HermitCrab analysis: defer the per-template clone in AnalysisAffixTemplateRule (1.054x wall-clock, not a step count)

Ouverte
  • Quelque chose ne va pas ?
#586 1 commentaire 0 réactions 0 personnes assignées Voir sur GitHub

Les mainteneurs répondent en général sous 1 jour

Personne n'a encore pris cette issue.

Évaluation

Difficulté
4/5
Temps estimé
3-5 jours
Accessibilité débutants
45/100
Type d'issue
Refactorisation
Clarté
Plutôt claire
Activité
Active
Stack technique
csharp
Domaine
performance

Piste de recherche

Start with the optimization record at docs/optimizations/deferred-template-clone.md and the disposition in docs/hermitcrab-perf-2026-09-disposition.md, then inspect AnalysisAffixTemplateRule.Apply and its emission points. First establish whether the target already clones lazily; measure steps or allocations and verify parity with a signature covering gloss, allomorph index, and syntactic feature structure. Done means the measured change is justified and parity holds.

Rédigé par le modèle d'indexation à partir du texte de l'issue.

Description

Hermit Crab performance

AnalysisAffixTemplateRule.Apply makes an internal engine clone of the input, freezes it once per template, and runs the slot DFS on the clone. The input is already frozen at every call site and slot rules never mutate it, so the DFS can run on the input directly. Only the two emission points that can pass the input through unchanged (ApplySlots terminal, ParallelApplySlots push) need a frozen clone. The traced path keeps the eager clone, because the trace manager stamps CurrentTrace onto the words it is given. The change is on the archived branch perf/hc-optimization-archive.

Measured. Clone-discard count (Amharic): 2,505 clones created, 1,659 discarded (66%). Wall-clock A/B on 45 Sena words, min of two interleaved rounds: 1.054x overall, 1.02x to 1.08x on heavy words. No step count was recorded. Parity was checked with WordAnalysisSignature, which on Sena compares morph counts only.

What a revival must do first. Re-measure in steps or allocations rather than wall time. Check parity with a gloss, allomorph-index and syntactic-feature-structure signature, not morph counts. The saving is allocation only, so a port that already clones lazily gains nothing.

Record: see the archived write-up below
Ledger: https://github.com/sillsdev/machine/pull/490#issuecomment-6701971116; disposition: https://github.com/sillsdev/machine/pull/490#issuecomment-6701971324
From #490 (closed).

Langage dominant
C#
Étoiles
29
Forks
16
Merge moyen
4 j 11 h
PR mergées (30 j)
14

Préparer son environnement

  • Aucun Dockerfile ni fichier Docker Compose
  • Propose un modèle de pull request
  • Aucun guide de contribution

Par où commencer

  1. Lisez l'issue en entier, puis le guide de contribution du projet.
  2. Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
  3. Forkez le dépôt et travaillez sur une branche.
  4. Ouvrez une pull request qui référence le numéro de l'issue.

Autres issues de sillsdev/machine

Toutes les issues de sillsdev/machine

Issues similaires

Plus d'issues C#

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.