WIP: v0.19 advanced type system #1507

Generated from source commit 1352ed3958 (source changes verified before committing).

Validated sequential rebootstrap, boot-clean, stage0, stage1, stdlib and tools with eight build cores. Post-rebuild: 21 focused control/return-inference/nullable cases passed, plus the separately compiled nine-case control suite. Tests used memory/time limits.

accept default-before-contract keyword arguments f38c48bdb2

bootstrap alternate typed keyword defaults 15c8aa3fc4

Generated from source commit f38c48bdb2 (source changes verified before committing).

Validated sequential rebootstrap, boot-clean, stage0, stage1, stdlib and tools with eight build cores. Post-rebuild: 32 focused cases passed, plus the separately compiled eleven-case sugar/interface suites. Tests used memory/time limits.

propagate stable-binding facts through normal completion e9962b1a12

bootstrap stable-binding normal-completion flow 2f938b6a50

Generated from source revision e9962b1a12 using the accepted 15c8aa3f bootstrap. Validated sequential clean stage0/stage1/stdlib/tools rebuild, 36 focused cases, and 27 compiler regression cases.

vyzo added 100 commits

2026-09-30 21:37:09 +00:00

add experimental structured optimizer flow IR 9a88919cf0

bootstrap experimental optimizer flow IR 784f1f8721

Generated from source revision 9a88919cf0 using the accepted 2f938b6a bootstrap. Validated clean stage0/stage1/stdlib/tools rebuild and all 76 focused/regression cases, including native direct-versus-IR round-trip fixtures. The new pass remains opt-in.

type flow IR and add contextual compiler BUG diagnostics 5c134fa79e

bootstrap typed flow IR and compiler bug diagnostics 001cba7b5a

Generated from source revision 5c134fa79e using accepted bootstrap 784f1f87. Clean stage0/stage1/stdlib/tools validation and all 78 regression cases passed. The later test-only qualification cleanup passed the 15-case flow suite without rebuilding.

move stable-binding normal-flow analysis onto the IR 32408b9d49

bootstrap IR-backed stable-binding analysis 941a4ea8df

Generated from source revision 32408b9d49 using accepted bootstrap 001cba7b. Validated clean stage0/stage1/stdlib/tools rebuild and all 83 regression cases, including 5,184 differential AST/IR comparisons. Syntax rewriting remains opt-in; per-query analysis allocation remains a profiling follow-up.

profile IR analysis and test fresh abort-summary queries 3615037dd4

Reject the structural-cache prototype: pinned optimizer-only measurements show no useful speedup. Keep the accepted compiler and bootstrap unchanged, document reproducible profiling and the annotation-pass hotspot, and retain two freshness/revalidation regressions. All 85 focused cases passed after restoring the compiler.

normalize annotation block runtime names once per lookup scope 13e089fa2a

bootstrap annotation block lookup normalization 7776f39f63

Generated from source revision 13e089fa2a using accepted bootstrap 941a4ea8. Clean stage0/stage1/stdlib/tools rebuild and all 89 focused regression cases passed. Pinned optimizer-only measurements retained the approximately 28% improvement on optimize-ann.ss after the clean bootstrap.

reuse the last raw block substitution within annotation optimization 0e96d3b13a

bootstrap annotation raw substitution reuse 8312c66a94

Generated from source revision 0e96d3b13a using accepted bootstrap 7776f39f. Clean stage0/stage1/stdlib/tools rebuild and all 95 focused cases passed. Pinned post-bootstrap optimize-ann measurements show approximately 36% less optimizer time and 13% less allocation; gxpkg remains essentially unchanged. Only raw substitutions are reused, never branch-sensitive optimized bodies or inferred facts.

add opt-in whole-bundle value routes and guarded joins edaa24b2d8

bootstrap whole-bundle flow view 7ccc92b916

Generated from source revision edaa24b2d8 using accepted bootstrap 8312c66a. Sequential clean stage0/stage1/stdlib/tools gate and all 106 focused cases passed. The value-routing view and rewriting remain opt-in; production normal-path analysis is unchanged. Default optimizer probes remain comparable on optimize-ann and gxpkg. No module inventories changed.

plan isolated ir optimizer schema and migration gates 3482698c55

require core Gerbil corpus in IR migration verification 1797850735

specify IR naming and audit initial optimizer node schemas 1baefe2f7f

preserve unspecified operand scheduling freedom in IR design f8f9a632d4

audit IR call and module boundaries and sketch transform interfaces a5505b3619

Audit optimizer IR comparisons, metadata ownership and naming boundaries c32d243434

Specify IR struct families and post-transition optimization scope 4873db9101

Use defir prefix for IR definition macros 614ac94e00

Specify IR value patterns and audit backend arity boundaries 834185679c

Plan proof-driven values checking after IR cutover 8695d4331b

Specify canonical IR bodies and scope-aware child edges 0d24d57f37

Specify declaration and foreign IR payload boundaries fd13c29393

Specify signature codec ordering and metadata compatibility ef30323123

Specify stateful match projection and syntax quotation carriers 7aec350868

Propose concrete IR macro grammar and traversal contracts 20c83fd723

Record reviewed IR macro surface and concrete transform port contracts 174d874302

Preserve typed Gerbil probes and evidence with a portable runner 714160976e

Specify optimizer IR context ownership and publication boundaries d2d553cb75

Audit optimizer inline callbacks and preserve boundary probes 93e59c88fb

Record approved call-local inline migration contract ab82a915b5

Separate IR binding occurrences from view-dependent projections 8f35c8ee69

Specify IR construction and cloning ownership contracts 49ff865906

Audit IR framing and phase-aware schema interfaces db44e560cb

Inventory inline declarations for replacement review e8ae55f93f

Convert utility inlining and named-slot initialization dc865ec7fb

Regenerate bootstrap for utility inlining cleanup 02fc3df166

Source and validation: dc865ec7. Clean stage0/stage1 and dependent builds passed; post-bootstrap core/stdlib tests passed with the agreed static-link and ensemble.TODO exclusions.

Review optimizer IR source tracking and auxiliary codecs eb4d38457b

Replace generic framing with the approved source policy and enclosing-location fallback. Audit pattern, clause and binding-group codecs; preserve research probes and evidence for 444 passing source/generation cases.

Record post-cutover compiler-owned contracts and interface dispatch, compile-time rejection of proven violations, untyped bidirectional inference, and fixed-size heterogeneous vector types. No product or bootstrap changes.

Audit annotation producers and plan rich callable specialization e302c8ac01

Audit module and phase payload codecs 2d2887387e

Add the 29-case module payload probe, runner registration, codec notes, and archived evidence for the 15-probe / 505-case research suite. Preserve failed fixtures and provenance separately from the IR implementation.

Add shared optimizer IR conversion foundation 0867997acb

Implement typed immutable IR records, strict annotation codecs, lexical and phased lowering, shared module ownership, optional verification, and reification with checked publication. Preserve opaque user annotations and source identity. Register the modules after optimize-base in all build inventories.

Validation: incremental stage1 build and all 63 IR conversion cases pass. Conversion-only core and stdlib corpus acceptance remains pending.

Refine IR lambda nodes, contracts, and conversion codecs f445c95e59

Represent case-lambda arms as registered lambda nodes, tighten procedure typing, and preserve foreign syntax tails and inline-rule quote payloads. Extend focused regression coverage and record symbolic dispatch and ListTuple design decisions.

Add source-only IR corpus verification probes a8a23daa48

Compile core and stdlib modules in shared sessions and compare generated artifacts structurally, accounting for compiler timestamps and generated symbol identities. Record the completed corpus gate and subsequent focused verification.

Checkpoint shared optimizer IR infrastructure and CFG inference c33f8e1ddd

Add captured binding and scope infrastructure, compiler assertions, IR traversal and transforms, explicit descriptor/session state, and general expression inference backed by the canonical CFG.

Preserve legacy facades, record the remaining migration work, and align source and bootstrap module inventories in dependency order.

Validation: focused source suites passed 207 cases and 14302 checks in separate processes. Bootstrap and native rebuild remain deferred until after cutover.

Checkpoint Typed Gerbil optimizer before cutover 3c56884215

Capture the complete candidate IR optimizer, typed descriptor interfaces,
binding construction and relocation, metadata emission, and driver/build
integration. Include the isolated validation harness and final review fixes
for unchanged transforms, obsolete arena records, callable remapping, and
compilation failure handling.

Validated with 15 source suites (240 cases, 14,557 checks), a subsequent
22-case/162-check ownership gate, seven compiler-module emission checks,
and fresh-process execution of the emitted flow fixture. Preserve evidence
and record cutover as pending. Native/bootstrap validation and direct IR
backend emission remain subsequent milestones.

Bootstrap runtime helpers required by the IR compiler d04dc651d3

Regenerate bootstrap with the IR compiler 7776c73cc1

Generated with the recovered stage1 compiler. Validated with boot-clean followed by stage0, stage1, stdlib, libgerbil and tools builds.

Migrate compiler regression tests and document bootstrap staging ad78f3e15f

Use explicit IR environments and worker-session metadata in compiler tests. Count checked and unchecked class predicates without changing folding expectations. Record runtime bootstrap recovery and clean native validation, including the missing static-library limitation.

Emit Gambit code directly from IR and migrate optimizer tests 21dffc5650

Regenerate bootstrap with the direct IR Gambit backend 7959c7d31a

Remove legacy AST compiler and isolate shared metadata support 0ec779ec4e

Regenerate bootstrap without the legacy AST compiler 55f58b57db

Record clean bootstrap validation and post-cutover IR review 7aa87170e7

Simplify IR and stage repeatable call optimization bd7488e5a8

Separate semantic call optimization, runtime helper lowering, and final unchecked
call lowering. Default to two bounded call sweeps, refreshing analysis only as
needed and collecting inline declarations once on the final descriptors.

Join mutable binding types across initializers and assignments, retain common
procedure returns without preserving unsafe dispatch claims, and track refinement
changes explicitly. Cover phase-1 initializers and avoid lazy mutable-cell seeds.

Remove unused IR records and production AST snapshots, index inline rules by
canonical descriptors, and render diagnostic provenance through source syntax.
Add regressions and record sweep measurements and native validation evidence.

Regenerate bootstrap for the IR pass overhaul b729800d72

Generate the bootstrap from source commit bd7488e5, including the simplified IR,
mutation bounds, staged inline lowering and two-sweep call optimization default.
Keep generated artifacts separate from the source changes.

Validation: stage0 and stage1 self-compilation, stdlib, libgerbil and tools all
rebuilt successfully. Artifact audits passed and the native compiler reports the
two-sweep default. Active stdlib passes 766 cases / 72076 checks. Gerbil passes
442/444 cases and 11228/11230 checks; only the two previously recorded static-link
failures remain, due to absent crypto, ssl, z and sqlite3 archives.

Record IR overhaul bootstrap validation 1b6dce950b

Add named result type folding to IR inference 3978304416

Regenerate bootstrap with named type fold support e965ba6cce

Enable numeric builtin type folds and record validation 247eab34b1

Record generic type constructor and erasure design considerations 6c885bdc10

Specify builtin type rule prototype and prioritize union types 55e0b6dff0

Default folding rule keys to canonical procedure names 1a9983b269

Connect folding identity facts with capabilities and read reuse 9b0a2852d5

Unify finalized container immutability with init-only slot guarantees 9934f58174

Add union type foundations for the compiler DSL 432e207368

Normalize union alternatives and support subtype/disjointness proofs, compound
annotations and signatures, and SSXI projection and reload. Preserve distinct
callable summaries and keep calls through unions checked. Expose a precise join
separately from the existing nominal inference joins.

Validation: focused source suites pass 172 cases and 3214 checks. Core, stdlib,
libgerbil, and tools rebuilds pass. Full stdlib passes 766 cases and 72076 checks;
Gerbil passes 450/452 cases and 11404/11406 checks, with only the two recorded
missing-static-library failures. Include validation evidence and source hashes.

Add deferred compiler type references and class slot metadata cd8fbe7a59

Regenerate bootstrap with unions and deferred type metadata d5247ff9b2

Run semantic analysis and emit SSXI for every compilation 35630d48b4

Validate declarations, returns, and known calls without requiring optimization. Require versioned compiled interfaces, reuse completed source-dependency SSXI, and preserve public wrapper result bounds.

Track SSXI in incremental builds, remove obsolete generation state, and cover baseline compilation and metadata failures. Record clean bootstrap self-build, assertion timings, and full-suite results; only the two existing host static-link failures remain.

Regenerate bootstrap for baseline semantic compilation a497c4f966

Generated from source checkpoint 35630d48 with compiler assertions enabled. Fresh stage0 and clean stage1 self-compilation passed, followed by stdlib, libgerbil, tools, and full-suite validation.

All stdlib tests pass; Gerbil retains only the two known host static-link failures. Generation hashes and validation evidence are recorded in the source checkpoint.

Compile module sources in dependency-ordered processes 46cfb9aaf6

Run isolated source workers and release dependents only after every phased
output and semantic interface is complete. Reuse compiler-driver native jobs,
prefer newer Scheme source over stale native objects, and retain pending
markers and diagnostics for failed builds.

Join workers cooperatively, return native slots on failure, honor explicit
native concurrency, and preserve companion-file build options. Preserve the
originating syntax context for inherited private slot types.

Validation: 773 stdlib cases and 72,135 checks passed. Gerbil retained only its
two known missing-static-archive failures. Focused tests passed with one and
eight cores; real artifact deletion rebuilt exactly six modules and the next
build was a no-op. Process-based ordinary and optimized executables passed.

Regenerate bootstrap for process-isolated compilation f14a7418f8

Record bootstrap validation and contract entry-fact examples bb390cf61a

Track implicit preludes in build ordering and freshness bd3593891a

Collect builtin contract macros in the core prelude a9e97c02ae

Add single-container runtime contract traversals 8f1165cbb3

Regenerate bootstrap for builtin contract traversals 4b7dc67d5b

Load plain semantic interfaces without expansion 354e4d57d5

Compile authored builtin metadata once and emit versioned .ssxi data for all modules. Decode declarations directly into isolated compiler sessions, preserve primitive signatures and literal rules, and migrate build outputs and incremental checks.

Validate with a fresh bootstrap, dependent builds, full suites, and artifact-removal rebuilds. Record pinned probes showing 40-44x faster metadata loading; retain the two known host static-link test failures in the evidence.

Regenerate bootstrap with plain SSXI interfaces 179b6c298c

Include the compiled builtin artifact and plain module metadata. Stage0, stage1/BACH, stdlib, libgerbil and tools build without legacy metadata loading; validate Gerbil and stdlib suites and incremental recovery.

Skip IR compaction for allocation-free unchanged transforms dd38fb4cab

Regenerate bootstrap with guarded IR compaction e9a724703e

Review expander allocation and repeated resolution costs 88e38cb817

compiler: add versioned matcher region optimizers 8856c1b898

bootstrap: include versioned matcher compiler support 45a886bc20

docs: record matcher bootstrap validation and stage1 benchmark plan 5e6d66caeb

Emit versioned matcher regions and preserve shared decomposition 6b757aaa9c

Regenerate bootstrap for versioned matcher generators e66e13d274

Preserve procedure effects and distinguish idempotent allocation 64f7ce291c

Regenerate bootstrap for procedure effect annotations 60bb01c5b3

Restore fixed-arity syntax-case splice lowering cb797370c2

Remove legacy matcher recognizers and continuation normalization 504253bb4c

Validate matcher bootstrap and checked procedure effect boundaries 792d44978f

Regenerate bootstrap after matcher optimizer cleanup 9c37d8b1cc

Port gxprof to v0.19 and enable it in the tools build a768e4f06c

Defer gxprof backtrace processing and add an eager collection option 7f57d36eb4

Record fresh compiler and cumulative expander profiles e72197a92e

Prevent reentrant eager profiler sampling fc2324ac06

Collect procedure names with a weak creator cache, guard nested sampling, and capture before installing cleanup frames. Verify eager and deferred sampling, analyzer compatibility, and cleanup after exceptions.

Optimize typed hash lookups and IR ID registries f4609b9f61

Use the base prelude for compiler and expander consumers, inline small hash wrappers, and preserve intermediate slot types so chained interface calls eliminate redundant casts. Export the implicit application hooks from the protobuf prelude.

Use immediate tables for generated IR IDs, guard exhaustion and malformed IDs, and record source/native performance baselines. Bootstrap source compilation improves about 2.3% in pinned controls. All 774 active stdlib cases pass; Gerbil has only its two existing missing-static-library failures. Generated bootstrap follows separately.

Refresh bootstrap after typed lookup optimizations 55572fdd46

Regenerate from f4609b9f, compile stage0, and use it to self-compile all 70 core modules to Scheme and semantic interfaces. Post-bootstrap validation passes all 774 active stdlib cases; Gerbil retains only its two existing missing-static-library failures.

vyzo added 10 commits

2026-10-01 16:45:25 +00:00

Declare load-scheme before loading bootstrap runtime 4ef4f94fb9

Split IR declarations into nested compilation units cb7ec4fbdd

Refresh bootstrap after splitting IR declaration modules 35e2f71edd

Prioritize native compilation by generated input size 29c7da0115

Refresh bootstrap with prioritized native compilation dcb76c1223

Reduce repeated work in IR type analysis f4bf3fe154

Refresh bootstrap for reduced IR analysis work 614bad20bd

Record context lookup cache distribution and visibility contract 0e1221215f

Cache inherited expander bindings with raw symbol tables 2e21cee634

Refresh bootstrap for cached expander resolution 21b5d3e245

This pull request is marked as a work in progress.

This branch is out-of-date with the base branch

View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.

git fetch -u origin v0.19-typed-gerbil:v0.19-typed-gerbil

git switch v0.19-typed-gerbil

Merge

Merge the changes and update on Forgejo.

Warning: The "Autodetect manual merge" setting is not enabled for this repository, you will have to mark this pull request as manually merged afterwards.

git switch v0.19-staging

git merge --no-ff v0.19-typed-gerbil

git switch v0.19-typed-gerbil

git rebase v0.19-staging

git switch v0.19-staging

git merge --ff-only v0.19-typed-gerbil

git switch v0.19-typed-gerbil

git rebase v0.19-staging

git switch v0.19-staging

git merge --no-ff v0.19-typed-gerbil

git switch v0.19-staging

git merge --squash v0.19-typed-gerbil