[colbymchenry]

Summary

In a tag-based CFML file, a <cfquery> or <cfscript> block the parser can't read put this warning on the whole file: "parse produced no symbols (tree has errors) — the file is indexed but contributes nothing to the graph". codegraph init printed it and codegraph status counted the file among those "could not be parsed, so their symbols are missing", though the file's component, methods and calls were all in the graph. The reporter saw it on 66 DAO and gateway files.

The issue's own example SQL (SELECT <cfif arguments.fields IS "count">COUNT(*)<cfelse>p.*</cfif> FROM pages p) parses cleanly with the bundled query grammar, as do about 110 other query shapes tried, so the reporter's actual trigger was not reproduced. The fix doesn't depend on it: any <cfquery> body the query grammar can't read produced the warning (the repro uses EQV in a <cfif> condition, one shape it can't read), and so did a statement-only <cfscript> block with a syntax error.

Root cause

Both delegates in cfml-extractor.ts parse the block with its own grammar through a fresh TreeSitterExtractor and copied its errors onto the file. That extractor warns when its tree has errors and it produced no symbols, which describes the block, not the file, and a <cfquery> body never declares a symbol. extractTagExpressions already drops that warning for the same reason; the two delegates didn't.

Fix

The delegates record the block's warning as what it is: one naming the block and its line (<cfquery> at line 6 has a syntax error — the rest of the file is indexed, but calls inside that block may be missing from the graph), with line set and its own code, partial_parse. So codegraph files --json shows which block failed, while init no longer prints a whole-file warning and status no longer counts the file as one whose symbols are missing (both key on parse_error). Other errors pass through as before. A file whose whole source fails to parse still gets the old warning.

Validation

• New __tests__/cfml-embedded-parse-warning.test.ts: the repro's DAO (class, both methods and the #getSiteId()# call inside the failing query are kept; one partial_parse warning names <cfquery> at line 6), a failing <cfscript> block, and getIndexHealth() no longer listing either file. All 3 fail on main's extractor and pass with the fix. status-index-health.test.ts passes.
• On the repro, main prints ▲ appsettingsdao.cfc: parse produced no symbols … during init and status reports "1 file could not be parsed, so its symbols are missing"; with the fix, neither.
• A/B against main f406fe0 (CFML has no kernel path), each arm built from its own sources:
  • CFML: FW/1 d7fb9add, BlogCFC 8758879a, ColdBox 13480b7c, ContentBox 7070b2c4, Preside a3d9028b and CFWheels 2b1a8a62: no change. In CFWheels' vendor/wheels core indexed as its own project (the default ignore skips vendor/), one file changed: tests/html.cfm, a template whose <cfscript> block at line 8 fails to parse, now records <cfscript> at line 8 has a syntax error … instead of the whole-file warning. Nodes, edges and references are identical in every repo.
  • Controls: express 0dee2c52 and flask d086db85 on the kernel path, full graph dumps byte-identical.
• Index time: median of 4 interleaved runs, CPU with workers, main vs fix: Preside 57.1 / 54.5 s, Wheels core 33.2 / 32.8 s. Within noise; the change only rewrites an error record.
• Full suite on the pushed commit 9fc2bc6 (rebased onto ef05970, whose test(daemon): the OS process lookup tests pass on a loaded Windows machine #2842 only touches daemon tests and process lookup), kernel built (CODEGRAPH_KERNEL_EXPECT=1): 752 files (746 passed, 6 skipped), 8,986 tests: 8,904 passed, 0 failed, 82 skipped. The pre-rebase head 35e6edc on f406fe0 also passed with 0 failures.

Fixes #2654

Left out

Making the bundled query grammar read more SQL is upstream work in tree-sitter-cfml, and without one of the reporter's files the shape that fails for them is unknown.

🤖 Generated with Claude Code