Extract typed C facts for search and lineage #24

JayKobe6k · 2026-09-03T22:41:00Z

JayKobe6k commented

2026-09-03 22:41:00 +00:00

Summary

expand the language-neutral IR with typed scopes, complete declaration types, aggregate and enum structure, signatures, operations, definition sites, unsupported facts, and validated CFG relationships
recursively lower owned libclang facts for initializers, assignments, expressions, direct and indirect calls, returns, branches, loops, and conservative unsupported constructs
preserve spelling and expansion provenance, deterministic identities, C17/C23 prototype semantics, and producer-before-consumer operation ordering
add comprehensive C fixtures, negative builder validation tests, deterministic repeated-indexing coverage, and supported-fact policy documentation

Tests

nix develop --command cargo fmt --all -- --check
nix develop --command cargo lint
nix develop --command cargo test --all-targets --all-features
nix develop --command cargo doc --no-deps --all-features
nix develop --command cargo check --target x86_64-pc-windows-gnu --all-targets --all-features
nix develop --command nixfmt --check flake.nix
nix flake check

Closes #7

## Summary - expand the language-neutral IR with typed scopes, complete declaration types, aggregate and enum structure, signatures, operations, definition sites, unsupported facts, and validated CFG relationships - recursively lower owned libclang facts for initializers, assignments, expressions, direct and indirect calls, returns, branches, loops, and conservative unsupported constructs - preserve spelling and expansion provenance, deterministic identities, C17/C23 prototype semantics, and producer-before-consumer operation ordering - add comprehensive C fixtures, negative builder validation tests, deterministic repeated-indexing coverage, and supported-fact policy documentation ## Tests - `nix develop --command cargo fmt --all -- --check` - `nix develop --command cargo lint` - `nix develop --command cargo test --all-targets --all-features` - `nix develop --command cargo doc --no-deps --all-features` - `nix develop --command cargo check --target x86_64-pc-windows-gnu --all-targets --all-features` - `nix develop --command nixfmt --check flake.nix` - `nix flake check` Closes #7

JayKobe6k added 1 commit

2026-09-03 22:41:01 +00:00

Extract typed C facts for lineage

Validate / Formatting, linting, and tests (push) Successful in 4m8s

Details

Validate / Formatting, linting, and tests (pull_request) Successful in 3m58s

Details

c17da99b23

JayKobe6k reviewed

2026-09-03 23:15:22 +00:00

JayKobe6k left a comment

Found two actionable semantic extraction issues.

src/parser/libclang.rs Outdated

					
				@ -377,0 +612,4 @@

				        } else if kind == EntityKind::VarDecl {

				            let block = self.ensure_current_block(callable, current)?;

				            let mut context = OperationContext::callable(callable, block);

				            self.lower_initialization(entity, &mut context)?;

[P1] Block-scope static declarations are emitted into the function entry block like automatic initializers, but C initializes these objects once before program start rather than on each function entry. Downstream reaching-definition and lineage analysis will therefore model a fresh assignment on every call. Please represent static-storage initialization outside the per-invocation CFG, or mark it with distinct initialization semantics.

[P1] Block-scope `static` declarations are emitted into the function entry block like automatic initializers, but C initializes these objects once before program start rather than on each function entry. Downstream reaching-definition and lineage analysis will therefore model a fresh assignment on every call. Please represent static-storage initialization outside the per-invocation CFG, or mark it with distinct initialization semantics.

src/parser/libclang.rs

					
				@ -377,0 +731,4 @@

				        let Some(declaration) = self.ensure_declaration(entity)? else {

				            return Ok(());

				        };

				        let Some(initializer) = variable_initializer(entity) else {

[P1] A variable declaration is lowered only through its initializer, so expressions in variably modified array bounds are skipped. For int a[bound()], this loses both the call to bound and any identifier uses in the bound, making the call graph and lineage facts incomplete for valid C. Please lower all evaluated declaration expressions, including VLA bounds, while still distinguishing the initializer value.

[P1] A variable declaration is lowered only through its initializer, so expressions in variably modified array bounds are skipped. For `int a[bound()]`, this loses both the call to `bound` and any identifier uses in the bound, making the call graph and lineage facts incomplete for valid C. Please lower all evaluated declaration expressions, including VLA bounds, while still distinguishing the initializer value.

JayKobe6k added 2 commits

2026-09-04 00:47:35 +00:00

Merge branch 'issue-6-project-sources' into issue-7-c-facts 8cb9f3cfc7

Fix declaration expression semantics

Validate / Formatting, linting, and tests (push) Successful in 2m26s

Details

Validate / Formatting, linting, and tests (pull_request) Successful in 2m7s

Details

962b47bd94

JayKobe6k commented

2026-09-04 00:48:08 +00:00

Review comment #931: Fixed in 962b47b. Block-scope static initializers are now translation-unit-owned static-storage operations and skipped by callable CFG lowering; IR validation accepts static-local targets while retaining automatic-local rejection. Focused IR/C-facts tests, full tests, rustdoc, and Windows GNU checks passed.

Review comment #931: Fixed in 962b47b. Block-scope static initializers are now translation-unit-owned static-storage operations and skipped by callable CFG lowering; IR validation accepts static-local targets while retaining automatic-local rejection. Focused IR/C-facts tests, full tests, rustdoc, and Windows GNU checks passed.

JayKobe6k commented

2026-09-04 00:48:08 +00:00

Review comment #929: Fixed in 962b47b. Evaluated VLA bounds are lowered before initializers for locals, block typedefs, pointer/multidimensional VLAs, and definition parameters, while prototypes and constant arrays remain non-executable. The C-facts regression and complete 101-unit/5-integration suite passed.

Review comment #929: Fixed in 962b47b. Evaluated VLA bounds are lowered before initializers for locals, block typedefs, pointer/multidimensional VLAs, and definition parameters, while prototypes and constant arrays remain non-executable. The C-facts regression and complete 101-unit/5-integration suite passed.

JayKobe6k changed target branch from issue-6-project-sources to main

2026-09-04 00:50:17 +00:00

JayKobe6k merged commit d551ac1a23 into main

2026-09-04 00:50:17 +00:00

JayKobe6k referenced this pull request from a commit

2026-09-04 00:50:19 +00:00

Merge pull request 'Extract typed C facts for search and lineage' (#24) from issue-7-c-facts into main

Sign in to join this conversation.

No reviewers