fix(context): single-source modelwindow table + add claude-sonnet-5 (gc-qin3c)#100
Open
zook-bot wants to merge 8 commits into
Open
fix(context): single-source modelwindow table + add claude-sonnet-5 (gc-qin3c)#100zook-bot wants to merge 8 commits into
zook-bot wants to merge 8 commits into
Conversation
…gc-yey06) (#88) Only unassigned routed work is pool demand, so a bare 'gc sling <pool> <bead>' of an existing open bead must clear a stale assignee or the re-pour is invisible to scale_check and stalls silently. In-progress beads are exempt. The order-sweep rig-dir stat gate that originally rode along in this commit was dropped: all configured rigs exist on disk, so it guarded a condition that does not occur, at the cost of threading an io.Writer through 3 call sites.
…olated git config (gc-v2z1p) (#89) `make check` runs `go test -p=4 ./...`; the parallel link phase of several large CGO/ICU test binaries peaks past the 16G tmpfs-backed /tmp and fails mid-link with "No space left on device". This surfaces as spurious rebase/refinery preflight failures (first hit hard by the gc-k0hvd upstream-rebase polecat, which worked around it by pointing TMPDIR at ext4). The flake is latent and recurs under load whenever /tmp is tight. Root cause: the TEST_ENV `env -i` wrapper passed TMPDIR through with a /tmp default, and GOTMPDIR is empty, so the Go toolchain's build/link work dir (go-build*/go-link*) and test-runtime temp both land on the tmpfs. Fix: default the wrapper's TMPDIR to /var/tmp (a persistent, non-tmpfs disk with more room; ext4 with 66G free on the Gas City host) instead of /tmp. Every `make test*` target flows through TEST_ENV, so the shard/parallel helper scripts it invokes inherit the same TMPDIR — one leverage point covers all preflight paths. An explicit TMPDIR export still overrides (e.g. `TMPDIR=/tmp make check`), so callers that want the old behavior keep it. Validation (fixed env, TMPDIR unset): - `go build -work` reports WORK=/var/tmp/go-build... (was /tmp) - a built binary sees os.TempDir()=/var/tmp - with TMPDIR=/tmp exported, WORK returns to /tmp/go-build... (override preserved) - `make -n test` expands the recipe with TMPDIR="${TMPDIR:-/var/tmp}" Environment/infra fix, not a code regression. No behavior change for tests themselves — they use whatever os.TempDir() returns. Second host-state leak, same wrapper: TEST_ENV allowlists HOME but not SSH_AUTH_SOCK, so a host ~/.gitconfig with commit.gpgsign=true plus gpg.format=ssh made every test that execs `git commit` fail with "Couldn't get agent socket?" / "failed to write commit object". That was being patched per-test by sprinkling testutil.IsolatedGitConfig(t) into upstream-owned test files — carry that grew with every new upstream test that execs git. Fix: point GIT_CONFIG_GLOBAL at a seeded config and GIT_CONFIG_SYSTEM at /dev/null in the same wrapper, so every test binary inherits the isolation and no per-test opt-in is needed. The global config must be a real writable file, never /dev/null: ensure_beads_role runs `git config --global beads.role maintainer` and cannot lock /dev/null. Tests that WRITE global config still call testutil.IsolatedGitConfig for a per-test file so writes don't leak through the shared one, and helpers that build an explicit env -i subprocess environment still append SharedIsolatedGitConfigEnv.
…#92) Gate the fork's zookanalytics/beads pseudo-version replace past the check-gomod-replace guard so Preflight/static-checks runs to completion and fork PRs can auto-land again (was FAILURE-at-guard on main). - gc-jwho8: allowlist the beads fork entry in scripts/check-gomod-replace.sh - gc-434qa (pre-open signoff): gate the allowlist on pseudo-version shape so the exception stays narrow, plus scripts/gomod_replace_guard_test.go coverage The FORK_REPLACE_ALLOWLIST entry is a TEMPORARY bridge: delete it when the beads fork converges upstream and go.mod's `replace` returns to a released semver tag. Closes gc-bvjbs.
These non-conflicted cmd/gc test files kept pre-rebase call shapes and only surfaced post-rebase (cmd/gc could not build mid-rebase until the resolveBdScopeTarget caller fix landed): - startControllerSocket gained demandDirty (gc-qedgc #69) — add missing nil arg - startCandidate's raw bead pointer became info session.Info (WI-6 R4): session:&bead -> info: SeedBead / InfoFromMeta - sortCandidatesByWakeFairness gained a cfg param
Rebase of origin/main onto upstream/main (adc03f1) grew the tracked-source AST counts (fork tests atop new upstream). Ratchet bootstrapPolicy + test-resources.toml baselines to the current census and regenerate the TESTING.md ledger block. Supersedes the pre-rebase census snapshot 7bba3de, which drops as empty against new upstream (the census is a pure function of the tree, plan §5e).
…ut arg (gc-u1g6k) Upstream adc03f1 added a checkTimeout time.Duration parameter to doDoctor (between the bool flags and the writers). One fork test call site was missed by the existing test-alignment commit; pass 0 to match every other call site.
…gc-qin3c) The fork core-path resolver landed via #85 (gc-os8fn) but trailed the reviewed upstream PR gastownhall#4527 (gc-z0vi2): the 1M marker list was missing claude-sonnet-5, so a bare "claude-sonnet-5" — a 1M-context model — resolved to the 200K default on the API/session-log path and the injector alike. That is the same under-report class #85 fixed for the bare opus-4-6/4-7/4-8 and sonnet-4-6 families, left open for the newer model. Bring the five fork-relevant files to match the reviewed PR commit (5a9daff): - add claude-sonnet-5 to the 1M markers - fold the familyWindows map + separate familyOrder slice into one ordered struct slice (longest-match-first inline); no behavior change for existing families - expand tests: sonnet-5 cases on both resolvers, and a table-driven injector test across the shared model table (opus-4-8 / sonnet-5 / gpt-5) Excludes the PR's upstream-CI-only files (.github/workflows/ci.yml, scripts/cipolicy/policy.go) — the fork carries its own CI shape and cipolicy execution digest, which must not be overwritten with upstream's.
johnzook
approved these changes
Jul 23, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Brings the fork's core-path context-window resolver up to the reviewed form of upstream PR gastownhall#4527 (
gc-z0vi2, head5a9daff84).The fork already carries the single-source
internal/modelwindowextraction from #85 (gc-os8fn), but that landed before the upstream review and trails it in two ways:claude-sonnet-5. A bareclaude-sonnet-5— a 1M-context model — resolves to the 200K default on both the API/session-log path (internal/sessionlog.ModelContextWindow) and the context-pressure injector (cmd/gc/context_inject.go). That is the same under-report class ModelContextWindow under-reports 1M window for bare claude-opus-4-8 (no [1m] suffix) — context_pct reads 100% at real ~38% (gc-os8fn) #85 fixed for the bareopus-4-6/4-7/4-8andsonnet-4-6families, left open for the newer model.familyWindowsmap plus a separatefamilyOrderslice; the reviewed PR folds them into one ordered struct slice (longest-match-first inline), and expands the tests.This PR checks out the five fork-relevant files from the reviewed PR commit
5a9daff84, making them byte-identical to it:internal/modelwindow/modelwindow.gointernal/modelwindow/modelwindow_test.gointernal/sessionlog/context.gointernal/sessionlog/context_test.gocmd/gc/context_inject_test.gocmd/gc/context_inject.goandinternal/modelwindow/testenv_import_test.gowere already identical on fork main and are untouched.Deliberately excluded
The upstream PR also touches two CI-shape-specific files that must not cross onto the fork base:
.github/workflows/ci.ymlscripts/cipolicy/policy.go(theexpectedCIExecutionHash/requiredFilterPaths)The fork carries its own CI workflow and cipolicy execution digest; overwriting them with upstream's values would break the fork's own policy check. The fork's cipolicy already covers
internal/modelwindow/**(landed with #85), so no policy change is needed here.Behavior
opus-4-6/4-7/4-8,sonnet-4-6,sonnet-5,fable,mythos→ 1M[1m]suffix still forces 1M for any Claude familyNo behavior change for any family already recognized on fork main; the only newly-resolved model is
claude-sonnet-5.Testing
go test ./internal/modelwindow/... ./internal/sessionlog/...→ okgo test ./cmd/gc/ -run TestContextInject→ all pass, including the newclaude-sonnet-5case inTestContextInjectResolvesWindowFromSharedModelTable5a9daff84, which passed full upstream CI on PR fix(context): single-source the model context-window table gastownhall/gascity#4527.Bead:
gc-qin3c(discovered-fromgc-z0vi2).