[sunerpy]

Tier 6 of the post-v1.5.0 upstream sync — eight independent surface defects, batched because each is small and none shares a design with another. Every one was reproduced on a real index before the fix.

An explicit Python import was ignored (#1518)

Worse than upstream's framing, which blames line-wrapped parenthesized import lists — the single-line form fails identically, and produces a false edge plus a missing one from the same cause:

pkg/mod.py      def real_target(): return 1
pkg/decoy.py    def real_target(): return 99
app.py          from pkg.mod import real_target       → caller()
app2.py         from pkg.mod import real_target as aliased → caller2()

before:  caller  -> real_target@pkg/decoy.py      ← bound to the DECOY
         caller2 -> (no edge at all)
after:   caller  -> real_target@pkg/mod.py
         caller2 -> real_target@pkg/mod.py        edges to decoy: 0

Root cause is neither of upstream's: the import's dotted source was never mapped to a path, and the fallback gated on an is_exported flag our Python extractor never sets. Both halves failed, so the reference fell through to name matching. The fix is gated narrowly so reference/golden/python/'s function-ref edges are untouched.

Go calls across module boundaries (#1521)

Two modules, ex/b requiring ex/a: 0 calls edges before, Use -> Helper@a/lib.go after. A module was identified by path without knowing its directory; modules are now located by walking each file's ancestors for a go.mod, and the longest matching path wins so ex/abc is not mistaken for a child of ex/a. Pinned by a file-order-independence test and that segment negative.

Declaration initializers (#1510, upstream PR #1511)

export const value = target(); attributed the call to the file, so callers target answered with a filename:

before:  file:d.ts       -> function:target
after:   constant:value  -> function:target

Controls pin that function-local initializers stay attributed to the function, and that a multi-declarator statement attributes per declarator.

The rest

• #1524 — index . inside a subdirectory rebuilt the parent index silently, reporting files the user did not ask about. It now warns, on sync/uninit/unlock too.
• #1512 — callers/callees/impact accept --file. Matching is segment-aligned, so other.ts never selects my_other.ts, and an unmatched filter is an error listing the defining files rather than an empty result that would read as "this symbol is dead".
• #1504 — the explore note stated a call budget nothing enforced, which made agents stop early. It now says there is no limit; a test bans all four cap phrasings and asserts the replacement is present, so the wording cannot creep back.
• #671 — MCP instructions list the 38 languages we actually support, generated from the Language registry rather than copied prose. This is the one golden this change regenerates (mcp/initialize.json, one field appended), verified against the live server.
• fix(cli): enable self_update zip extraction for Windows releases #23 — installer gains VS Code, Copilot CLI and JetBrains targets. While testing I broke all three config contracts at once and found the --print-config assertion did not catch a broken install, so it now reads the written file instead.

Verification

• make ci green: 132 suites / 3304 passed / 0 failed (baseline 129 / 3231).
• CURRENT_EXTRACTION_VERSION still 7; no schema, NodeKind or EdgeKind change; crates/codegraph-resolve/src/frameworks/ untouched.
• Extraction goldens (all 15 corpora) byte-identical; reference/ movement is only mcp/initialize.json.
• reference/golden/erlang/edges.json still holds its 2 legitimate recursive calls self-edges.

Self-index delta — controlled, same source, old vs new binary

edge_from_file:calls   4 → 3
edge:calls         22715 → 22716

Exactly two edges, both intended. file:sample.astro → method:fetch became constant:posts@sample.astro → method:fetch — #1510 working. And a new edge RunCheckoutFlow@…/e2e/flow.go → Double@…/math.go: that fixture has a go.mod nested under crates/codegraph-cli/tests/fixtures/go_test_convention/ with none at the repo root, so this repository's own tree was a latent instance of #1521.

One finding worth separate attention

The codegraph-bench golden suite writes JSON from the committed colby.db and compares it to the committed JSON — so it guards database↔JSON consistency but never re-runs the extractor, and therefore cannot detect an extraction change on its own. I confirmed both halves: corrupting colby.db does fail the suite, and no test in that crate invokes the extractor. Real byte-stability requires regenerating from the corpora, which is what the tiers that changed extraction did. Regenerating also surfaces three diffs (godot/schema.sql, mini/refs.json, mini/schema.sql) that are pre-existing on clean main and unrelated to this change; mini/refs.json gaining a this.value function_ref row looks like content drift rather than ordering noise and deserves its own look.

[codecov]

Codecov Report

❌ Patch coverage is 98.69679% with 15 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
crates/codegraph-resolve/src/import_resolver.rs	92.30%	8 Missing ⚠️
...tes/codegraph-cli/src/installer/targets/copilot.rs	98.99%	6 Missing ⚠️
crates/codegraph-extract/src/walker.rs	98.64%	1 Missing ⚠️

@@            Coverage Diff             @@
##             main     #215      +/-   ##
==========================================
+ Coverage   95.25%   95.30%   +0.04%     
==========================================
  Files         137      139       +2     
  Lines       74928    76041    +1113     
==========================================
+ Hits        71375    72473    +1098     
- Misses       3553     3568      +15     

Files with missing lines	Coverage Δ
crates/codegraph-cli/src/installer/registry.rs	98.60% <100.00%> (+0.02%)	⬆️
crates/codegraph-cli/src/installer/types.rs	100.00% <100.00%> (ø)
crates/codegraph-cli/src/main.rs	90.85% <100.00%> (+0.29%)	⬆️
crates/codegraph-mcp/src/engine.rs	97.60% <100.00%> (+0.04%)	⬆️
crates/codegraph-mcp/src/instructions.rs	100.00% <100.00%> (ø)
crates/codegraph-mcp/src/rmcp_handler.rs	92.67% <100.00%> (ø)
crates/codegraph-mcp/src/server.rs	97.01% <100.00%> (ø)
crates/codegraph-resolve/src/context.rs	98.47% <100.00%> (+0.33%)	⬆️
crates/codegraph-resolve/src/snapshot_context.rs	97.24% <100.00%> (-0.80%)	⬇️
crates/codegraph-resolve/src/types.rs	100.00% <100.00%> (ø)
... and 3 more

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.