HermitCrab: do literal-identity and natural-class restoration on hc.dll match the port's fixes? (unmeasured)

未关闭
  • 哪里不对？
#609 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
48/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
csharp
领域
backend, testing

调研方向

Read PanGloss CHANGELOG origin/main lines 202–230 and docs/hermitcrab-2026-10-21-search-cost-round.md lines 59–113, then inspect the C# implementation in hc.dll for the two described identity/restoration patterns. Replay both patterns and count steps; add conformance cases to #480 for any divergence, or record that C# already matches. The issue does not name a specific test or code entry point.

由索引模型根据 Issue 内容生成。

描述

Claude-drafted.

Two identity fixes cut the PanGloss port's work sharply, and nobody has replayed them on hc.dll. Affix literal segments match only themselves on large featureless inventories: 98.5% less work on 7 spot words. Deleted natural-class members restore only their class: 94.7% less work on a 7,206-word list. Both were measured on the port (HC-Rust) only, in its own work units.

What we know:

  • The port's changelog says the literal fix matches HermitCrab ("as in HermitCrab", 0.9.0, line 233). The class fix cites divergence entry 103 and makes no such claim (line 209).
  • The ledger says C# gives a featureless segment a StrRep EqualTo constraint, so its literal matches only itself. That is in the code at CharacterDefinitionTable.cs#L73-L74. For the class case the ledger says C# unifies against the class's StrRep union, with no code reference (inferred).
  • The port ran 18 lookups on the slowest word on C#'s own export, against 49 for C# and 1,655 on its own snapshot. That gap is what led to both fixes.

Not known: whether hc.dll's work steps match the port's after the fixes, for either pattern.

Proposed change (C#): replay the two patterns on hc.dll and count steps. Add conformance cases to the conformance suite for any divergence. If C# already behaves this way, record that.

Related: #480 (closed; its fixtures are now in the conformance suite). PanGloss: sillsdev/PanGloss#111 (the pointer issue for this one, open); the fixes are in PanGloss CHANGELOG 0.9.0 (divergence entries 090 and 103).

Next: replay both patterns on hc.dll and count work steps.

Evidence -- the port-side measurements
  • Literal identity (ledger row 24): 7 spot words 98.5% less work (2,365,602 to 35,913); worst variant 99.8% less (50,778,583 to 83,677); steps on full lists fell 46-55x. Parses identical on three private lists and on Amharic, Sena and Indonesian.
  • Natural-class restoration (ledger row 25): held list of 7,206 words 94.7% less work (303,966,311 to 16,220,190); the 16 words that hit a 1 s limit 99.9% less. Parses identical everywhere.
  • Cause of A: above 64 character definitions the port lost the literal's identity, so an enclitic matched any final segment. Cause of B: the port's re-inserted class node matched any segment and its root trie ignored the member set.
  • The three lookup counts (49, 18, 1,655) were supplied to that round and not reproduced by it.
Method rule -- how the bugs were found

Run the port on the oracle's own export of the same grammar and compare a cheap count, such as lookups per word. A large gap between the port's two inputs is a port bug. The round's rule: do this before measuring any algorithmic lead on a port.

Sources -- pinned
主要语言
C#
星标
29
派生
16
平均合并
6 天 7 小时
30 天内合并 PR
14

环境准备

  • 没有 Dockerfile 或 Docker Compose 文件
  • 有 Pull Request 模板
  • 没有贡献指南

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

sillsdev/machine 的其他 Issue

查看 sillsdev/machine 的全部 Issue

相似的 Issue

更多 C# Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。