasOpen := t.file != nil\r\n\r\n\t\tif t.file == nil {\r\n\t\t\tif ch != ChangeDeleted {\r\n\t\t\t\terr = t.openFile()\r\n\t\t\t\tif err != nil {\r\n\t\t\t\t\tif t.cfg.Retry && t.cfg.Follow {\r\n\t\t\t\t\t\tcontinue\r\n\t\t\t\t\t}\r\n\t\t\t\t\treturn\r\n\t\t\t\t}\r\n\t\t\t} else {\r\n\t\t\t\tcontinue\r\n\t\t\t}\r\n\t\t}\r\n\r\n\t\tswitch ch {\r\n\t\tcase ChangeCreated:\r\n\t\t\tif fileWasOpen {\r\n\t\t\t\terr = t.reopenFile()\r\n\t\t\t\tif err != nil {\r\n\t\t\t\t\treturn\r\n\t\t\t\t}\r\n\t\t\t}\r\n\t\tcase ChangeDeleted:\r\n\t\t\terr = t.closeFile()\r\n\t\t\tif err != nil {\r\n\t\t\t\treturn\r\n\t\t\t}\r\n\t\tcase ChangeModified:\r\n\t\t\tt.reader.Reset(t.file)\r\n\t\tcase ChangeTruncated:\r\n\t\t\tif !t.cfg.NoSeek {\r\n\t\t\t\t_, err = t.file.Seek(0, io.SeekStart)\r\n\t\t\t\tif err != nil {\r\n\t\t\t\t\treturn\r\n\t\t\t\t}\r\n\t\t\t}\r\n\t\t}\r\n\t}\r\n}\r\n\r\nfunc (t *Tail) tryReadLine(b []byte) (n int, found bool, err error) {\r\n\tnBuf := t.reader.Buffered()\r\n\tif nBuf == 0 {\r\n\t\tif _, err := t.reader.Peek(1); err != nil {\r\n\t\t\tif errors.Is(err, io.EOF) {\r\n\t\t\t\treturn 0, false, nil\r\n\t\t\t}\r\n\t\t\treturn 0, false, err\r\n\t\t}\r\n\t\tnBuf = t.reader.Buffered()\r\n\t}\r\n\r\n\tpeeked, err := t.reader.Peek(nBuf)\r\n\tif err != nil && !errors.Is(err, io.EOF) {\r\n\t\treturn 0, false, err\r\n\t}\r\n\r\n\tidx := bytes.IndexByte(peeked, t.cfg.Delimiter)\r\n\tif idx < 0 {\r\n\t\treturn 0, false, nil\r\n\t}\r\n\r\n\tif len(b) < idx {\r\n\t\treturn 0, false, ErrBufferTooShort\r\n\t}\r\n\r\n\tcopied := copy(b, peeked[:idx])\r\n\tt.reader.Discard(idx + 1)\r\n\treturn copied, true, nil\r\n}\r\n\r\nfunc (t *Tail) Close() error {\r\n\tt.cancel()\r\n\treturn t.closeFile()\r\n}\r\n\r\nfunc (t *Tail) closeFile() (err error) {\r\n\tif t.file != nil {\r\n\t\terr = t.file.Close()\r\n\t\tt.file = nil\r\n\t}\r\n\treturn\r\n}\r\n\r\nfunc (t *Tail) openFile() error {\r\n\tfile, err := os.Open(t.filename)\r\n\tif err != nil {\r\n\t\treturn err\r\n\t}\r\n\r\n\tif !t.cfg.NoSeek && t.firstOpen {\r\n\t\tif _, err := file.Seek(t.cfg.Seek.Offset, t.cfg.Seek.Whence); err != nil {\r\n\t\t\tfile.Close()\r\n\t\t\treturn err\r\n\t\t}\r\n\t}\r\n\tt.firstOpen = false\r\n\r\n\tt.file = file\r\n\tt.reader.Reset(t.file)\r\n\treturn nil\r\n}\r\n\r\nfunc (t *Tail) reopenFile() (err error) {\r\n\terr = t.closeFile()\r\n\tif err != nil {\r\n\t\treturn\r\n\t}\r\n\terr = t.openFile()\r\n\treturn\r\n}\r\n```\r\n\r\n合并了 `ChangeMoved` 和 `ChangeDeleted`（全部当作 `ChangeDeleted`）\r\n\r\n#### FSM\r\n![tail.svg|690x496](/uploads/image/20/4f/204f18deb48bc0b1155a91674881dc586e2e2b91deecf707e7909bafebc2291b.svg)\r\n\r\n> ref: https://github.com/nxadm/tail\r\n\r\n好复杂，能 syslog over UDP 就别搞这种（","cooked":"

Tail

package tail\n\nimport (\n\t\"bufio\"\n\t\"bytes\"\n\t\"context\"\n\t\"errors\"\n\t\"fmt\"\n\t\"io\"\n\t\"os\"\n\t\"sync\"\n\t\"time\"\n)\n\nconst (\n\tpollInterval = 250 * time.Millisecond\n)\n\ntype fakeReader struct{}\n\nfunc (*fakeReader) Read(p []byte) (n int, err error) {\n\treturn 0, nil\n}\n\ntype Config struct {\n\t// Seek to before tailing.\n\tSeek SeekInfo\n\n\t// Provide appended data as the file grows.\n\tFollow bool\n\n\t// Keep trying to open a file if it is inaccessible.\n\t// Ignored when Follow is false.\n\tRetry bool\n\n\t// Use PollWatcher for file changes.\n\t// FSNotifyWatcher is used if false.\n\tPoll bool\n\n\t// Disable seeking.\n\t// Enable this for named pipes (mkfifo).\n\tNoSeek bool\n\n\t// Line delimiter.\n\tDelimiter byte\n}\n\nfunc DefaultConfig() Config {\n\treturn Config{\n\t\tDelimiter: '\\n',\n\t}\n}\n\n// SeekInfo\ntype SeekInfo struct {\n\tOffset int64\n\tWhence int\n}\n\ntype Tail struct {\n\tcfg      Config\n\tfilename string\n\tfile     *os.File\n\twatcher  Watcher\n\n\treader *bufio.Reader\n\n\tfirstOpen bool\n\n\tmu     sync.Mutex\n\tctx    context.Context\n\tcancel context.CancelFunc\n}\n\nfunc New(ctx context.Context, filename string, cfg Config) (*Tail, error) {\n\n\tvar watcher Watcher\n\tvar err error\n\tif cfg.Poll {\n\t\twatcher, err = NewPollWatcher(filename, pollInterval)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to create PollWatcher: %w\", err)\n\t\t}\n\t} else {\n\t\twatcher, err = NewFSNotifyWatcher(filename)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to create FSNotifyWatcher: %w\", err)\n\t\t}\n\t}\n\n\treturn newTailWithWatcher(ctx, filename, cfg, watcher), nil\n}\n\nfunc newTailWithWatcher(ctx context.Context, filename string, cfg Config, watcher Watcher) *Tail {\n\tctx, cancel := context.WithCancel(ctx)\n\n\treturn &Tail{\n\t\tcfg:       cfg,\n\t\tfilename:  filename,\n\t\twatcher:   watcher,\n\t\treader:    bufio.NewReader(&fakeReader{}),\n\t\tfirstOpen: true,\n\t\tctx:       ctx,\n\t\tcancel:    cancel,\n\t}\n}\n\nvar (\n\tErrBufferTooShort = errors.New(\"buffer is too short to read complete line\")\n)\n\nfunc (t *Tail) ReadLine(b []byte) (n int, err error) {\n\tt.mu.Lock()\n\tdefer t.mu.Unlock()\n\n\tfor {\n\t\t// Try to read a complete line already in the buffer\n\t\tif t.file != nil {\n\t\t\tn, found, err := t.tryReadLine(b)\n\t\t\tif found || err != nil {\n\t\t\t\treturn n, err\n\t\t\t}\n\t\t\t// No complete line in buffer\n\t\t\tif !t.cfg.Follow {\n\t\t\t\treturn 0, io.EOF\n\t\t\t}\n\t\t}\n\n\t\t// Eagerly open file\n\t\tif t.file == nil && !t.cfg.Follow {\n\t\t\tif err := t.openFile(); err == nil {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t}\n\n\t\t// Wait for next file change\n\t\tch, bErr := t.watcher.BlockUntilChange(t.ctx)\n\t\tif bErr != nil {\n\t\t\treturn 0, bErr\n\t\t}\n\n\t\tfileWasOpen := t.file != nil\n\n\t\tif t.file == nil {\n\t\t\tif ch != ChangeDeleted {\n\t\t\t\terr = t.openFile()\n\t\t\t\tif err != nil {\n\t\t\t\t\tif t.cfg.Retry && t.cfg.Follow {\n\t\t\t\t\t\tcontinue\n\t\t\t\t\t}\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tcontinue\n\t\t\t}\n\t\t}\n\n\t\tswitch ch {\n\t\tcase ChangeCreated:\n\t\t\tif fileWasOpen {\n\t\t\t\terr = t.reopenFile()\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t}\n\t\tcase ChangeDeleted:\n\t\t\terr = t.closeFile()\n\t\t\tif err != nil {\n\t\t\t\treturn\n\t\t\t}\n\t\tcase ChangeModified:\n\t\t\tt.reader.Reset(t.file)\n\t\tcase ChangeTruncated:\n\t\t\tif !t.cfg.NoSeek {\n\t\t\t\t_, err = t.file.Seek(0, io.SeekStart)\n\t\t\t\tif err != nil {\n\t\t\t\t\treturn\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n}\n\nfunc (t *Tail) tryReadLine(b []byte) (n int, found bool, err error) {\n\tnBuf := t.reader.Buffered()\n\tif nBuf == 0 {\n\t\tif _, err := t.reader.Peek(1); err != nil {\n\t\t\tif errors.Is(err, io.EOF) {\n\t\t\t\treturn 0, false, nil\n\t\t\t}\n\t\t\treturn 0, false, err\n\t\t}\n\t\tnBuf = t.reader.Buffered()\n\t}\n\n\tpeeked, err := t.reader.Peek(nBuf)\n\tif err != nil && !errors.Is(err, io.EOF) {\n\t\treturn 0, false, err\n\t}\n\n\tidx := bytes.IndexByte(peeked, t.cfg.Delimiter)\n\tif idx < 0 {\n\t\treturn 0, false, nil\n\t}\n\n\tif len(b) < idx {\n\t\treturn 0, false, ErrBufferTooShort\n\t}\n\n\tcopied := copy(b, peeked[:idx])\n\tt.reader.Discard(idx + 1)\n\treturn copied, true, nil\n}\n\nfunc (t *Tail) Close() error {\n\tt.cancel()\n\treturn t.closeFile()\n}\n\nfunc (t *Tail) closeFile() (err error) {\n\tif t.file != nil {\n\t\terr = t.file.Close()\n\t\tt.file = nil\n\t}\n\treturn\n}\n\nfunc (t *Tail) openFile() error {\n\tfile, err := os.Open(t.filename)\n\tif err != nil {\n\t\treturn err\n\t}\n\n\tif !t.cfg.NoSeek && t.firstOpen {\n\t\tif _, err := file.Seek(t.cfg.Seek.Offset, t.cfg.Seek.Whence); err != nil {\n\t\t\tfile.Close()\n\t\t\treturn err\n\t\t}\n\t}\n\tt.firstOpen = false\n\n\tt.file = file\n\tt.reader.Reset(t.file)\n\treturn nil\n}\n\nfunc (t *Tail) reopenFile() (err error) {\n\terr = t.closeFile()\n\tif err != nil {\n\t\treturn\n\t}\n\terr = t.openFile()\n\treturn\n}\n

合并了 ChangeMoved 和 ChangeDeleted（全部当作 ChangeDeleted）

FSM

$\"tail.svg\"$

\n
ref: GitHub - nxadm/tail: [Revamped] Go package for reading from continuously updated files (tail -f)
\n

好复杂，能 syslog over UDP 就别搞这种（

\n","createdAt":"2026-07-14T04:27:02.289Z","updatedAt":"2026-07-14T11:44:54.348Z","replyToPostNumber":null,"replyCount":2,"likeCount":0,"wordCount":698,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":3,"publicVersion":3,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72572,"postNumber":16,"raw":"Minor issue\r\n![Screenshot From 2026-07-14 12-24-23.png|690x77](/uploads/image/95/69/9569895e53cd1911a317db9e05b271c7b88dd4772dd1f31d679569eb5eb7821f.avif)","cooked":"

Minor issue
\n $\"Screenshot$

\n","createdAt":"2026-07-14T04:28:04.311Z","updatedAt":"2026-07-14T04:28:04.311Z","replyToPostNumber":13,"replyCount":1,"likeCount":0,"wordCount":6,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72573,"postNumber":17,"raw":"![github.svg|16x16](/uploads/image/33/56/33565671098c29c555f8e34d4932ff907f8aeafea1d3436d8b60a2e75d423319.svg)\r\n已配置，刷新一下","cooked":"

$\"github.svg\"$
\n已配置，刷新一下

\n","createdAt":"2026-07-14T06:40:02.210Z","updatedAt":"2026-07-14T06:40:02.210Z","replyToPostNumber":16,"replyCount":0,"likeCount":0,"wordCount":9,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"👍","count":1,"users":[{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":4668,"username":"IO","name":null,"avatarUrl":"/uploads/image/86/f2/86f267c0aaa73835a999b7cc3a7e7a701605574eda9a2c2a56f5ddbbba64d1a9.avif","admin":true,"moderator":false,"signatureRaw":"","signatureCooked":null}},{"id":72578,"postNumber":18,"raw":"[quote=\"stateless_trash_can, post:15, topic:15591\"]\r\nTail\r\n[/quote]\r\n\r\n我是垃圾桶，这写的就是垃圾","cooked":"

\n\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\tstateless_trash_can:\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t

\n
Tail
\n

\n\t\t\t\t\t\t

我是垃圾桶，这写的就是垃圾

\n","createdAt":"2026-07-15T09:52:57.562Z","updatedAt":"2026-07-15T09:53:20.231Z","replyToPostNumber":15,"replyCount":1,"likeCount":1,"wordCount":18,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"❤️","count":1,"users":[{"id":2902,"username":"1313","name":"ᐕ)ﾉ"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72581,"postNumber":19,"raw":"![image.png|447x447](/uploads/image/9f/9b/9f9b1af823bdcdba1af14a3e7d75a24d355c30632d66386cdbd4231649985fc9.avif)\r\n\r\nhttps://www.rockstargames.com/newswire/article/25kkao323155a8/presenting-junk-energy-skydives","cooked":"

$\"image.png\"$

\n","createdAt":"2026-07-15T12:13:25.368Z","updatedAt":"2026-07-15T12:13:25.368Z","replyToPostNumber":18,"replyCount":0,"likeCount":0,"wordCount":2,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"❤️","count":1,"users":[{"id":2902,"username":"1313","name":"ᐕ)ﾉ"}],"hasReacted":false},{"reaction":"🤣️","count":1,"users":[{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":4668,"username":"IO","name":null,"avatarUrl":"/uploads/image/86/f2/86f267c0aaa73835a999b7cc3a7e7a701605574eda9a2c2a56f5ddbbba64d1a9.avif","admin":true,"moderator":false,"signatureRaw":"","signatureCooked":null}},{"id":72585,"postNumber":20,"raw":"[quote=\"stateless_trash_can, post:15, topic:15591\"]\r\nTail\r\n[/quote]\r\n\r\n重构\r\n\r\n```go\r\npackage tail\r\n\r\nimport (\r\n\t\"bufio\"\r\n\t\"bytes\"\r\n\t\"context\"\r\n\t\"errors\"\r\n\t\"fmt\"\r\n\t\"io\"\r\n\t\"os\"\r\n\t\"sync\"\r\n\t\"time\"\r\n)\r\n\r\nconst (\r\n\tpollInterval = 250 * time.Millisecond\r\n)\r\n\r\ntype fakeReader struct{}\r\n\r\nfunc (*fakeReader) Read(p []byte) (n int, err error) {\r\n\treturn 0, nil\r\n}\r\n\r\ntype Config struct {\r\n\t// Seek to before tailing.\r\n\tSeek SeekInfo\r\n\r\n\t// Provide appended data as the file grows.\r\n\tFollow bool\r\n\r\n\t// Keep trying to open a file if it is inaccessible.\r\n\t// Requires Follow to be set.\r\n\tRetry bool\r\n\r\n\t// Use PollWatcher for file changes.\r\n\t// FSNotifyWatcher is used if false.\r\n\tPoll bool\r\n\r\n\t// Disable seeking.\r\n\t// Enable this for named pipes (mkfifo).\r\n\tNoSeek bool\r\n\r\n\t// MaxLineSize sets the maximum line length in bytes.\r\n\t// Lines exceeding this are returned in chunks of MaxLineSize with error ErrMaxLineSizeExceeded.\r\n\t// Defaults to 4096 when <= 0.\r\n\tMaxLineSize int\r\n\r\n\t// Line delimiter.\r\n\tDelimiter byte\r\n}\r\n\r\nfunc DefaultConfig() Config {\r\n\treturn Config{\r\n\t\tDelimiter: '\\n',\r\n\t\tMaxLineSize: 4096,\r\n\t}\r\n}\r\n\r\n// SeekInfo\r\ntype SeekInfo struct {\r\n\tOffset int64\r\n\tWhence int\r\n}\r\n\r\ntype tailState int8\r\n\r\nconst (\r\n\tstateInit tailState = iota\r\n\tstateReady\r\n\tstateEOF\r\n\tstateNotOpen\r\n\tstateFailed\r\n)\r\n\r\ntype Tail struct {\r\n\tcfg Config\r\n\tfilename string\r\n\r\n\tfile *os.File\r\n\twatcher Watcher\r\n\treader *bufio.Reader\r\n\r\n\tstate tailState\r\n\tfailErr error\r\n\r\n\tmu sync.Mutex\r\n\tctx context.Context\r\n\tcancel context.CancelFunc\r\n}\r\n\r\nfunc New(ctx context.Context, filename string, cfg Config) (*Tail, error) {\r\n\tif !cfg.Follow && cfg.Retry {\r\n\t\treturn nil, errors.New(\"Retry set without Follow\")\r\n\t}\r\n\r\n\tvar watcher Watcher\r\n\tvar err error\r\n\tif cfg.Poll {\r\n\t\twatcher, err = NewPollWatcher(filename, pollInterval)\r\n\t\tif err != nil {\r\n\t\t\treturn nil, fmt.Errorf(\"failed to create PollWatcher: %w\", err)\r\n\t\t}\r\n\t} else {\r\n\t\twatcher, err = NewFSNotifyWatcher(filename)\r\n\t\tif err != nil {\r\n\t\t\treturn nil, fmt.Errorf(\"failed to create FSNotifyWatcher: %w\", err)\r\n\t\t}\r\n\t}\r\n\r\n\treturn newTailWithWatcher(ctx, filename, cfg, watcher), nil\r\n}\r\n\r\nfunc newTailWithWatcher(ctx context.Context, filename string, cfg Config, watcher Watcher) *Tail {\r\n\tctx, cancel := context.WithCancel(ctx)\r\n\r\n\tif cfg.MaxLineSize <= 0 {\r\n\t\tcfg.MaxLineSize = 4096\r\n\t}\r\n\r\n\treturn &Tail{\r\n\t\tcfg: cfg,\r\n\t\tfilename: filename,\r\n\t\twatcher: watcher,\r\n\t\treader: bufio.NewReaderSize(&fakeReader{}, cfg.MaxLineSize),\r\n\t\tstate: stateInit,\r\n\t\tctx: ctx,\r\n\t\tcancel: cancel,\r\n\t}\r\n}\r\n\r\nvar (\r\n\tErrMaxLineSizeExceeded = errors.New(\"line exceeds MaxLineSize\")\r\n\tErrShortBuffer = io.ErrShortBuffer\r\n\tErrClosed = errors.New(\"closed\")\r\n)\r\n\r\nfunc (t *Tail) ReadLine(b []byte) (int, error) {\r\n\tt.mu.Lock()\r\n\tdefer t.mu.Unlock()\r\n\r\n\tfor {\r\n\t\tswitch t.state {\r\n\t\tcase stateInit:\r\n\t\t\t// Try to open file\r\n\t\t\tfile, err := os.Open(t.filename)\r\n\t\t\tif err != nil {\r\n\t\t\t\tt.failErr = err\r\n\t\t\t\tt.state = stateNotOpen\r\n\t\t\t\tcontinue\r\n\t\t\t} else {\r\n\t\t\t\tif !t.cfg.NoSeek {\r\n\t\t\t\t\t_, err = file.Seek(t.cfg.Seek.Offset, t.cfg.Seek.Whence)\r\n\t\t\t\t\tif err != nil {\r\n\t\t\t\t\t\tt.failErr = err\r\n\t\t\t\t\t\tt.state = stateFailed\r\n\t\t\t\t\t\treturn 0, err\r\n\t\t\t\t\t}\r\n\t\t\t\t}\r\n\t\t\t\tt.file = file\r\n\t\t\t\tt.reader.Reset(file)\r\n\t\t\t\tt.state = stateReady\r\n\t\t\t\tcontinue\r\n\t\t\t}\r\n\r\n\t\tcase stateReady:\r\n\t\t\t// Try to read line\r\n\t\t\tn := t.reader.Buffered()\r\n\t\t\tif n == 0 {\r\n\t\t\t\tif _, err := t.reader.Peek(1); err != nil {\r\n\t\t\t\t\tif errors.Is(err, io.EOF) {\r\n\t\t\t\t\t\tt.state = stateEOF\r\n\t\t\t\t\t\tcontinue\r\n\t\t\t\t\t} else {\r\n\t\t\t\t\t\tt.failErr = err\r\n\t\t\t\t\t\tt.state = stateFailed\r\n\t\t\t\t\t\treturn 0, err\r\n\t\t\t\t\t}\r\n\t\t\t\t}\r\n\t\t\t\tn = t.reader.Buffered()\r\n\t\t\t}\r\n\r\n\t\t\tpeeked, err := t.reader.Peek(n)\r\n\t\t\tif err != nil && !errors.Is(err, io.EOF) {\r\n\t\t\t\tt.failErr = err\r\n\t\t\t\tt.state = stateFailed\r\n\t\t\t\treturn 0, err\r\n\t\t\t}\r\n\r\n\t\t\tidx := bytes.IndexByte(peeked, t.cfg.Delimiter)\r\n\t\t\tif idx < 0 {\r\n\t\t\t\t// No delimiter found\r\n\t\t\t\tif n >= t.cfg.MaxLineSize {\r\n\t\t\t\t\t// Line exceeds MaxLineSize\r\n\t\t\t\t\t// Return chunk\r\n\t\t\t\t\tn = t.cfg.MaxLineSize\r\n\t\t\t\t\tif len(b) < n {\r\n\t\t\t\t\t\treturn 0, ErrShortBuffer\r\n\t\t\t\t\t}\r\n\t\t\t\t\tpeeked, _ = t.reader.Peek(n)\r\n\t\t\t\t\tcopied := copy(b, peeked[:n])\r\n\t\t\t\t\tt.reader.Discard(n)\r\n\t\t\t\t\treturn copied, ErrMaxLineSizeExceeded\r\n\t\t\t\t}\r\n\t\t\t\tt.state = stateEOF\r\n\t\t\t\tcontinue\r\n\t\t\t}\r\n\r\n\t\t\t// Delimiter found\r\n\t\t\t// Chunk if line exceeds MaxLineSize\r\n\t\t\tif idx >= t.cfg.MaxLineSize {\r\n\t\t\t\tn = t.cfg.MaxLineSize\r\n\t\t\t\tif len(b) < n {\r\n\t\t\t\t\treturn 0, ErrShortBuffer\r\n\t\t\t\t}\r\n\t\t\t\tpeeked, _ = t.reader.Peek(n)\r\n\t\t\t\tcopied := copy(b, peeked[:n])\r\n\t\t\t\tt.reader.Discard(n)\r\n\t\t\t\treturn copied, ErrMaxLineSizeExceeded\r\n\t\t\t}\r\n\r\n\t\t\tif len(b) < idx {\r\n\t\t\t\treturn 0, ErrShortBuffer\r\n\t\t\t}\r\n\r\n\t\t\tcopied := copy(b, peeked[:idx])\r\n\t\t\tt.reader.Discard(idx + 1)\r\n\t\t\treturn copied, nil\r\n\r\n\t\tcase stateEOF:\r\n\t\t\tif !t.cfg.Follow {\r\n\t\t\t\tn := t.reader.Buffered()\r\n\t\t\t\tif n > 0 {\r\n\t\t\t\t\t// Probe for more data\r\n\t\t\t\t\t_, peekErr := t.reader.Peek(n + 1)\r\n\t\t\t\t\tif errors.Is(peekErr, io.EOF) {\r\n\t\t\t\t\t\t// True EOF\r\n\t\t\t\t\t\t// Return as final line\r\n\t\t\t\t\t\tpeeked, _ := t.reader.Peek(n)\r\n\t\t\t\t\t\tcopied := copy(b, peeked[:n])\r\n\t\t\t\t\t\tt.reader.Discard(n)\r\n\t\t\t\t\t\treturn copied, nil\r\n\t\t\t\t\t}\r\n\t\t\t\t\tif peekErr != nil {\r\n\t\t\t\t\t\tt.failErr = peekErr\r\n\t\t\t\t\t\tt.state = stateFailed\r\n\t\t\t\t\t\treturn 0, peekErr\r\n\t\t\t\t\t}\r\n\t\t\t\t\t// Has more\r\n\t\t\t\t\tt.state = stateReady\r\n\t\t\t\t\tcontinue\r\n\t\t\t\t}\r\n\t\t\t\tt.failErr = io.EOF\r\n\t\t\t\tt.state = stateFailed\r\n\t\t\t\treturn 0, io.EOF\r\n\t\t\t}\r\n\r\n\t\t\tch, err := t.watcher.BlockUntilChange(t.ctx)\r\n\t\t\tif err != nil {\r\n\t\t\t\tt.failErr = err\r\n\t\t\t\tt.state = stateFailed\r\n\t\t\t\treturn 0, err\r\n\t\t\t}\r\n\r\n\t\t\tswitch ch {\r\n\t\t\tcase ChangeModified, ChangeCreated:\r\n\t\t\t\t// Drain EOF error in bufio.Reader\r\n\t\t\t\tif n := t.reader.Buffered(); n > 0 && n < t.cfg.MaxLineSize {\r\n\t\t\t\t\tt.reader.Peek(n + 1) // trigger file read\r\n\t\t\t\t}\r\n\t\t\t\tt.state = stateReady\r\n\t\t\t\tcontinue\r\n\t\t\tcase ChangeDeleted:\r\n\t\t\t\t// Drain buffer from old file\r\n\t\t\t\tif n := t.reader.Buffered(); n > 0 {\r\n\t\t\t\t\tif len(b) < n {\r\n\t\t\t\t\t\treturn 0, ErrShortBuffer\r\n\t\t\t\t\t}\r\n\t\t\t\t\tpeeked, _ := t.reader.Peek(n)\r\n\t\t\t\t\tcopied := copy(b, peeked[:n])\r\n\t\t\t\t\tt.reader.Discard(n)\r\n\t\t\t\t\tt.file.Close()\r\n\t\t\t\t\tt.file = nil\r\n\t\t\t\t\tt.state = stateNotOpen\r\n\t\t\t\t\treturn copied, nil\r\n\t\t\t\t}\r\n\t\t\t\terr = t.file.Close()\r\n\t\t\t\tif err != nil {\r\n\t\t\t\t\tt.failErr = err\r\n\t\t\t\t\tt.state = stateFailed\r\n\t\t\t\t\treturn 0, err\r\n\t\t\t\t}\r\n\t\t\t\tt.file = nil\r\n\t\t\t\tt.state = stateNotOpen\r\n\t\t\t\tcontinue\r\n\r\n\t\t\tcase ChangeTruncated:\r\n\t\t\t\tif !t.cfg.NoSeek {\r\n\t\t\t\t\t_, err := t.file.Seek(0, io.SeekStart)\r\n\t\t\t\t\tif err != nil {\r\n\t\t\t\t\t\tt.failErr = err\r\n\t\t\t\t\t\tt.state = stateFailed\r\n\t\t\t\t\t\treturn 0, err\r\n\t\t\t\t\t}\r\n\t\t\t\t}\r\n\t\t\t\tt.reader.Reset(t.file)\r\n\t\t\t\tt.state = stateReady\r\n\t\t\t\tcontinue\r\n\t\t\t}\r\n\r\n\t\tcase stateNotOpen:\r\n\t\t\tif !t.cfg.Retry {\r\n\t\t\t\tt.state = stateFailed\r\n\t\t\t\treturn 0, errors.New(\"cannot open file\")\r\n\t\t\t}\r\n\r\n\t\t\tch, err := t.watcher.BlockUntilChange(t.ctx)\r\n\t\t\tif err != nil {\r\n\t\t\t\tt.failErr = err\r\n\t\t\t\tt.state = stateFailed\r\n\t\t\t\treturn 0, err\r\n\t\t\t}\r\n\r\n\t\t\tswitch ch {\r\n\t\t\tcase ChangeCreated, ChangeModified, ChangeTruncated:\r\n\t\t\t\t// Try to open file\r\n\t\t\t\tfile, err := os.Open(t.filename)\r\n\t\t\t\tif err != nil {\r\n\t\t\t\t\tcontinue\r\n\t\t\t\t} else {\r\n\t\t\t\t\tt.file = file\r\n\t\t\t\t\tt.reader.Reset(file)\r\n\t\t\t\t\tt.state = stateReady\r\n\t\t\t\t\tcontinue\r\n\t\t\t\t}\r\n\r\n\t\t\tcase ChangeDeleted:\r\n\t\t\t\tt.failErr = errors.New(\"got ChangeDeleted in stateNotOpen\")\r\n\t\t\t\tt.state = stateFailed\r\n\t\t\t\treturn 0, t.failErr\r\n\r\n\t\t\t}\r\n\r\n\t\tcase stateFailed:\r\n\t\t\treturn 0, t.failErr\r\n\t\t}\r\n\t}\r\n}\r\n\r\nfunc (t *Tail) Close() error {\r\n\tt.cancel()\r\n\tt.mu.Lock()\r\n\tdefer t.mu.Unlock()\r\n\r\n\tt.failErr = ErrClosed\r\n\tt.state = stateFailed\r\n\tif t.file != nil {\r\n\t\tif err := t.file.Close(); err != nil {\r\n\t\t\treturn err\r\n\t\t}\r\n\t}\r\n\treturn nil\r\n}\r\n```","cooked":"

\n\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\tstateless_trash_can:\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t

\n
Tail
\n

\n\t\t\t\t\t\t

重构

package tail\n\nimport (\n\t\"bufio\"\n\t\"bytes\"\n\t\"context\"\n\t\"errors\"\n\t\"fmt\"\n\t\"io\"\n\t\"os\"\n\t\"sync\"\n\t\"time\"\n)\n\nconst (\n\tpollInterval = 250 * time.Millisecond\n)\n\ntype fakeReader struct{}\n\nfunc (*fakeReader) Read(p []byte) (n int, err error) {\n\treturn 0, nil\n}\n\ntype Config struct {\n\t// Seek to before tailing.\n\tSeek SeekInfo\n\n\t// Provide appended data as the file grows.\n\tFollow bool\n\n\t// Keep trying to open a file if it is inaccessible.\n\t// Requires Follow to be set.\n\tRetry bool\n\n\t// Use PollWatcher for file changes.\n\t// FSNotifyWatcher is used if false.\n\tPoll bool\n\n\t// Disable seeking.\n\t// Enable this for named pipes (mkfifo).\n\tNoSeek bool\n\n\t// MaxLineSize sets the maximum line length in bytes.\n\t// Lines exceeding this are returned in chunks of MaxLineSize with error ErrMaxLineSizeExceeded.\n\t// Defaults to 4096 when <= 0.\n\tMaxLineSize int\n\n\t// Line delimiter.\n\tDelimiter byte\n}\n\nfunc DefaultConfig() Config {\n\treturn Config{\n\t\tDelimiter:   '\\n',\n\t\tMaxLineSize: 4096,\n\t}\n}\n\n// SeekInfo\ntype SeekInfo struct {\n\tOffset int64\n\tWhence int\n}\n\ntype tailState int8\n\nconst (\n\tstateInit tailState = iota\n\tstateReady\n\tstateEOF\n\tstateNotOpen\n\tstateFailed\n)\n\ntype Tail struct {\n\tcfg      Config\n\tfilename string\n\n\tfile    *os.File\n\twatcher Watcher\n\treader  *bufio.Reader\n\n\tstate   tailState\n\tfailErr error\n\n\tmu     sync.Mutex\n\tctx    context.Context\n\tcancel context.CancelFunc\n}\n\nfunc New(ctx context.Context, filename string, cfg Config) (*Tail, error) {\n\tif !cfg.Follow && cfg.Retry {\n\t\treturn nil, errors.New(\"Retry set without Follow\")\n\t}\n\n\tvar watcher Watcher\n\tvar err error\n\tif cfg.Poll {\n\t\twatcher, err = NewPollWatcher(filename, pollInterval)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to create PollWatcher: %w\", err)\n\t\t}\n\t} else {\n\t\twatcher, err = NewFSNotifyWatcher(filename)\n\t\tif err != nil {\n\t\t\treturn nil, fmt.Errorf(\"failed to create FSNotifyWatcher: %w\", err)\n\t\t}\n\t}\n\n\treturn newTailWithWatcher(ctx, filename, cfg, watcher), nil\n}\n\nfunc newTailWithWatcher(ctx context.Context, filename string, cfg Config, watcher Watcher) *Tail {\n\tctx, cancel := context.WithCancel(ctx)\n\n\tif cfg.MaxLineSize <= 0 {\n\t\tcfg.MaxLineSize = 4096\n\t}\n\n\treturn &Tail{\n\t\tcfg:      cfg,\n\t\tfilename: filename,\n\t\twatcher:  watcher,\n\t\treader:   bufio.NewReaderSize(&fakeReader{}, cfg.MaxLineSize),\n\t\tstate:    stateInit,\n\t\tctx:      ctx,\n\t\tcancel:   cancel,\n\t}\n}\n\nvar (\n\tErrMaxLineSizeExceeded = errors.New(\"line exceeds MaxLineSize\")\n\tErrShortBuffer         = io.ErrShortBuffer\n\tErrClosed              = errors.New(\"closed\")\n)\n\nfunc (t *Tail) ReadLine(b []byte) (int, error) {\n\tt.mu.Lock()\n\tdefer t.mu.Unlock()\n\n\tfor {\n\t\tswitch t.state {\n\t\tcase stateInit:\n\t\t\t// Try to open file\n\t\t\tfile, err := os.Open(t.filename)\n\t\t\tif err != nil {\n\t\t\t\tt.failErr = err\n\t\t\t\tt.state = stateNotOpen\n\t\t\t\tcontinue\n\t\t\t} else {\n\t\t\t\tif !t.cfg.NoSeek {\n\t\t\t\t\t_, err = file.Seek(t.cfg.Seek.Offset, t.cfg.Seek.Whence)\n\t\t\t\t\tif err != nil {\n\t\t\t\t\t\tt.failErr = err\n\t\t\t\t\t\tt.state = stateFailed\n\t\t\t\t\t\treturn 0, err\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tt.file = file\n\t\t\t\tt.reader.Reset(file)\n\t\t\t\tt.state = stateReady\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase stateReady:\n\t\t\t// Try to read line\n\t\t\tn := t.reader.Buffered()\n\t\t\tif n == 0 {\n\t\t\t\tif _, err := t.reader.Peek(1); err != nil {\n\t\t\t\t\tif errors.Is(err, io.EOF) {\n\t\t\t\t\t\tt.state = stateEOF\n\t\t\t\t\t\tcontinue\n\t\t\t\t\t} else {\n\t\t\t\t\t\tt.failErr = err\n\t\t\t\t\t\tt.state = stateFailed\n\t\t\t\t\t\treturn 0, err\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tn = t.reader.Buffered()\n\t\t\t}\n\n\t\t\tpeeked, err := t.reader.Peek(n)\n\t\t\tif err != nil && !errors.Is(err, io.EOF) {\n\t\t\t\tt.failErr = err\n\t\t\t\tt.state = stateFailed\n\t\t\t\treturn 0, err\n\t\t\t}\n\n\t\t\tidx := bytes.IndexByte(peeked, t.cfg.Delimiter)\n\t\t\tif idx < 0 {\n\t\t\t\t// No delimiter found\n\t\t\t\tif n >= t.cfg.MaxLineSize {\n\t\t\t\t\t// Line exceeds MaxLineSize\n\t\t\t\t\t// Return chunk\n\t\t\t\t\tn = t.cfg.MaxLineSize\n\t\t\t\t\tif len(b) < n {\n\t\t\t\t\t\treturn 0, ErrShortBuffer\n\t\t\t\t\t}\n\t\t\t\t\tpeeked, _ = t.reader.Peek(n)\n\t\t\t\t\tcopied := copy(b, peeked[:n])\n\t\t\t\t\tt.reader.Discard(n)\n\t\t\t\t\treturn copied, ErrMaxLineSizeExceeded\n\t\t\t\t}\n\t\t\t\tt.state = stateEOF\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\t\t// Delimiter found\n\t\t\t// Chunk if line exceeds MaxLineSize\n\t\t\tif idx >= t.cfg.MaxLineSize {\n\t\t\t\tn = t.cfg.MaxLineSize\n\t\t\t\tif len(b) < n {\n\t\t\t\t\treturn 0, ErrShortBuffer\n\t\t\t\t}\n\t\t\t\tpeeked, _ = t.reader.Peek(n)\n\t\t\t\tcopied := copy(b, peeked[:n])\n\t\t\t\tt.reader.Discard(n)\n\t\t\t\treturn copied, ErrMaxLineSizeExceeded\n\t\t\t}\n\n\t\t\tif len(b) < idx {\n\t\t\t\treturn 0, ErrShortBuffer\n\t\t\t}\n\n\t\t\tcopied := copy(b, peeked[:idx])\n\t\t\tt.reader.Discard(idx + 1)\n\t\t\treturn copied, nil\n\n\t\tcase stateEOF:\n\t\t\tif !t.cfg.Follow {\n\t\t\t\tn := t.reader.Buffered()\n\t\t\t\tif n > 0 {\n\t\t\t\t\t// Probe for more data\n\t\t\t\t\t_, peekErr := t.reader.Peek(n + 1)\n\t\t\t\t\tif errors.Is(peekErr, io.EOF) {\n\t\t\t\t\t\t// True EOF\n\t\t\t\t\t\t// Return as final line\n\t\t\t\t\t\tpeeked, _ := t.reader.Peek(n)\n\t\t\t\t\t\tcopied := copy(b, peeked[:n])\n\t\t\t\t\t\tt.reader.Discard(n)\n\t\t\t\t\t\treturn copied, nil\n\t\t\t\t\t}\n\t\t\t\t\tif peekErr != nil {\n\t\t\t\t\t\tt.failErr = peekErr\n\t\t\t\t\t\tt.state = stateFailed\n\t\t\t\t\t\treturn 0, peekErr\n\t\t\t\t\t}\n\t\t\t\t\t// Has more\n\t\t\t\t\tt.state = stateReady\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\t\t\t\tt.failErr = io.EOF\n\t\t\t\tt.state = stateFailed\n\t\t\t\treturn 0, io.EOF\n\t\t\t}\n\n\t\t\tch, err := t.watcher.BlockUntilChange(t.ctx)\n\t\t\tif err != nil {\n\t\t\t\tt.failErr = err\n\t\t\t\tt.state = stateFailed\n\t\t\t\treturn 0, err\n\t\t\t}\n\n\t\t\tswitch ch {\n\t\t\tcase ChangeModified, ChangeCreated:\n\t\t\t\t// Drain EOF error in bufio.Reader\n\t\t\t\tif n := t.reader.Buffered(); n > 0 && n < t.cfg.MaxLineSize {\n\t\t\t\t\tt.reader.Peek(n + 1) // trigger file read\n\t\t\t\t}\n\t\t\t\tt.state = stateReady\n\t\t\t\tcontinue\n\t\t\tcase ChangeDeleted:\n\t\t\t\t// Drain buffer from old file\n\t\t\t\tif n := t.reader.Buffered(); n > 0 {\n\t\t\t\t\tif len(b) < n {\n\t\t\t\t\t\treturn 0, ErrShortBuffer\n\t\t\t\t\t}\n\t\t\t\t\tpeeked, _ := t.reader.Peek(n)\n\t\t\t\t\tcopied := copy(b, peeked[:n])\n\t\t\t\t\tt.reader.Discard(n)\n\t\t\t\t\tt.file.Close()\n\t\t\t\t\tt.file = nil\n\t\t\t\t\tt.state = stateNotOpen\n\t\t\t\t\treturn copied, nil\n\t\t\t\t}\n\t\t\t\terr = t.file.Close()\n\t\t\t\tif err != nil {\n\t\t\t\t\tt.failErr = err\n\t\t\t\t\tt.state = stateFailed\n\t\t\t\t\treturn 0, err\n\t\t\t\t}\n\t\t\t\tt.file = nil\n\t\t\t\tt.state = stateNotOpen\n\t\t\t\tcontinue\n\n\t\t\tcase ChangeTruncated:\n\t\t\t\tif !t.cfg.NoSeek {\n\t\t\t\t\t_, err := t.file.Seek(0, io.SeekStart)\n\t\t\t\t\tif err != nil {\n\t\t\t\t\t\tt.failErr = err\n\t\t\t\t\t\tt.state = stateFailed\n\t\t\t\t\t\treturn 0, err\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tt.reader.Reset(t.file)\n\t\t\t\tt.state = stateReady\n\t\t\t\tcontinue\n\t\t\t}\n\n\t\tcase stateNotOpen:\n\t\t\tif !t.cfg.Retry {\n\t\t\t\tt.state = stateFailed\n\t\t\t\treturn 0, errors.New(\"cannot open file\")\n\t\t\t}\n\n\t\t\tch, err := t.watcher.BlockUntilChange(t.ctx)\n\t\t\tif err != nil {\n\t\t\t\tt.failErr = err\n\t\t\t\tt.state = stateFailed\n\t\t\t\treturn 0, err\n\t\t\t}\n\n\t\t\tswitch ch {\n\t\t\tcase ChangeCreated, ChangeModified, ChangeTruncated:\n\t\t\t\t// Try to open file\n\t\t\t\tfile, err := os.Open(t.filename)\n\t\t\t\tif err != nil {\n\t\t\t\t\tcontinue\n\t\t\t\t} else {\n\t\t\t\t\tt.file = file\n\t\t\t\t\tt.reader.Reset(file)\n\t\t\t\t\tt.state = stateReady\n\t\t\t\t\tcontinue\n\t\t\t\t}\n\n\t\t\tcase ChangeDeleted:\n\t\t\t\tt.failErr = errors.New(\"got ChangeDeleted in stateNotOpen\")\n\t\t\t\tt.state = stateFailed\n\t\t\t\treturn 0, t.failErr\n\n\t\t\t}\n\n\t\tcase stateFailed:\n\t\t\treturn 0, t.failErr\n\t\t}\n\t}\n}\n\nfunc (t *Tail) Close() error {\n\tt.cancel()\n\tt.mu.Lock()\n\tdefer t.mu.Unlock()\n\n\tt.failErr = ErrClosed\n\tt.state = stateFailed\n\tif t.file != nil {\n\t\tif err := t.file.Close(); err != nil {\n\t\t\treturn err\n\t\t}\n\t}\n\treturn nil\n}\n

\n","createdAt":"2026-07-15T15:17:32.222Z","updatedAt":"2026-07-15T15:17:32.222Z","replyToPostNumber":15,"replyCount":0,"likeCount":0,"wordCount":973,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72726,"postNumber":21,"raw":"# go maps delete and clone\r\ngo maps 即使删除 kv pairs 也不会自动 shrink，如果长期使用同一 map 增删不同的 key 会导致 map 内存占用无限增长：\r\n> In the official standard Go runtime implementation, the backing array of a map will never shrink, even if all entries are deleted from the map. This is a form of memory wasting. But in practice, this is seldom a problem and and actually often good for program performances. - https://go101.org/optimizations/6-map.html\r\n\r\n> https://github.com/VictoriaMetrics/VictoriaMetrics/issues/5379\r\n\r\n解决方法是当 map 到达一定大小之后建立新 map 并复制所有 kv pairs：\r\n```go\r\n\t\tnewCap := 0\r\n\t\tfor _, idx := range s.idxMap {\r\n\t\t\tif isOneIterationBefore(idx, s.idx) {\r\n\t\t\t\tnewCap++\r\n\t\t\t}\r\n\t\t}\r\n\t\tnewMap := make(map[uint64]uint32, newCap)\r\n\t\tfor k, idx := range s.idxMap {\r\n\t\t\tif isOneIterationBefore(idx, s.idx) {\r\n\t\t\t\tnewMap[k] = idx\r\n\t\t\t}\r\n\t\t}\r\n\t\ts.idxMap = newMap\r\n```\r\n这里计算需要保留的 kv pair 个数并以此为 capacity hint 建立新 map\r\n\r\n但是这样每次重建可能过于激进，产生更多 GC 压力\r\n\r\n```\r\n 71 . . \t\tnewCap := 0 \r\n 72 . . \t\tfor _, idx := range s.idxMap { \r\n 73 . . \t\t\tif isOneIterationBefore(idx, s.idx) { \r\n 74 . . \t\t\t\tnewCap++ \r\n 75 . . \t\t\t} \r\n 76 . . \t\t} \r\n 77 161.71MB 161.71MB \t\tnewMap := make(map[uint64]uint32, newCap) \r\n 78 . . \t\tfor k, idx := range s.idxMap { \r\n 79 . . \t\t\tif isOneIterationBefore(idx, s.idx) { \r\n 80 3.50MB 3.50MB \t\t\t\tnewMap[k] = idx \r\n 81 . . \t\t\t} \r\n 82 . . \t\t} \r\n```\r\n\r\n设置一个 mapResizeFactor，当 cap > use*factor 时才重建\r\n```go\r\n\t\toldCap := len(s.idxMap)\r\n\t\tnewCap := 0\r\n\t\tfor _, idx := range s.idxMap {\r\n\t\t\tif isOneIterationBefore(idx, s.idx) {\r\n\t\t\t\tnewCap++\r\n\t\t\t}\r\n\t\t}\r\n\t\tif oldCap >= newCap*mapResizeFactor {\r\n\t\t\tnewMap := make(map[uint64]uint32, newCap)\r\n\t\t\tfor k, idx := range s.idxMap {\r\n\t\t\t\tif isOneIterationBefore(idx, s.idx) {\r\n\t\t\t\t\tnewMap[k] = idx\r\n\t\t\t\t}\r\n\t\t\t}\r\n\t\t\ts.idxMap = newMap\r\n\t\t} else {\r\n\t\t\tmaps.DeleteFunc(s.idxMap, func(k uint64, idx uint32) bool {\r\n\t\t\t\treturn !isOneIterationBefore(idx, s.idx)\r\n\t\t\t})\r\n\t\t}\r\n```\r\n\r\n另外这里需要遍历旧 map 两遍，因为先删除不需要的 kv pairs 更低效，而且删除后再复制本质上还是复制所有的 kv pairs 到新 map\r\n\r\n如果使用 maps.Clone 就无法达到缩小 map 的效果，因为 maps.Clone 是链接到 runtime 包的实现，直接复制了原 map 的内存结构\r\n\r\n```go\r\n// clone is implemented in the runtime package.\r\n//\r\n//go:linkname clone maps.clone\r\nfunc clone(m any) any\r\n\r\n// Clone returns a copy of m. This is a shallow clone:\r\n// the new keys and values are set using ordinary assignment.\r\nfunc Clone[M ~map[K]V, K comparable, V any](m M) M {\r\n\t// Preserve nil in case it matters.\r\n\tif m == nil {\r\n\t\treturn nil\r\n\t}\r\n\treturn clone(m).(M)\r\n}\r\n```\r\n\r\n```go\r\n// mapclone for implementing maps.Clone\r\n//\r\n//go:linkname mapclone maps.clone\r\nfunc mapclone(m any) any {\r\n\te := efaceOf(&m)\r\n\ttyp := (*abi.MapType)(unsafe.Pointer(e._type))\r\n\tmap_ := (*maps.Map)(e.data)\r\n\tmap_ = map_.Clone(typ)\r\n\te.data = (unsafe.Pointer)(map_)\r\n\treturn m\r\n}\r\n```\r\n\r\n```go\r\nfunc (m *Map) Clone(typ *abi.MapType) *Map {\r\n\t// Note: this should never be called with a nil map.\r\n\tif m.writing != 0 {\r\n\t\tfatal(\"concurrent map clone and map write\")\r\n\t}\r\n\r\n\t// Shallow copy the Map structure.\r\n\tm2 := new(Map)\r\n\t*m2 = *m\r\n\tm = m2\r\n\r\n\t// We need to just deep copy the dirPtr field.\r\n\tif m.dirPtr == nil {\r\n\t\t// delayed group allocation, nothing to do.\r\n\t} else if m.dirLen == 0 {\r\n\t\t// Clone one group.\r\n\t\toldGroup := groupReference{data: m.dirPtr}\r\n\t\tnewGroup := groupReference{data: newGroups(typ, 1).data}\r\n\t\tcloneGroup(typ, newGroup, oldGroup)\r\n\t\tm.dirPtr = newGroup.data\r\n\t} else {\r\n\t\t// Clone each (different) table.\r\n\t\toldDir := unsafe.Slice((**table)(m.dirPtr), m.dirLen)\r\n\t\tnewDir := make([]*table, m.dirLen)\r\n\t\tfor i, t := range oldDir {\r\n\t\t\tif i > 0 && t == oldDir[i-1] {\r\n\t\t\t\tnewDir[i] = newDir[i-1]\r\n\t\t\t\tcontinue\r\n\t\t\t}\r\n\t\t\tnewDir[i] = t.clone(typ)\r\n\t\t}\r\n\t\tm.dirPtr = unsafe.Pointer(&newDir[0])\r\n\t}\r\n\r\n\treturn m\r\n}\r\n```","cooked":"

go maps delete and clone

go maps 即使删除 kv pairs 也不会自动 shrink，如果长期使用同一 map 增删不同的 key 会导致 map 内存占用无限增长：

\n
In the official standard Go runtime implementation, the backing array of a map will never shrink, even if all entries are deleted from the map. This is a form of memory wasting. But in practice, this is seldom a problem and and actually often good for program performances. - Maps -Go 101
\n

\n
vmstorage GC overhead grows on long running instance · Issue #5379 · VictoriaMetrics/VictoriaMetrics
\n

解决方法是当 map 到达一定大小之后建立新 map 并复制所有 kv pairs：

\t\tnewCap := 0\n\t\tfor _, idx := range s.idxMap {\n\t\t\tif isOneIterationBefore(idx, s.idx) {\n\t\t\t\tnewCap++\n\t\t\t}\n\t\t}\n\t\tnewMap := make(map[uint64]uint32, newCap)\n\t\tfor k, idx := range s.idxMap {\n\t\t\tif isOneIterationBefore(idx, s.idx) {\n\t\t\t\tnewMap[k] = idx\n\t\t\t}\n\t\t}\n\t\ts.idxMap = newMap\n

这里计算需要保留的 kv pair 个数并以此为 capacity hint 建立新 map

但是这样每次重建可能过于激进，产生更多 GC 压力

     71            .          .           \t\tnewCap := 0 \n     72            .          .           \t\tfor _, idx := range s.idxMap { \n     73            .          .           \t\t\tif isOneIterationBefore(idx, s.idx) { \n     74            .          .           \t\t\t\tnewCap++ \n     75            .          .           \t\t\t} \n     76            .          .           \t\t} \n     77     161.71MB   161.71MB           \t\tnewMap := make(map[uint64]uint32, newCap) \n     78            .          .           \t\tfor k, idx := range s.idxMap { \n     79            .          .           \t\t\tif isOneIterationBefore(idx, s.idx) { \n     80       3.50MB     3.50MB           \t\t\t\tnewMap[k] = idx \n     81            .          .           \t\t\t} \n     82            .          .           \t\t} \n

设置一个 mapResizeFactor，当 cap > use*factor 时才重建

\t\toldCap := len(s.idxMap)\n\t\tnewCap := 0\n\t\tfor _, idx := range s.idxMap {\n\t\t\tif isOneIterationBefore(idx, s.idx) {\n\t\t\t\tnewCap++\n\t\t\t}\n\t\t}\n\t\tif oldCap >= newCap*mapResizeFactor {\n\t\t\tnewMap := make(map[uint64]uint32, newCap)\n\t\t\tfor k, idx := range s.idxMap {\n\t\t\t\tif isOneIterationBefore(idx, s.idx) {\n\t\t\t\t\tnewMap[k] = idx\n\t\t\t\t}\n\t\t\t}\n\t\t\ts.idxMap = newMap\n\t\t} else {\n\t\t\tmaps.DeleteFunc(s.idxMap, func(k uint64, idx uint32) bool {\n\t\t\t\treturn !isOneIterationBefore(idx, s.idx)\n\t\t\t})\n\t\t}\n

另外这里需要遍历旧 map 两遍，因为先删除不需要的 kv pairs 更低效，而且删除后再复制本质上还是复制所有的 kv pairs 到新 map

如果使用 maps.Clone 就无法达到缩小 map 的效果，因为 maps.Clone 是链接到 runtime 包的实现，直接复制了原 map 的内存结构

// clone is implemented in the runtime package.\n//\n//go:linkname clone maps.clone\nfunc clone(m any) any\n\n// Clone returns a copy of m.  This is a shallow clone:\n// the new keys and values are set using ordinary assignment.\nfunc Clone[M ~map[K]V, K comparable, V any](m M) M {\n\t// Preserve nil in case it matters.\n\tif m == nil {\n\t\treturn nil\n\t}\n\treturn clone(m).(M)\n}\n

// mapclone for implementing maps.Clone\n//\n//go:linkname mapclone maps.clone\nfunc mapclone(m any) any {\n\te := efaceOf(&m)\n\ttyp := (*abi.MapType)(unsafe.Pointer(e._type))\n\tmap_ := (*maps.Map)(e.data)\n\tmap_ = map_.Clone(typ)\n\te.data = (unsafe.Pointer)(map_)\n\treturn m\n}\n

func (m *Map) Clone(typ *abi.MapType) *Map {\n\t// Note: this should never be called with a nil map.\n\tif m.writing != 0 {\n\t\tfatal(\"concurrent map clone and map write\")\n\t}\n\n\t// Shallow copy the Map structure.\n\tm2 := new(Map)\n\t*m2 = *m\n\tm = m2\n\n\t// We need to just deep copy the dirPtr field.\n\tif m.dirPtr == nil {\n\t\t// delayed group allocation, nothing to do.\n\t} else if m.dirLen == 0 {\n\t\t// Clone one group.\n\t\toldGroup := groupReference{data: m.dirPtr}\n\t\tnewGroup := groupReference{data: newGroups(typ, 1).data}\n\t\tcloneGroup(typ, newGroup, oldGroup)\n\t\tm.dirPtr = newGroup.data\n\t} else {\n\t\t// Clone each (different) table.\n\t\toldDir := unsafe.Slice((**table)(m.dirPtr), m.dirLen)\n\t\tnewDir := make([]*table, m.dirLen)\n\t\tfor i, t := range oldDir {\n\t\t\tif i > 0 && t == oldDir[i-1] {\n\t\t\t\tnewDir[i] = newDir[i-1]\n\t\t\t\tcontinue\n\t\t\t}\n\t\t\tnewDir[i] = t.clone(typ)\n\t\t}\n\t\tm.dirPtr = unsafe.Pointer(&newDir[0])\n\t}\n\n\treturn m\n}\n

\n","createdAt":"2026-07-24T18:29:26.201Z","updatedAt":"2026-07-24T18:29:26.201Z","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":706,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72771,"postNumber":22,"raw":"完蛋力，我是傻逼，以前写 `sync.Pool` 都写错了 😭\r\n\r\n> We’re using a pool of `[]byte`. Generally (though not always), when you pass a value to an interface, it may cause the value to be placed on the heap. This happens here too, not just with slices but with anything you pass to pool.Put() that isn’t a pointer.
- https://victoriametrics.com/blog/go-sync-pool\r\n\r\n```go\r\nvar bufPool = sync.Pool{\r\n\tNew: func() any {\r\n\t\t// The Pool's New function should generally only return pointer\r\n\t\t// types, since a pointer can be put into the return interface\r\n\t\t// value without an allocation:\r\n\t\treturn new(bytes.Buffer)\r\n\t},\r\n}\r\n```\r\n>https://github.com/golang/go/blob/2580d0e08d5e9f979b943758d3c49877fb2324cb/src/sync/example_pool_test.go#L15","cooked":"

完蛋力，我是傻逼，以前写 sync.Pool 都写错了 😭

\n
We’re using a pool of []byte. Generally (though not always), when you pass a value to an interface, it may cause the value to be placed on the heap. This happens here too, not just with slices but with anything you pass to pool.Put() that isn’t a pointer.
- Go sync.Pool and the Mechanics Behind It
\n

var bufPool = sync.Pool{\n\tNew: func() any {\n\t\t// The Pool's New function should generally only return pointer\n\t\t// types, since a pointer can be put into the return interface\n\t\t// value without an allocation:\n\t\treturn new(bytes.Buffer)\n\t},\n}\n

\n
go/src/sync/example_pool_test.go at 2580d0e08d5e9f979b943758d3c49877fb2324cb · golang/go
\n

\n","createdAt":"2026-07-26T16:52:25.815Z","updatedAt":"2026-07-26T16:54:07.946Z","replyToPostNumber":null,"replyCount":1,"likeCount":1,"wordCount":111,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"❤️","count":1,"users":[{"id":5905,"username":"PipaQinse233","name":"PipaQinse233"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72816,"postNumber":23,"raw":"大于 8 byte 的类型赋值到接口类型都会逃逸到堆，产生堆分配\r\n\r\n可以使用 go 编译器进行逃逸分析：\r\n\r\n```shell\r\ngo build -gcflags=\"-m\" /path/to/src.go\r\n```\r\n\r\n## `sync.Pool` 案例分析\r\n\r\n#### `slice.go` - 切片池\r\n```go\r\npackage heaptest\r\n\r\nimport \"sync\"\r\n\r\nvar slicePool = sync.Pool{\r\n\tNew: func() any {\r\n\t\treturn make([]byte, 10)\r\n\t},\r\n}\r\n\r\nfunc getSlice() []byte {\r\n\treturn slicePool.Get().([]byte)\r\n}\r\n\r\nfunc putSlice(s []byte) {\r\n\tslicePool.Put(s)\r\n}\r\n```\r\n```\r\n./slice.go:7:14: make([]byte, 10) escapes to heap\r\n./slice.go:15:15: leaking param: s\r\n./slice.go:16:16: s escapes to heap\r\n```\r\n这里 `slicePool.Put(s)` 会产生一次堆分配，因为 slice header 有 24 bytes，无法放入接口类型\r\n\r\n\r\n#### `slicePtr.go` - 切片指针池\r\n```go\r\npackage heaptest\r\n\r\nimport \"sync\"\r\n\r\nvar slicePtrPool = sync.Pool{\r\n\tNew: func() any {\r\n\t\tb := make([]byte, 10)\r\n\t\treturn &b\r\n\t},\r\n}\r\n\r\nfunc getPtrSlice() *[]byte {\r\n\treturn slicePtrPool.Get().(*[]byte)\r\n}\r\n\r\nfunc putPtrSlice(s *[]byte) {\r\n\tslicePtrPool.Put(s)\r\n}\r\n```\r\n```\r\n./slicePtr.go:7:3: moved to heap: b\r\n./slicePtr.go:7:12: make([]byte, 10) escapes to heap\r\n./slicePtr.go:16:18: leaking param: s\r\n```\r\n这样 slice header 和 slice 本身都分配在堆上，Put 时仅储存 slice header 指针（8 byte）\r\n\r\n#### `map.go` - 表池\r\n```go\r\npackage heaptest\r\n\r\nimport \"sync\"\r\n\r\nvar mapPool = sync.Pool{\r\n\tNew: func() any {\r\n\t\treturn make(map[int]int)\r\n\t},\r\n}\r\n\r\nfunc getMap() map[int]int {\r\n\treturn mapPool.Get().(map[int]int)\r\n}\r\n\r\nfunc putMap(m map[int]int) {\r\n\tclear(m)\r\n\tmapPool.Put(m)\r\n}\r\n```\r\n```\r\n./map.go:7:14: make(map[int]int) escapes to heap\r\n./map.go:15:13: leaking param: m\r\n```\r\nmap 类型本来就是指向 map 内容的指针，所以可以直接赋值给接口类型无需堆分配\r\n类似的 chan, func() 也是指针所以也可以高效赋值给接口类型","cooked":"

大于 8 byte 的类型赋值到接口类型都会逃逸到堆，产生堆分配

可以使用 go 编译器进行逃逸分析：

go build -gcflags=\"-m\" /path/to/src.go\n

`sync.Pool` 案例分析

`slice.go` - 切片池

package heaptest\n\nimport \"sync\"\n\nvar slicePool = sync.Pool{\n\tNew: func() any {\n\t\treturn make([]byte, 10)\n\t},\n}\n\nfunc getSlice() []byte {\n\treturn slicePool.Get().([]byte)\n}\n\nfunc putSlice(s []byte) {\n\tslicePool.Put(s)\n}\n

./slice.go:7:14: make([]byte, 10) escapes to heap\n./slice.go:15:15: leaking param: s\n./slice.go:16:16: s escapes to heap\n

这里 slicePool.Put(s) 会产生一次堆分配，因为 slice header 有 24 bytes，无法放入接口类型

`slicePtr.go` - 切片指针池

package heaptest\n\nimport \"sync\"\n\nvar slicePtrPool = sync.Pool{\n\tNew: func() any {\n\t\tb := make([]byte, 10)\n\t\treturn &b\n\t},\n}\n\nfunc getPtrSlice() *[]byte {\n\treturn slicePtrPool.Get().(*[]byte)\n}\n\nfunc putPtrSlice(s *[]byte) {\n\tslicePtrPool.Put(s)\n}\n

./slicePtr.go:7:3: moved to heap: b\n./slicePtr.go:7:12: make([]byte, 10) escapes to heap\n./slicePtr.go:16:18: leaking param: s\n

这样 slice header 和 slice 本身都分配在堆上，Put 时仅储存 slice header 指针（8 byte）

`map.go` - 表池

package heaptest\n\nimport \"sync\"\n\nvar mapPool = sync.Pool{\n\tNew: func() any {\n\t\treturn make(map[int]int)\n\t},\n}\n\nfunc getMap() map[int]int {\n\treturn mapPool.Get().(map[int]int)\n}\n\nfunc putMap(m map[int]int) {\n\tclear(m)\n\tmapPool.Put(m)\n}\n

./map.go:7:14: make(map[int]int) escapes to heap\n./map.go:15:13: leaking param: m\n

map 类型本来就是指向 map 内容的指针，所以可以直接赋值给接口类型无需堆分配
\n类似的 chan, func() 也是指针所以也可以高效赋值给接口类型

\n","createdAt":"2026-07-29T14:08:50.152Z","updatedAt":"2026-07-29T14:10:02.190Z","replyToPostNumber":22,"replyCount":1,"likeCount":1,"wordCount":325,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"❤️","count":1,"users":[{"id":4668,"username":"IO","name":null}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72817,"postNumber":24,"raw":"## 跑分\r\n#### 错误写法\r\n```\r\ngoos: linux\r\ngoarch: amd64\r\npkg: github.com/HT4w5/kv\r\ncpu: Intel(R) Core(TM) Ultra 9 185H\r\n=== RUN BenchmarkCacheSetDelete\r\nBenchmarkCacheSetDelete\r\n=== RUN BenchmarkCacheSetDelete/Size-33554432\r\nBenchmarkCacheSetDelete/Size-33554432\r\nBenchmarkCacheSetDelete/Size-33554432-22 53940969 22.58 ns/op 708.57 MB/s 30 B/op 0 allocs/op\r\n=== RUN BenchmarkCacheSetDelete/Size-1073741824\r\nBenchmarkCacheSetDelete/Size-1073741824\r\nBenchmarkCacheSetDelete/Size-1073741824-22 61089979 16.99 ns/op 941.60 MB/s 24 B/op 1 allocs/op\r\nPASS\r\nok github.com/HT4w5/kv 2.310s\r\n```\r\n\r\n#### 正确写法\r\n```\r\ngoos: linux\r\ngoarch: amd64\r\npkg: github.com/HT4w5/kv\r\ncpu: Intel(R) Core(TM) Ultra 9 185H\r\n=== RUN BenchmarkCacheSetDelete\r\nBenchmarkCacheSetDelete\r\n=== RUN BenchmarkCacheSetDelete/Size-33554432\r\nBenchmarkCacheSetDelete/Size-33554432\r\nBenchmarkCacheSetDelete/Size-33554432-22 91227044 13.39 ns/op 1194.55 MB/s 0 B/op 0 allocs/op\r\n=== RUN BenchmarkCacheSetDelete/Size-1073741824\r\nBenchmarkCacheSetDelete/Size-1073741824\r\nBenchmarkCacheSetDelete/Size-1073741824-22 116488242 10.45 ns/op 1531.02 MB/s 0 B/op 0 allocs/op\r\nPASS\r\nok github.com/HT4w5/kv 3.400s\r\n```","cooked":"

跑分

错误写法

goos: linux\ngoarch: amd64\npkg: github.com/HT4w5/kv\ncpu: Intel(R) Core(TM) Ultra 9 185H\n=== RUN   BenchmarkCacheSetDelete\nBenchmarkCacheSetDelete\n=== RUN   BenchmarkCacheSetDelete/Size-33554432\nBenchmarkCacheSetDelete/Size-33554432\nBenchmarkCacheSetDelete/Size-33554432-22                53940969                22.58 ns/op      708.57 MB/s          30 B/op          0 allocs/op\n=== RUN   BenchmarkCacheSetDelete/Size-1073741824\nBenchmarkCacheSetDelete/Size-1073741824\nBenchmarkCacheSetDelete/Size-1073741824-22              61089979                16.99 ns/op      941.60 MB/s          24 B/op          1 allocs/op\nPASS\nok      github.com/HT4w5/kv     2.310s\n

正确写法

goos: linux\ngoarch: amd64\npkg: github.com/HT4w5/kv\ncpu: Intel(R) Core(TM) Ultra 9 185H\n=== RUN   BenchmarkCacheSetDelete\nBenchmarkCacheSetDelete\n=== RUN   BenchmarkCacheSetDelete/Size-33554432\nBenchmarkCacheSetDelete/Size-33554432\nBenchmarkCacheSetDelete/Size-33554432-22                91227044                13.39 ns/op     1194.55 MB/s           0 B/op          0 allocs/op\n=== RUN   BenchmarkCacheSetDelete/Size-1073741824\nBenchmarkCacheSetDelete/Size-1073741824\nBenchmarkCacheSetDelete/Size-1073741824-22              116488242               10.45 ns/op     1531.02 MB/s           0 B/op          0 allocs/op\nPASS\nok      github.com/HT4w5/kv     3.400s\n

\n","createdAt":"2026-07-29T15:44:09.692Z","updatedAt":"2026-07-29T15:44:09.692Z","replyToPostNumber":23,"replyCount":0,"likeCount":0,"wordCount":113,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72831,"postNumber":25,"raw":"才发现 Go `atomic.Uint64` 只有 `Add()` 没有 `Sub()` 所以想减只能 `Add(^uint64(delta - 1))` 🤯","cooked":"

才发现 Go atomic.Uint64 只有 Add() 没有 Sub() 所以想减只能 Add(^uint64(delta - 1)) 🤯

\n","createdAt":"2026-07-31T04:51:32.187Z","updatedAt":"2026-07-31T04:51:32.187Z","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":21,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72858,"postNumber":26,"raw":"https://victoriametrics.com/blog/go-1-27/index.html\r\n\r\n有意思的：\r\n- 泛型方法\r\n- uuid\r\n- json/v2\r\n- simd","cooked":"\n

有意思的：

泛型方法
uuid
json/v2
simd

\n","createdAt":"2026-08-02T18:18:48.462Z","updatedAt":"2026-08-02T18:18:48.462Z","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":17,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72954,"postNumber":27,"raw":"# 造轮子笑传之 chunked cache bucket\r\n起因是我需要 [fastcache](https://github.com/VictoriaMetrics/fastcache)，但是 [fastcache](https://github.com/VictoriaMetrics/fastcache) 没有迭代器，所以干脆自己仿制一个就当学习了\r\n\r\n## Public Interface\r\nhttps://pkg.go.dev/github.com/HT4w5/kv@v0.2.0\r\n\r\n```\r\n type Cache\r\n func New(size int) *Cache\r\n func (c *Cache) Del(k []byte)\r\n func (c *Cache) Get(dst, k []byte) (res []byte)\r\n func (c *Cache) Has(k []byte) (found bool)\r\n func (c *Cache) HasGet(dst, k []byte) (res []byte, found bool)\r\n func (c *Cache) Iterator() *Iterator\r\n func (c *Cache) LoadStats(stats *Stats)\r\n func (c *Cache) Reset()\r\n func (c *Cache) Set(k, v []byte)\r\ntype Iterator\r\n func (it *Iterator) GetNext(kDst, vDst []byte) ([]byte, []byte, bool)\r\ntype Stats\r\n```\r\n\r\n## 架构\r\n```shell\r\n┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐\r\n│ │ │ map[hash]idx │ │ │\r\n│ [512]bucket◄────┼─┤ [n]chunks◄──────┼─┤ [65536]byte │\r\n│ │ │ idx │ │ useCount │\r\n└─────────────────┘ └─────────────────┘ └─────────────────┘\r\n cache bucket chunk \r\n```\r\n- Cache 分 512 个 bucket，每个 bucket 有独立的读写锁\r\n- 每个 bucket 是由 n 个 chunk 构成的 ring buffer，idx 记录下一次写入位置，map[hash]idx 记录现有键值对位置\r\n- 每个 chunk 是 [65536]byte\r\n- n 可根据 Cache 大小进行调整，至少为 1，整个 Cache 至少 65536\\*1\\*512 B = 32 MiB\r\n\r\n#### 键值对格式\r\n```shell\r\n 0 1 2 3\r\n 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| kLen (uint16le) | vLen (uint16le) |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| |\r\n+ k (kLen bytes) +\r\n| |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| |\r\n+ v (vLen bytes) +\r\n| |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n```\r\n- 键值对不可跨 chunk 存在，键值对记录最长为 65536 bytes,因此 kLen, vLen 可用 uint16 记录\r\n\r\n#### idx 格式\r\n```shell\r\n 0 1 2 3\r\n 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n|W| chunkIndex (47 bits) |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| chunkIndex (cont.) | localIndex (16 bits) |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n```\r\n- W: wrap，ring 每走一圈翻转一次\r\n- chunkIndex: chunk 编号\r\n- localIndex: chunk 上的 byte 编号\r\n- 对给定的 idxA, idxB，若 idxA 为键值对 idx，idxB 为 bucket.idx，判断 idxA 键值对仍有效（未被 ring 写入覆盖）的方法：`int64(idxA - idxB) < 0` (isValid).\r\n\r\n#### 写入流程\r\n1. `Set(k, v []byte)` 得到待写入键值对\r\n1. 计算 `hash = xxhash3(k)`\r\n1. 计算 bucket 编号 `bktIdx = hash%512`，对 bucket[bktIdx] 进行操作\r\n1. 获取 bucket 写入锁\r\n1. 检查 hash 是否已存在 map 中，如果是，使旧键值对所在 chunk.useCount--，若归零则将 chunk 放回 chunkPool\r\n1. 检查 idx 能否写入键值对:\r\n 1. 若当前 chunk 剩余空间足以写入键值对记录，则下一步\r\n 1. 若当前 chunk 剩余空间不足:\r\n 1. 若当前 chunk 非最后一个，前进至下一 chunk 并下一步\r\n 1. 若当前 chunk 是最后一个，回到第一个 chunk，翻转 idx.W，清除 map 中所有被覆盖键值对记录（根据 isValid 判断，删除键值对记录同时所在 chunk.useCount--，若归零则将 chunk 放回 chunkPool），再下一步\r\n1. 若当前 chunk 未 allocate: 从 chunkPool 获取空闲 chunk 并分配给当前位置\r\n1. 写入\r\n1. 更新 bucket.idx\r\n\r\n#### 读取流程\r\n1. `Get(dst, k []byte)` 得到键\r\n1. 计算 `hash = xxhash3(k)`\r\n1. 计算 bucket 编号 `bktIdx = hash%512`，对 bucket[bktIdx] 进行操作\r\n1. 获取 bucket 读取锁\r\n1. 检查 hash 是否在 map 中，若在则下一步，否则 miss\r\n1. 根据 isValid 检查 idx 是否有效，若无效则 miss\r\n1. 根据 idx 计算 chunkIdx, localIdx\r\n1. 读取 kLen, vLen, K，若 k != K，则哈希碰撞\r\n1. 追加值到 dst\r\n\r\n#### 删除流程\r\n1. `Del(k []byte)` 得到键\r\n1. 计算 `hash = xxhash3(k)`\r\n1. 计算 bucket 编号 `bktIdx = hash%512`，对 bucket[bktIdx] 进行操作\r\n1. 获取 bucket 写入锁\r\n1. 检查 hash 是否在 map 中，若否则 nop\r\n1. 若记录 idx 所在 chunk 已分配，chunk.useCount--，若归零则将 chunk 放回 chunkPool\r\n1. 从 map 中删除 hash 对应的记录\r\n\r\n#### 迭代器流程\r\n1. Cache 迭代器桥接 512 个 bucket 迭代器\r\n2. Bucket 迭代器创建时复制 map[hash]idx，每次 `GetNext(kDst, vDst []byte)` 获取读取锁，检验 idx isValid 并检验 hash 是否相符，若满足则追加 KV，否则跳过该记录使用下一条记录，直到没有更多记录\r\n\r\n要困死了 🫠","cooked":"

造轮子笑传之 chunked cache bucket

起因是我需要 fastcache，但是 fastcache 没有迭代器，所以干脆自己仿制一个就当学习了

Public Interface

\n\n

 type Cache\n    func New(size int) *Cache\n    func (c *Cache) Del(k []byte)\n    func (c *Cache) Get(dst, k []byte) (res []byte)\n    func (c *Cache) Has(k []byte) (found bool)\n    func (c *Cache) HasGet(dst, k []byte) (res []byte, found bool)\n    func (c *Cache) Iterator() *Iterator\n    func (c *Cache) LoadStats(stats *Stats)\n    func (c *Cache) Reset()\n    func (c *Cache) Set(k, v []byte)\ntype Iterator\n    func (it *Iterator) GetNext(kDst, vDst []byte) ([]byte, []byte, bool)\ntype Stats\n

架构

┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐\n│                 │ │ map[hash]idx    │ │                 │\n│ [512]bucket◄────┼─┤ [n]chunks◄──────┼─┤ [65536]byte     │\n│                 │ │     idx         │ │   useCount      │\n└─────────────────┘ └─────────────────┘ └─────────────────┘\n      cache               bucket              chunk        \n

Cache 分 512 个 bucket，每个 bucket 有独立的读写锁
每个 bucket 是由 n 个 chunk 构成的 ring buffer，idx 记录下一次写入位置，map[hash]idx 记录现有键值对位置
每个 chunk 是 [65536]byte
n 可根据 Cache 大小进行调整，至少为 1，整个 Cache 至少 65536*1*512 B = 32 MiB

键值对格式

 0                   1                   2                   3\n 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|          kLen (uint16le)      |          vLen (uint16le)      |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                                                               |\n+                        k (kLen bytes)                         +\n|                                                               |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                                                               |\n+                        v (vLen bytes)                         +\n|                                                               |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n

键值对不可跨 chunk 存在，键值对记录最长为 65536 bytes,因此 kLen, vLen 可用 uint16 记录

idx 格式

 0                   1                   2                   3\n 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|W|                     chunkIndex (47 bits)                    |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|     chunkIndex (cont.)        |      localIndex (16 bits)     |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n

W: wrap，ring 每走一圈翻转一次
chunkIndex: chunk 编号
localIndex: chunk 上的 byte 编号
对给定的 idxA, idxB，若 idxA 为键值对 idx，idxB 为 bucket.idx，判断 idxA 键值对仍有效（未被 ring 写入覆盖）的方法：int64(idxA - idxB) < 0 (isValid).

写入流程

Set(k, v []byte) 得到待写入键值对
计算 hash = xxhash3(k)
计算 bucket 编号 bktIdx = hash%512，对 bucket[bktIdx] 进行操作
获取 bucket 写入锁
检查 hash 是否已存在 map 中，如果是，使旧键值对所在 chunk.useCount--，若归零则将 chunk 放回 chunkPool
检查 idx 能否写入键值对:\n
1. 若当前 chunk 剩余空间足以写入键值对记录，则下一步
2. 若当前 chunk 剩余空间不足:\n
  1. 若当前 chunk 非最后一个，前进至下一 chunk 并下一步
  2. 若当前 chunk 是最后一个，回到第一个 chunk，翻转 idx.W，清除 map 中所有被覆盖键值对记录（根据 isValid 判断，删除键值对记录同时所在 chunk.useCount--，若归零则将 chunk 放回 chunkPool），再下一步
  \n
\n
若当前 chunk 未 allocate: 从 chunkPool 获取空闲 chunk 并分配给当前位置
写入
更新 bucket.idx

读取流程

Get(dst, k []byte) 得到键
计算 hash = xxhash3(k)
计算 bucket 编号 bktIdx = hash%512，对 bucket[bktIdx] 进行操作
获取 bucket 读取锁
检查 hash 是否在 map 中，若在则下一步，否则 miss
根据 isValid 检查 idx 是否有效，若无效则 miss
根据 idx 计算 chunkIdx, localIdx
读取 kLen, vLen, K，若 k != K，则哈希碰撞
追加值到 dst

删除流程

Del(k []byte) 得到键
计算 hash = xxhash3(k)
计算 bucket 编号 bktIdx = hash%512，对 bucket[bktIdx] 进行操作
获取 bucket 写入锁
检查 hash 是否在 map 中，若否则 nop
若记录 idx 所在 chunk 已分配，chunk.useCount--，若归零则将 chunk 放回 chunkPool
从 map 中删除 hash 对应的记录

迭代器流程

Cache 迭代器桥接 512 个 bucket 迭代器
Bucket 迭代器创建时复制 map[hash]idx，每次 GetNext(kDst, vDst []byte) 获取读取锁，检验 idx isValid 并检验 hash 是否相符，若满足则追加 KV，否则跳过该记录使用下一条记录，直到没有更多记录

要困死了 🫠

\n","createdAt":"2026-08-06T15:34:13.465Z","updatedAt":"2026-08-07T05:23:01.452Z","replyToPostNumber":null,"replyCount":0,"likeCount":1,"wordCount":982,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"❤️","count":1,"users":[{"id":8040,"username":"dvm","name":"Dvm"}],"hasReacted":false}],"wiki":false,"version":2,"publicVersion":2,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72980,"postNumber":28,"raw":"和 cache lab 搏斗一天了，还是做不到满分吗？力竭了 🫠\r\n![image.png|574x500](/uploads/image/45/5d/455d3411f8e07a1e01c36e0290fffe8f8b5b69451c3e7fbc5a4a0c0001cb1ee6.avif)","cooked":"

和 cache lab 搏斗一天了，还是做不到满分吗？力竭了 🫠
\n $\"image.png\"$

\n","createdAt":"2026-08-08T12:53:30.106Z","updatedAt":"2026-08-08T12:53:30.106Z","replyToPostNumber":null,"replyCount":1,"likeCount":1,"wordCount":23,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"❤️","count":1,"users":[{"id":8040,"username":"dvm","name":"Dvm"}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72984,"postNumber":29,"raw":"终于成了 😋\r\n\r\nhttps://github.com/HT4w5/csim\r\n\r\n[报告](https://ht4w5.github.io/csim)","cooked":"

终于成了 😋

\n\n

报告

\n","createdAt":"2026-08-08T15:50:42.669Z","updatedAt":"2026-08-08T15:51:33.597Z","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":10,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":72985,"postNumber":30,"raw":"之前 64x64 一直不满，是因为 tiled8sb4 写入 4x4 块的顺序不是最优，导致多了 400 多个 miss\r\n![image.png|690x372](/uploads/image/3b/3e/3b3ebba941943a67a1db39c298c999860d57f57ca77c5ff0b6f69dab523eb555.avif)\r\n（左：tiled8sb4rev2，右：tiled8sb4）","cooked":"

之前 64x64 一直不满，是因为 tiled8sb4 写入 4x4 块的顺序不是最优，导致多了 400 多个 miss
\n $\"image.png\"$
\n（左：tiled8sb4rev2，右：tiled8sb4）

\n","createdAt":"2026-08-08T15:58:14.589Z","updatedAt":"2026-08-08T15:59:29.983Z","replyToPostNumber":28,"replyCount":0,"likeCount":0,"wordCount":38,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73083,"postNumber":31,"raw":"[quote=\"IO, post:13, topic:15591\"]\r\n不支持progrssive decoding\r\n[/quote]\r\n\r\n最近突然发现我的Firefox浏览器上，AVIF也可以Progrssive rendering了😮，是我的错觉吗？\r\n![Screenshot_20260812_094942.png|300x150](/uploads/image/dc/9c/dc9c928324d44bc6a98564c12872984f4c94a7525c3811fac9f01eb4735010af.avif)","cooked":"

\n\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\tIO:\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t

\n
不支持progrssive decoding
\n

\n\t\t\t\t\t\t

最近突然发现我的Firefox浏览器上，AVIF也可以Progrssive rendering了😮，是我的错觉吗？
\n $\"Screenshot_20260812_094942.png\"$

\n","createdAt":"2026-08-12T01:52:45.251Z","updatedAt":"2026-08-12T01:58:01.381Z","replyToPostNumber":13,"replyCount":1,"likeCount":0,"wordCount":38,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":4668,"username":"IO","name":null,"avatarUrl":"/uploads/image/86/f2/86f267c0aaa73835a999b7cc3a7e7a701605574eda9a2c2a56f5ddbbba64d1a9.avif","admin":true,"moderator":false,"signatureRaw":"","signatureCooked":null}},{"id":73092,"postNumber":32,"raw":"似乎是 bug，这张图实际上是 `png`，只是 URL 后缀为 `.avif` 😇\r\n\r\n`avif` 似乎有某种 experimental progressive rendering 叫 layered\r\n\r\n>The AV1 Image File Format also supports multi-layer images as specified in [AV1] to be stored both in image items and image sequences. The AV1 Image File Format supports progressive image decoding through layered images.\r\n\r\nhttps://aomediacodec.github.io/av1-avif/#general\r\n\r\n>This creates two layers, one at 1/8th resolution and 'minimum' quality, and one at full resolution and good quality. These layers work similarly to frames of a video, with the second layer using data from the first to save space (a P-frame in video terms).\r\n\r\nhttps://jakearchibald.com/2025/present-and-future-of-progressive-image-rendering/#avif\r\n\r\n似乎表现为先渲染低分辨率的 layer，目前 Firefox 未支持\r\n\r\nhttps://bugzilla.mozilla.org/show_bug.cgi?id=1712813","cooked":"

似乎是 bug，这张图实际上是 png，只是 URL 后缀为 .avif 😇

avif 似乎有某种 experimental progressive rendering 叫 layered

\n
The AV1 Image File Format also supports multi-layer images as specified in [AV1] to be stored both in image items and image sequences. The AV1 Image File Format supports progressive image decoding through layered images.
\n

\n\n

\n
This creates two layers, one at 1/8th resolution and 'minimum' quality, and one at full resolution and good quality. These layers work similarly to frames of a video, with the second layer using data from the first to save space (a P-frame in video terms).
\n

\n\n

似乎表现为先渲染低分辨率的 layer，目前 Firefox 未支持

\n","createdAt":"2026-08-12T13:43:08.709Z","updatedAt":"2026-08-16T19:16:19.946Z","replyToPostNumber":31,"replyCount":0,"likeCount":0,"wordCount":134,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"😂","count":1,"users":[{"id":4668,"username":"IO","name":null}],"hasReacted":false}],"wiki":false,"version":2,"publicVersion":2,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73119,"postNumber":33,"raw":"有没有门友做过 malloclab，你们是如何测得 AVG_LIBC_THRUPUT 的？\r\n\r\n我用 libc malloc 实现要求的接口并执行 mdriver 得到的值（41800 Kops/s）比给出的参考值（400 Kops/s）大得多，自己的 malloc 实现在 AVG_LIBC_THRUPUT=41800e3 下效率分只有 1/100 🥹\r\n\r\n是因为我的实现太垃圾还是 AVG_LIBC_THRUPUT 的评估有问题 🤔","cooked":"

有没有门友做过 malloclab，你们是如何测得 AVG_LIBC_THRUPUT 的？

我用 libc malloc 实现要求的接口并执行 mdriver 得到的值（41800 Kops/s）比给出的参考值（400 Kops/s）大得多，自己的 malloc 实现在 AVG_LIBC_THRUPUT=41800e3 下效率分只有 1/100 🥹

是因为我的实现太垃圾还是 AVG_LIBC_THRUPUT 的评估有问题 🤔

\n","createdAt":"2026-08-14T12:53:10.617Z","updatedAt":"2026-08-14T12:53:10.617Z","replyToPostNumber":null,"replyCount":2,"likeCount":0,"wordCount":88,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73123,"postNumber":34,"raw":"好消息：我是傻逼，文档里写了 `mdriver -l` 即可评估 libc malloc\r\n坏消息：我是傻逼，`mdriver -l` 测得的 `AVG_LIBC_THRUPUT` 高达 48000e3 🤡","cooked":"

好消息：我是傻逼，文档里写了 mdriver -l 即可评估 libc malloc
\n坏消息：我是傻逼，mdriver -l 测得的 AVG_LIBC_THRUPUT 高达 48000e3 🤡

\n","createdAt":"2026-08-15T08:42:24.409Z","updatedAt":"2026-08-15T08:42:24.409Z","replyToPostNumber":33,"replyCount":0,"likeCount":0,"wordCount":40,"liked":false,"bookmarked":false,"polls":[],"reactions":[{"reaction":"😂","count":1,"users":[{"id":4668,"username":"IO","name":null}],"hasReacted":false}],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73147,"postNumber":35,"raw":"还有暗广\r\n```\r\n$ ./mdriver-rbt -lv\r\nTeam Name:ateam\r\nMember 1 :Harry Bovik:bovik@cs.cmu.edu\r\nUsing default tracefiles in traces/\r\nMeasuring performance with gettimeofday().\r\n\r\nResults for libc malloc:\r\ntrace valid util ops secs Kops\r\n 0 yes 0% 5694 0.000172 33162\r\n 1 yes 0% 5848 0.000126 46376\r\n 2 yes 0% 6648 0.000358 18549\r\n 3 yes 0% 5380 0.000379 14188\r\n 4 yes 0% 14400 0.000201 71820\r\n 5 yes 0% 4800 0.000327 14688\r\n 6 yes 0% 4800 0.000257 18699\r\n 7 yes 0% 12000 0.000134 89686\r\n 8 yes 0% 24000 0.000272 88235\r\n 9 yes 0% 14401 0.000373 38598\r\n10 yes 0% 14401 0.000082175622\r\nTotal 0% 112372 0.002680 41925\r\n\r\nResults for mm malloc:\r\ntrace valid util ops secs Kops\r\n 0 yes 99% 5694 0.000131 43599\r\n 1 yes 100% 5848 0.000142 41270\r\n 2 yes 99% 6648 0.000161 41395\r\n 3 yes 100% 5380 0.000130 41480\r\n 4 yes 99% 14400 0.000113127321\r\n 5 yes 96% 4800 0.000328 14625\r\n 6 yes 95% 4800 0.000322 14930\r\n 7 yes 55% 12000 0.000199 60211\r\n 8 yes 50% 24000 0.000487 49271\r\n 9 yes 68% 14401 0.000170 84563\r\n10 yes 76% 14401 0.000086168236\r\nTotal 85% 112372 0.002268 49553\r\n\r\nPerf index = 51 (util) + 40 (thru) = 91/100\r\n```","cooked":"

还有暗广

$ ./mdriver-rbt -lv\nTeam Name:ateam\nMember 1 :Harry Bovik:bovik@cs.cmu.edu\nUsing default tracefiles in traces/\nMeasuring performance with gettimeofday().\n\nResults for libc malloc:\ntrace  valid  util     ops      secs  Kops\n 0       yes    0%    5694  0.000172 33162\n 1       yes    0%    5848  0.000126 46376\n 2       yes    0%    6648  0.000358 18549\n 3       yes    0%    5380  0.000379 14188\n 4       yes    0%   14400  0.000201 71820\n 5       yes    0%    4800  0.000327 14688\n 6       yes    0%    4800  0.000257 18699\n 7       yes    0%   12000  0.000134 89686\n 8       yes    0%   24000  0.000272 88235\n 9       yes    0%   14401  0.000373 38598\n10       yes    0%   14401  0.000082175622\nTotal           0%  112372  0.002680 41925\n\nResults for mm malloc:\ntrace  valid  util     ops      secs  Kops\n 0       yes   99%    5694  0.000131 43599\n 1       yes  100%    5848  0.000142 41270\n 2       yes   99%    6648  0.000161 41395\n 3       yes  100%    5380  0.000130 41480\n 4       yes   99%   14400  0.000113127321\n 5       yes   96%    4800  0.000328 14625\n 6       yes   95%    4800  0.000322 14930\n 7       yes   55%   12000  0.000199 60211\n 8       yes   50%   24000  0.000487 49271\n 9       yes   68%   14401  0.000170 84563\n10       yes   76%   14401  0.000086168236\nTotal          85%  112372  0.002268 49553\n\nPerf index = 51 (util) + 40 (thru) = 91/100\n

\n","createdAt":"2026-08-16T16:34:05.381Z","updatedAt":"2026-08-16T16:34:05.381Z","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":193,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73148,"postNumber":36,"raw":"这个方法感觉到头了 🫠\r\n```\r\n$ make && ./mdriver-rbt -v\r\nmake: Nothing to be done for 'all'.\r\nTeam Name:ateam\r\nMember 1 :Harry Bovik:bovik@cs.cmu.edu\r\nUsing default tracefiles in traces/\r\nMeasuring performance with gettimeofday().\r\n\r\nResults for mm malloc:\r\ntrace valid util ops secs Kops\r\n 0 yes 99% 5694 0.000140 40642\r\n 1 yes 99% 5848 0.000149 39354\r\n 2 yes 99% 6648 0.000166 39976\r\n 3 yes 100% 5380 0.000122 43990\r\n 4 yes 99% 14400 0.000107134705\r\n 5 yes 96% 4800 0.000323 14874\r\n 6 yes 95% 4800 0.000326 14724\r\n 7 yes 96% 12000 0.000231 51903\r\n 8 yes 85% 24000 0.000497 48261\r\n 9 yes 80% 14401 0.000119121119\r\n10 yes 89% 14401 0.000092155686\r\nTotal 94% 112372 0.002273 49442\r\n\r\nPerf index = 57 (util) + 40 (thru) = 97/100\r\n```\r\n\r\n本来还想拿 PKU ICS 的 malloclab traces 测一下，结果发现 PKU 的 driver 是 64 位的 :tieba_hehe:\r\n本来第一遍我就按照 64 位写的结果最后才看到 Makefile 里写的 `-m32`，真是草了","cooked":"

这个方法感觉到头了 🫠

$ make && ./mdriver-rbt -v\nmake: Nothing to be done for 'all'.\nTeam Name:ateam\nMember 1 :Harry Bovik:bovik@cs.cmu.edu\nUsing default tracefiles in traces/\nMeasuring performance with gettimeofday().\n\nResults for mm malloc:\ntrace  valid  util     ops      secs  Kops\n 0       yes   99%    5694  0.000140 40642\n 1       yes   99%    5848  0.000149 39354\n 2       yes   99%    6648  0.000166 39976\n 3       yes  100%    5380  0.000122 43990\n 4       yes   99%   14400  0.000107134705\n 5       yes   96%    4800  0.000323 14874\n 6       yes   95%    4800  0.000326 14724\n 7       yes   96%   12000  0.000231 51903\n 8       yes   85%   24000  0.000497 48261\n 9       yes   80%   14401  0.000119121119\n10       yes   89%   14401  0.000092155686\nTotal          94%  112372  0.002273 49442\n\nPerf index = 57 (util) + 40 (thru) = 97/100\n

本来还想拿 PKU ICS 的 malloclab traces 测一下，结果发现 PKU 的 driver 是 64 位的 $\":tieba_hehe:\"$
\n本来第一遍我就按照 64 位写的结果最后才看到 Makefile 里写的 -m32，真是草了

\n","createdAt":"2026-08-16T19:08:53.106Z","updatedAt":"2026-08-16T19:12:08.770Z","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":182,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73149,"postNumber":37,"raw":"[quote=\"stateless_trash_can, post:33, topic:15591\"]\r\n自己的 malloc 实现在 AVG_LIBC_THRUPUT=41800e3 下效率分只有 1/100\r\n[/quote]\r\n\r\nhttps://github.com/HT4w5/cmu15213-labs/blob/main/malloclab/mm_lifo.c\r\n\r\n这里当时使用 LIFO free list + 每 n 次请求后统计 size 区间众数和 95% size，得到 default_size 和 max_size，然后限制 coalescing 至 max_size 以及大于 default_size 才 split，但是事实证明完全没用\r\n\r\n由于每次统计都需要 qsort 20 个 call bucket，同时 n 太小，导致效率极低（如果 n 太大则在有限的 trace 内无法触发统计），最终也无法知道这种方式在长期运行下是否有意义 🤔\r\n\r\n话说这个算法还是学习自 [github.com/valyala/bytebufferpool](https://github.com/valyala/bytebufferpool)\r\n\r\n具体是这里\r\n\r\nhttps://github.com/valyala/bytebufferpool/blob/18533face0dfe7042f8157bba9010bd7f8df54b1/pool.go#L81","cooked":"

\n\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\tstateless_trash_can:\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t

\n
自己的 malloc 实现在 AVG_LIBC_THRUPUT=41800e3 下效率分只有 1/100
\n

\n\t\t\t\t\t\t

\n\n

这里当时使用 LIFO free list + 每 n 次请求后统计 size 区间众数和 95% size，得到 default_size 和 max_size，然后限制 coalescing 至 max_size 以及大于 default_size 才 split，但是事实证明完全没用

由于每次统计都需要 qsort 20 个 call bucket，同时 n 太小，导致效率极低（如果 n 太大则在有限的 trace 内无法触发统计），最终也无法知道这种方式在长期运行下是否有意义 🤔

话说这个算法还是学习自 github.com/valyala/bytebufferpool

具体是这里

\n","createdAt":"2026-08-16T19:23:40.927Z","updatedAt":"2026-08-16T19:28:48.542Z","replyToPostNumber":33,"replyCount":0,"likeCount":0,"wordCount":162,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":2,"publicVersion":2,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73150,"postNumber":38,"raw":"[quote=\"stateless_trash_can, post:36, topic:15591\"]\r\n```\r\n$ make && ./mdriver-rbt -v\r\nmake: Nothing to be done for 'all'.\r\nTeam Name:ateam\r\nMember 1 :Harry Bovik:bovik@cs.cmu.edu\r\nUsing default tracefiles in traces/\r\nMeasuring performance with gettimeofday().\r\n\r\nResults for mm malloc:\r\ntrace valid util ops secs Kops\r\n 0 yes 99% 5694 0.000140 40642\r\n 1 yes 99% 5848 0.000149 39354\r\n 2 yes 99% 6648 0.000166 39976\r\n 3 yes 100% 5380 0.000122 43990\r\n 4 yes 99% 14400 0.000107134705\r\n 5 yes 96% 4800 0.000323 14874\r\n 6 yes 95% 4800 0.000326 14724\r\n 7 yes 96% 12000 0.000231 51903\r\n 8 yes 85% 24000 0.000497 48261\r\n 9 yes 80% 14401 0.000119121119\r\n10 yes 89% 14401 0.000092155686\r\nTotal 94% 112372 0.002273 49442\r\n\r\nPerf index = 57 (util) + 40 (thru) = 97/100\r\n```\r\n[/quote]\r\n\r\n\r\n# mm_rbt\r\n使用红黑树 free list 实现的 malloc\r\n\r\nhttps://github.com/HT4w5/cmu15213-labs/blob/main/malloclab/mm_rbt.c\r\n\r\n### block\r\n除全局数据以外，所有的堆内存都由 block 构成\r\n\r\nblock 的结构：\r\n```c\r\nstruct size_header {\r\n uint32_t psize;\r\n uint32_t csize;\r\n};\r\n\r\nstruct block_header {\r\n struct size_header size;\r\n uint32_t color;\r\n union {\r\n struct {\r\n union {\r\n struct {\r\n struct block_header* left;\r\n struct block_header* right;\r\n };\r\n struct block_header* child[2];\r\n };\r\n struct block_header* parent;\r\n };\r\n struct {\r\n struct block_header* root;\r\n struct block_header* max;\r\n struct block_header* min;\r\n };\r\n };\r\n};\r\n```\r\n#### size_header\r\nsize_header 记录上一个 block（在堆上）和当前 block 的大小和是否 free\r\n`uint32_t size_header::psize` 和 `uint32_t size_header::csize`:\r\n```shell\r\n 0 1 2 3\r\n 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n|F| block_size (31 bits) |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n```\r\n- F: 块是否空闲\r\n- block_size: 块的有效空间\r\n\r\n>由于 size_header 内有上一 block 的大小/空闲信息，堆顶必须维护一个 stub size_header 记录最后一个 block 的信息，以方便新 block 的创建\r\n\r\n#### block_header\r\n当 block 空闲时，block_header 的全部内容有效（否则只有 size_header 有效）\r\n此时 block 为红黑树节点\r\n\r\n当 block 为红黑树 header 节点时：\r\n```shell\r\n 0 1 2 3\r\n 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| psize |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| csize |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| color |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| root |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| max |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| min |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n```\r\n- root: 红黑树根节点，当树空时指向 sentinel 节点\r\n- max/min: 指向最大/最小节点，用于 O(1) max/min 查询\r\n- color: 节点颜色，此时恒为 black\r\n\r\n当 block 为红黑树普通节点时：\r\n```shell\r\n 0 1 2 3\r\n 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| psize |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| csize |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| color |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| left / child[0] |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| right / child[1] |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n| parent |\r\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\r\n```\r\n- left / child[0]: 左子树\r\n- right / child[1]: 右子树\r\n- parent: 父节点\r\n\r\n#### 红黑树\r\n参考 https://en.wikipedia.org/wiki/Red%E2%80%93black_tree#Implementation\r\n\r\n>红黑树是一种自平衡二叉查找树\r\n\r\n>它的操作有着良好的最坏情况运行时间\r\n\r\n\r\n#### mm_malloc\r\n1. 查询红黑树，找到最合适的 free block，若未找到则进入 2.\r\n 1.1 若 free_size - size >= block_header 大小，则拆分 free block 为此次分配的 block 和新的 free block 并放回红黑树\r\n2. 若未找到，若 size <= 128 进入 3.，在堆上新建合适大小的 block 并返回\r\n3. 小块预分配：扩展 size 到最近的 power of 2，并在此次分配的同时预分配 x(size) 个大小为 size 的 free block 加入到红黑树中（这样可以将类似的小块集中在一起，缓解小块带来的 external fragmentation）\r\n\r\n>查询红黑树可以先对比 O(1) max 节点的大小，如果 size 比这个还大就可以直接新建了\r\n\r\n>小块预分配让我的分数从 91 增长到 97，不过感觉有点面向测试设计了（\r\n\r\n#### mm_free\r\n1. 向更高地址查询空闲 block，直到遇到非空闲 block\r\n2. 向低位地址查询空闲 block，直到遇到非空闲 block\r\n3. 将所有查询到的空闲 block 从红黑树中删除，并合并为一个空闲 block 重新插入红黑树\r\n\r\n> 这里其实可以限制最大合并大小，但是实际上没有这样做，也许可以进一步探索？（因为效率分满了所以好像没有意义）\r\n\r\n#### mm_realloc\r\n1. 若 size > block_size，进入 2.，若 block_size - size >= block_header 大小，则拆分原 block 为返回的 block 和新的 free block 并放回红黑树，返回原 block\r\n2. 向高位地址统计空闲 block，若总空闲空间 < size 则回落至 malloc/copy/free，否则进入 3.\r\n3. 若空闲空间 > size 则停止统计，将统计的空间 block 从红黑树中删除，合并为一个空闲 block\r\n4. 若空闲 block 大小 - size > block_header 大小，则拆分空闲 block 为分配 block 和新的空闲 block 并放回红黑树\r\n5. 返回原 block","cooked":"

\n\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\tstateless_trash_can:\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t

\n\t\t\t\t\t\t\t

$ make && ./mdriver-rbt -v\nmake: Nothing to be done for 'all'.\nTeam Name:ateam\nMember 1 :Harry Bovik:bovik@cs.cmu.edu\nUsing default tracefiles in traces/\nMeasuring performance with gettimeofday().\n\nResults for mm malloc:\ntrace  valid  util     ops      secs  Kops\n 0       yes   99%    5694  0.000140 40642\n 1       yes   99%    5848  0.000149 39354\n 2       yes   99%    6648  0.000166 39976\n 3       yes  100%    5380  0.000122 43990\n 4       yes   99%   14400  0.000107134705\n 5       yes   96%    4800  0.000323 14874\n 6       yes   95%    4800  0.000326 14724\n 7       yes   96%   12000  0.000231 51903\n 8       yes   85%   24000  0.000497 48261\n 9       yes   80%   14401  0.000119121119\n10       yes   89%   14401  0.000092155686\nTotal          94%  112372  0.002273 49442\n\nPerf index = 57 (util) + 40 (thru) = 97/100\n

\n\t\t\t\t\t\t

mm_rbt

使用红黑树 free list 实现的 malloc

\n\n

block

除全局数据以外，所有的堆内存都由 block 构成

block 的结构：

struct size_header {\n    uint32_t psize;\n    uint32_t csize;\n};\n\nstruct block_header {\n    struct size_header size;\n    uint32_t color;\n    union {\n        struct {\n            union {\n                struct {\n                    struct block_header* left;\n                    struct block_header* right;\n                };\n                struct block_header* child[2];\n            };\n            struct block_header* parent;\n        };\n        struct {\n            struct block_header* root;\n            struct block_header* max;\n            struct block_header* min;\n        };\n    };\n};\n

size_header

size_header 记录上一个 block（在堆上）和当前 block 的大小和是否 free
\nuint32_t size_header::psize 和 uint32_t size_header::csize:

 0                   1                   2                   3\n 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|F|                     block_size (31 bits)                    |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n

F: 块是否空闲
block_size: 块的有效空间

\n
由于 size_header 内有上一 block 的大小/空闲信息，堆顶必须维护一个 stub size_header 记录最后一个 block 的信息，以方便新 block 的创建
\n

block_header

当 block 空闲时，block_header 的全部内容有效（否则只有 size_header 有效）
\n此时 block 为红黑树节点

当 block 为红黑树 header 节点时：

 0                   1                   2                   3\n 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                            psize                              |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                            csize                              |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                            color                              |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                             root                              |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                             max                               |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                             min                               |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n

root: 红黑树根节点，当树空时指向 sentinel 节点
max/min: 指向最大/最小节点，用于 O(1) max/min 查询
color: 节点颜色，此时恒为 black

当 block 为红黑树普通节点时：

 0                   1                   2                   3\n 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                            psize                              |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                            csize                              |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                            color                              |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                      left / child[0]                          |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                      right / child[1]                         |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n|                            parent                             |\n+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+\n

left / child[0]: 左子树
right / child[1]: 右子树
parent: 父节点

红黑树

参考 Red–black tree - Wikipedia

\n
红黑树是一种自平衡二叉查找树
\n

\n
它的操作有着良好的最坏情况运行时间
\n

mm_malloc

查询红黑树，找到最合适的 free block，若未找到则进入 2.
\n1.1 若 free_size - size >= block_header 大小，则拆分 free block 为此次分配的 block 和新的 free block 并放回红黑树
若未找到，若 size <= 128 进入 3.，在堆上新建合适大小的 block 并返回
小块预分配：扩展 size 到最近的 power of 2，并在此次分配的同时预分配 x(size) 个大小为 size 的 free block 加入到红黑树中（这样可以将类似的小块集中在一起，缓解小块带来的 external fragmentation）

\n
查询红黑树可以先对比 O(1) max 节点的大小，如果 size 比这个还大就可以直接新建了
\n

\n
小块预分配让我的分数从 91 增长到 97，不过感觉有点面向测试设计了（
\n

mm_free

向更高地址查询空闲 block，直到遇到非空闲 block
向低位地址查询空闲 block，直到遇到非空闲 block
将所有查询到的空闲 block 从红黑树中删除，并合并为一个空闲 block 重新插入红黑树

\n
这里其实可以限制最大合并大小，但是实际上没有这样做，也许可以进一步探索？（因为效率分满了所以好像没有意义）
\n

mm_realloc

若 size > block_size，进入 2.，若 block_size - size >= block_header 大小，则拆分原 block 为返回的 block 和新的 free block 并放回红黑树，返回原 block
向高位地址统计空闲 block，若总空闲空间 < size 则回落至 malloc/copy/free，否则进入 3.
若空闲空间 > size 则停止统计，将统计的空间 block 从红黑树中删除，合并为一个空闲 block
若空闲 block 大小 - size > block_header 大小，则拆分空闲 block 为分配 block 和新的空闲 block 并放回红黑树
返回原 block

\n","createdAt":"2026-08-16T20:14:38.685Z","updatedAt":"2026-08-16T20:34:13.532Z","replyToPostNumber":null,"replyCount":2,"likeCount":0,"wordCount":1147,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":2,"publicVersion":2,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73152,"postNumber":39,"raw":"C 好恶心啊我要写 Go :innocent_mouse:","cooked":"

C 好恶心啊我要写 Go $\":innocent_mouse:\"$

\n","createdAt":"2026-08-16T20:26:57.249Z","updatedAt":"2026-08-16T20:26:57.249Z","replyToPostNumber":38,"replyCount":0,"likeCount":0,"wordCount":10,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73157,"postNumber":40,"raw":"说实在的红黑树只是照着文档写出来的，要是自己独立实现恐怕还是做不到，实在是惭愧","cooked":"

说实在的红黑树只是照着文档写出来的，要是自己独立实现恐怕还是做不到，实在是惭愧

\n","createdAt":"2026-08-17T04:45:16.761Z","updatedAt":"2026-08-17T04:45:16.761Z","replyToPostNumber":38,"replyCount":0,"likeCount":0,"wordCount":39,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73200,"postNumber":41,"raw":"Go 1.27 快发行吧我要用 uuid 😩","cooked":"

Go 1.27 快发行吧我要用 uuid 😩

\n","createdAt":"2026-08-19T14:05:01.147Z","updatedAt":"2026-08-19T14:05:01.147Z","replyToPostNumber":null,"replyCount":1,"likeCount":0,"wordCount":11,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73208,"postNumber":42,"raw":"何意味，刚说完就发了\r\n\r\nhttps://go.dev/blog/go1.27","cooked":"

何意味，刚说完就发了

\n","createdAt":"2026-08-20T04:34:55.284Z","updatedAt":"2026-08-20T04:43:16.491Z","replyToPostNumber":41,"replyCount":0,"likeCount":0,"wordCount":11,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":2,"publicVersion":2,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73217,"postNumber":43,"raw":"https://cursor.com/blog/git-at-any-scale","cooked":"","createdAt":"2026-08-21T00:34:59.327Z","updatedAt":"2026-08-21T00:34:59.327Z","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":1,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}},{"id":73227,"postNumber":44,"raw":"DeepSeek 出 vision 了\r\nhttps://api-docs.deepseek.com/guides/vision/","cooked":"

DeepSeek 出 vision 了

\n","createdAt":"2026-08-21T15:24:11.515Z","updatedAt":"2026-08-21T15:24:11.515Z","replyToPostNumber":null,"replyCount":0,"likeCount":0,"wordCount":5,"liked":false,"bookmarked":false,"polls":[],"reactions":[],"wiki":false,"version":1,"publicVersion":1,"deletedAt":null,"isAnonymous":false,"user":{"id":7830,"username":"stateless_trash_can","name":"无状态垃圾桶","avatarUrl":"/uploads/image/f4/69/f469375c27cd72c6970425e94b999be00fed8fe6234b977b14f5bdb65da24b4e.avif","admin":false,"moderator":false,"signatureRaw":"

","signatureCooked":"

"}}],"hasMore":false,"lastReadPostNumber":44,"hasUrlPostNumber":true}

SplitArgs

FSM

Go 实现

方法论

Ring Buffer Read/Write

结构体

读取

A

B

C

D

写入

A

B

C

D

ID Counter

SimpleIDCounter

RandomShardedIDCounter

跑分

Log file change watcher

接口

Polling 实现

FSNotify 实现

Tail

FSM

go maps delete and clone

sync.Pool 案例分析

slice.go - 切片池

slicePtr.go - 切片指针池

map.go - 表池

跑分

错误写法

正确写法

造轮子笑传之 chunked cache bucket

Public Interface

架构

键值对格式

idx 格式

写入流程

读取流程

删除流程

迭代器流程

mm_rbt

block

size_header

block_header

红黑树

mm_malloc

mm_free

mm_realloc

`sync.Pool` 案例分析

`slice.go` - 切片池

`slicePtr.go` - 切片指针池

`map.go` - 表池