Skip to content

pay down grandfathered scan findings, then remove the baseline [📡#443] #1493

Description

@donjor-agent

🤖 claude/anthropic claude-opus-5-5 (high)

Problem

The whole-series scan grandfathers legacy findings, and nothing ever pays them down.

Part Evidence
Baseline resolveBaselineKeys in scripts/fork-scan.ts: with --replay-of and --since equal to --target, a finding already on the trunk renders advisory
Origin c341365 fix(fork): sync battery fails only on findings the replay introduced; 197 legacy findings had turned every sync red
Fold proves tree-equal, so it never touches the files that carry a finding
PR CI scans only its own commits, so an old finding is never seen again
Result the 2026-10-02 sync reported 172 findings, all advisory

Inventory

Merge base 094fb23 (v0.0.45-nightly.20261001.2552), head dd89a3c:

vp run fork:scan --head origin/hyprws --target <merge-base> --since <merge-base> \
  --replay-of origin/hyprws --no-typecheck
historical: 172 finding(s) already on origin/hyprws against 094fb230e90…, advisory
Kind Count Owner
upstream-test: fork blocks in an upstream test file 93 28 commits
upstream-test: changed or removed upstream test lines 41 26 commits
replaced-export 10 6 commits
additive:tests: upstream test lines gone, no sibling 28 files, 132 lines tree-level
By domain
Domain upstream-test replaced-export
upstream-fixes 28 0
custom-agents 23 2
zmux-estate 21 0
multi-window 19 1
fork-meta 11 0
github-issues 11 0
worktrunk-hooks 7 1
workspace-files 4 1
backend-attach 2 5
browser-bookmarks 2 0
distribution 2 0
markdown-editing 2 0
thread-ordering 2 0
additive:tests area Files
apps/server 11
apps/web 8
apps/desktop 5
apps/mobile, packages/client-runtime, packages/contracts, scripts 1 each

Decision

  1. Pay each finding down in the commit that owns it.
  2. Remove the baseline: any finding refuses.
Finding Pay-down
Fork test block move to the .fork. sibling
Changed upstream test line restore it
Replaced export extend from a fork-owned sibling

Side defect, moot on removal: the gate compares --since and --target as strings, so a short sha against a full sha silently disables the baseline.
scripts/fork-ci.ts still prints replay-of in its header line; it goes with the baseline.

Close conditions

  • Inventory: the scan reports 0 findings
  • Baseline: resolveBaselineKeys removed
  • Guard: a pre-existing finding fails the scan

Origin: the 2026-10-02 fork sync reported 172 grandfathered findings as advisory.

Carried from #689

Triage on 2026-10-14 folded the upstream-tests invariant in.

  • In-place edits: 0 on the next walk

Activity

  1. changed the title [-]pay down grandfathered scan findings, then remove the baseline[/-] [+]pay down grandfathered scan findings, then remove the baseline [✨#689][/+] on Oct 2, 2026
  2. donjor-agent commented on Oct 2, 2026

    @donjor-agent
    Author

    🤖 claude/anthropic claude-opus-5-5 (high)

    Plan settled; work starts with the tooling batch.

    The inventory on 0e13bec is unchanged at 172 findings. They split into three groups:

    Group Findings Paid by
    History-only: a later mover already moved the block 69 blocks, 1 export one history fold that splits the movers into their owners
    False positive: a restored upstream case counts as a fork block 4 the scan rule (tooling batch)
    Live at the tip 20 blocks, 41 changed files, 9 exports fixup! PRs in seven file-cluster batches

    Decisions

    # Decision
    Movers fork-fold accepts a non-repair commit split across owners when each owner cites it under Squashes:
    Harness files DesktopWindow.test.ts and server.test.ts blocks move to register-function siblings; the stale deferral text goes
    Landing one PR per owner titled fixup! <owner subject>: squash landing collapses a multi-commit PR into one commit
    Fold cadence each batch folds at the next sync
    serverRuntimeState, EnvironmentResourceNotFoundReason decided in their batches: restore upstream's export, with no duplicated logic and no client-visible wire change left half done

    Order

    1. Tooling batch: the block-rule false positive, the fold split guard, the docs.
    2. History fold, published with the lease after prove.
    3. Tip batches, in parallel by disjoint files.
    4. Baseline removal, once the scan reads 0.
  3. 269 remaining items

  4. added 15 commits that reference this issue on Oct 23, 2026
    0a2ecb8
    93ea470
    671021b
    3054f56
    6fcc3b9
    eb85c53
    e5e2f8a
    0778722
    0425d3d
    cd96be8
    bb8c9a3
    0f1699c
    93a9f39
    1d7f5a9
    0c2371c
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

dxdeveloper experience, CLIs, skills

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions