shard sema and codegen

[alii]

[alii]

@coderabbitai off

[coderabbitai]

Caution

Review failed

Pull request was closed or merged during review

Walkthrough

This PR introduces LLVM backend partitioning for parallel codegen through a new LlvmPartitionSet container, adds comprehensive parallel semantic analysis infrastructure with synchronization primitives and work queue management, extends build/linker support for multi-partition LLVM object handling, improves type resolution under concurrent execution with WIP types and namespace completion, and adds performance diagnostics.

Changes

Cohort / File(s)	Summary
Build System & Configuration build.zig, src/main.zig	Added llvm-has-polly build option for Polly library linking, --llvm-shard-stats CLI flag for per-shard codegen diagnostics, increased Zig compiler memory limit, and instrumented startup with phase-timing markers.
Compilation & Work Management src/Compilation.zig	Added llvm_shard_stats option, mutex-protected work queues, switched to LlvmPartitionSet for LLVM emission, parallelized .analyze_func jobs under ZIG_PARALLEL_SEMA, and introduced per-shard statistics dumping via dumpLlvmShardStats.
Intern Pool & Type Metadata src/InternPool.zig	Introduced seqlock-style writing flag for Nav.Repr.Bits, added tid/index repacking helpers, extended concurrency helpers for field/type mutations, added function-analysis queue state (is_queued), introduced namespace publish/spin mechanism with sentinels, and refactored getOrPutKey to prevent ABBA deadlocks via lockShardsSorted.
Type System & Resolution src/Type.zig, src/Air/types_resolved.zig	Added eagerResolved predicate, hardened parallel sema with childComptimeOnly and .parallel_sema handling, added namespace completion waits (awaitNamespaceTypeFinished), improved resolution gating with pre-checks, and introduced resolveTypesFully to force resolution during parallel analysis.
Semantic Analysis Core src/Sema.zig, src/Zcu.zig, src/Zcu/PerThread.zig	Added comprehensive parallel sema synchronization: recursive sema_lock, per-unit claim/wait mechanism (claimOrWait/releaseClaim), deferred reference buffering, additional mutexes (compile log, cimport errors), updated analysis failure tracking, improved namespace scanning with atomic generation marking, and replaced analysis_in_progress with aipPut/aipRemove helpers.
LLVM Backend Partitioning src/codegen/llvm.zig, src/zig_llvm.cpp	Introduced PartitionSet container for sharded LLVM emission with per-shard locking and parallel emit via worker threads, extended Object with partition_id and partition_set, added sharded naming helpers (shardedNavName), refactored debug type lowering for unresolved types using eagerResolved checks, and extracted optimization pipeline into runOptimizationPipeline helper for per-partition execution.
Linker Object Handling src/link.zig, src/link/Lld.zig, src/link/MachO.zig, src/link/MachO/relocatable.zig	Added resolveZcuObjectPaths to compute partitioned object file paths based on llvm_codegen_threads, updated Lld to initialize zcu_object_partition_count and iterate over object paths, added appendZcuObjectInputs helper to MachO, and updated linker APIs to accept arena allocator for path construction.
MachO Symbol & Export Handling src/link/MachO/Object.zig, src/link/MachO/Symbol.zig, src/link/MachO/file.zig	Modified symbol visibility handling to preserve .hidden visibility in tentative definition conversion (emitting N_PEXT instead of unconditional .global), updated setOutputSym to set N_PEXT for hidden symbols, and changed export marking to include non-.local symbols (not just .global).
Target & Repository src/target.zig, src/Air.zig, .gitignore	Changed .stage2_llvm backend to support .separate_thread feature (returns true instead of false), re-exported resolveTypesFully from Air, and added .claude/ and bun-cache/ to ignore list.

🚥 Pre-merge checks | ✅ 2

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The pull request title 'shard sema and codegen' directly and concisely describes the main objective of the changeset, which implements sharded semantic analysis and code generation infrastructure.
Description check	✅ Passed	While the PR description consists only of an image reference (a Sonic 'Gotta go FAST' illustration), it is thematically related to the performance optimization goals of the changeset and conveys the intent of the work.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}