replace_ops does not re-mint inputs[].grow_id — bulk-replayed autogrow slots inherit the source document's identities

未关闭
  • 哪里不对？
#1,035 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
1-2 天
新手友好度
35/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
技术栈
python
领域
cli

调研方向

Read docs/op-vocabulary-v1.md §8.8 and inspect replace_ops to trace how it remaps inputs[].link and handles inputs[].grow_id. Compare the op replay path with the apply_specs behavior described in the issue. The owner must decide whether to remap grow_id or document why it is carried through; done means that decision is implemented and the acceptance condition in the issue is met.

由索引模型根据 Issue 内容生成。

描述

bug

Problem

replace_ops re-mints every node and link identity, but it does not re-mint inputs[].grow_id on an autogrow slot. The grown slots in the emitted add_node payload keep the source document's link ids as their grow_id.

docs/op-vocabulary-v1.md §8.8 states the rule as "Identity is re-minted, never inherited… Every node and link gets a fresh mint_id and every interior reference (inputs[].link, outputs[].links, the links tuples) is remapped onto it." grow_id is an interior identity reference and is not in that list, so it is carried through verbatim.

Repro

Build a workflow whose BatchImagesNode autogrow base has three wired slots, then emit and replay a batch:

source document:  images.image0  link=1369904620697626  grow_id=1369904620697626
add_node payload: images.image0  link=4047451378665163  grow_id=1369904620697626   <-- stale
after apply_op:   images.image0  link=4047451378665163  grow_id=1369904620697626   <-- stale
after apply_specs: images.image0 link=<fresh>           grow_id=<fresh, == link>

The link is remapped; the grow_id is not. _apply_connect sets grow_id to the newly minted link id, which is why the spec path (which re-mints each node from the catalog and wires it with real connect ops) comes out consistent while the op path does not.

Impact

  1. Duplicate identity across replicas. grow_id is the convergence identity of a grown slot. Two documents seeded from the same template by replace_ops end up with grown slots that share a grow_id while holding different link ids — the duplicate-identity-on-first-merge hazard §1.5 / §8.6 exist to prevent. The whole reason replace_ops re-mints is to avoid exactly this.
  2. The two consumers of one artifact diverge. §8.8 promises the same array replays through apply_op and apply_specs. After the fix in #848 both paths reproduce the same slot names, wiring and link count, but their grow_id values differ in kind: the op path inherits the source's, the spec path mints fresh ones equal to the link id.

Expected

Either replace_ops remaps grow_id alongside inputs[].link, or §8.8 states explicitly that grow_id is carried verbatim and says why that is safe. The current state is neither — the doc enumerates what gets remapped and omits this field without comment, so the behaviour is undocumented rather than decided.

This touches frozen mint semantics, so it needs an owner ruling rather than a drive-by patch.

Provenance

Raised as a review finding on https://github.com/Comfy-Org/comfy-cli/pull/848 (tests/comfy_cli/command/test_workflow_edit.py line 2239) and verified by direct probe at head 8da9bf27ec8b0f56fab0ac23aad2695fbc777ed3. #848 deliberately does not assert grow_id parity between the two paths, so it neither fixes nor pins this; see #1034 for the separate _slot_ref defect that PR does fix.

Task Decomposition

Not decomposed: the investigation is complete and the remaining work is one owner decision followed by either a few-line remap in replace_ops or a documentation sentence in §8.8. Complexity tier: analytical (the decision), trivial (the change). Dependency: an owner ruling on frozen mint semantics. Acceptance: grow_id values after an apply_op replay are disjoint from the source document's, or §8.8 records why they need not be.

主要语言
Python
星标
1k
派生
176
平均合并
1 天 11 小时
30 天内合并 PR
87

环境准备

  • 没有 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

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。

Comfy-Org/comfy-cli 的其他 Issue

查看 Comfy-Org/comfy-cli 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。