fix(php): preserve use-statement facts and stop group-form use function/const from claiming class names - #2502
Conversation
PHP `imports` edges now carry `use_kind` / `alias` / `target_fqn` metadata,
mirroring `_import_csharp`. The already-correct `use`-parser inside
`_resolve_php_type_references` was extracted into shared helpers
(`_php_use_clause_fact`, `_php_use_clause_context`,
`_php_use_declaration_facts`) consumed by both the resolution pass and the
capture path, replacing `_import_php`'s lossy `raw.split("\\")[-1]`. Group use
`use A\{B, C as X};`, aliases, leading-backslash absolutes and
`use function` / `use const` are all handled in one place, so a clause
dispatched on its own (as `_import_php` is) can still spell its own FQN by
reading the group prefix and keyword off the parent declaration.
Strictly metadata-only: no resolver behavior change, `_PHP_CONFIG.import_types`
untouched, edge targets still keyed on the imported short name. Full
`extract()` output with metadata stripped, before vs after, over a corpus
covering plain / aliased / group / aliased-group / `function` / `const` /
group-function / group-const / leading-backslash `use`, trait `use`,
inheritance, interfaces and a typed member call: 16 nodes / 27 edges,
byte-identical (sha256 f6c6168f).
Group-form `use function A\{f, g};` and `use const A\{K};` put the keyword on
the declaration rather than the clause, so those names enter the class-name map
today; that pre-existing bug is deliberately preserved bit-for-bit here via
`apply_declaration_kind=False` and fixed in the follow-up commit. The new
metadata already reports the correct kind.
Note for consumers: the `use_kind` vocabulary is `class`/`function`/`const`
with `alias` as a separate key (unlike C#'s `using_kind == "alias"`), and
`_resolve_php_type_references` re-points `imports` edges without touching
metadata, so `metadata.target_fqn` is the reliable read rather than the target
node's label.
Tests: 8 new, all through the public `extract()` seam; 7 failed against
unfixed code (the 8th is the targets-unchanged guard, green by construction).
Full suite 3984 passed / 36 skipped (baseline 3976/36 + 8).
Adapted from fork PR #29
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Group-form `use function A\{f, g};` and `use const A\{K};` put the keyword on
the *declaration* node, not the clause, so those names wrongly entered
`_resolve_php_type_references`'s class-name map. A group-imported function or
constant whose short name was also used in a class position in the same file
therefore re-pointed that `inherits`/`implements`/`mixes_in`/`imports`/
`references` edge onto an external stub labeled with an FQN that names a
function or a constant, not a class.
The shared parser added in the previous commit already computed the correct
kind behind an `apply_declaration_kind=False` compatibility flag, which existed
only to keep that commit metadata-only. This removes the flag and its call
site, leaving one code path that always honors the declaration-level keyword,
so both spellings agree. Strictly subtractive: it can only remove a class-name
claim, never add one. The reference then falls back to the namespace-relative
FQN or to the legacy unique-label rewire, exactly as the unbraced form always
did.
Pre-existing, and rare in practice because it needs the same short name used
both as a group-imported function/constant and in a class position within one
file.
Tests: 4 new, all through the public `extract()` seam, each braced form paired
with its semantically equivalent unbraced control; 3 failed against unfixed
code, and the over-subtraction guard (`use App\Cms\{Page};` still claims the
class name, decoy `App\Models\Page` gets no edge) passes on both sides by
design. Full suite 3988 passed / 36 skipped (3984/36 + 4).
`grep -rn apply_declaration_kind` across the repo now returns nothing.
Adapted from fork PR #30
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
![graphify-labs[bot]](https://avatars.githubusercontent.com/in/4356270?s=60&v=4){kind=small}
There was a problem hiding this comment.
Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).
Graphify reviewed this change.
Looks safe to merge — no coupling regressions and no blocking issues, checked against the code graph (not a self-assessment).
Graphify review — findings
This PR reworks how PHP use statements are parsed during graph extraction. It introduces a shared use-statement parser (_php_use_clause_fact, _php_use_clause_context, _php_use_declaration_facts) in resolution.py and updates both _import_php (the imports edge capture) and _resolve_php_type_references to use it, so group-use syntax, aliases, absolute names, and function/const keyword placement are handled consistently. It also adds use_kind, alias, and target_fqn metadata to PHP imports edges, and adds a new test file plus CHANGELOG entries. The surface area spans graphify/extract.py, graphify/extractors/resolution.py, a new tests/test_php_group_use_kind.py, and CHANGELOG updates.
No blocking issues surfaced. 3 lower-confidence candidates did not survive cross-model review.
Analysis details — impact, health, verification
Impact & health
Graphify review
Impact — 1673 functions depend on the 511 functions this change touches.
Health — this change adds coupling hotspots:
- worse:
extract()— 379 callers, 39 callees
Verification — 1673 functions in the blast radius were not formally verified this run (proofs are advisory here).
Gate & verification
graphify gate
PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.
Advisory (not blocking):
- verification_scope: 1543 function(s) in the blast radius were not formally verified this run
· 1 more finding(s) on lines outside this diff (see the check run).
1 participant
Two coupled PHP
use-statement fixes, backported from our fork where they landed with full red-first evidence (lawnstarter#29, lawnstarter#30). This PR is independent of #2492 (it touches only pre-existing code); a heads-up was posted there for context.Commit 1 — capture
useFQN/alias/kind metadata on imports edges (metadata-only)_import_phpthrew away everything but the trailing short name (raw.split("\\")[-1]), unlike_import_csharp, which preserves the written FQN and alias. The already-correctuse-parser inside_resolve_php_type_references(group use, aliases,function/constrejection) is extracted into shared helpers consumed by both paths, and PHPimportsedges now carryuse_kind/alias/target_fqnmetadata.No behavior change: metadata-stripped
extract()output over a corpus covering plain / aliased / group / aliased-group /function/const/ group-function / group-const / leading-backslashuse, traituse, inheritance, and interfaces is sha256-identical before and after.Commit 2 — group-form
use function/use constno longer claim class names (strictly subtractive)The grammar puts the
function/constkeyword on the clause foruse function A\f;but on the declaration foruse function A\{f, g};(verified on tree-sitter-php 0.24.1)._resolve_php_type_referencesonly ever honored the clause-level keyword, so group-form function/const names wrongly entered the class-name map — e.g.use function Vendor\Sdk\{render};plusclass I extends Rendermis-repoints the inheritance. Now the declaration-level keyword is always honored; one code path, no compatibility flag.Test evidence
v8@07b9143): 3976 passed / 36 skipped.use A\{B};still claims the class name, decoy gets no edge) → full suite 3988/36.extract()seam.🤖 Generated with Claude Code