Repository navigation
Document tsbindgen in the README - #397
Merged
Merged
Conversation
How to run it, and why it can emit nothing: included declarations need --filter, static functions and non-literal macros are skipped, `()` has no prototype before C23, and a C++ header named .h needs `-- -x c++` (only structs, enums and extern "C" functions are kept). Adds tsbindgen to the VS Code workspace. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ASDAlexander77
added a commit
that referenced
this pull request
Sep 28, 2026
Every positive and negative program was dumped under -mm=rc on #397 and the plan now states what each produces: make() returns +1 with no retain on either side; a throw out of a try-less function leaks its local in rc too; the alias test needs `let` (a folded const emits no retain); the unknown producer is a declared function (.slice needs the default library). Also the verified call-op names, the --di-only variable name, number-typed accumulators, and the isHandOver move spelled out. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ASDAlexander77
added a commit
that referenced
this pull request
Sep 29, 2026
…ing (#399) * Design spec for -mm=own, a single-owner memory model A fourth memory model beside gc, rc and none: ownership inferred at compile time over the ownership ops MLIRGen already emits, moves and borrows instead of counts, and a compile error where neither can be proven. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * Satisfy markdownlint in the own-model spec Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * Written review of the -mm=own spec against main at #397 Six amendments in a new section 10: a lowering error does not fail the compile today (prerequisite for every model); the phase-0 gate is a pass, not the lowering; the default library is refused under own; the header word keeps its store on native targets; expected-error tests reuse the call-arity ctest pattern; and almost every retain is a birth reference, so a gate rejecting every retain compiles only heap-free programs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Plan for -mm=own phase 0, with the birth take Phase 0 accepts a fresh value's single acquisition - rc's count provably never above one - as a move, and reports every other retain. Seven tasks: the lowering-error prerequisite (PR #398), the model and flag, the isRefCounted() audit site by site, destroy-shaped lowering with a retain backstop, the inference pass, positive/negative/no-counting tests, and a rebuilt measuring script. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Check every phase-0 test program against rc affine IR Every positive and negative program was dumped under -mm=rc on #397 and the plan now states what each produces: make() returns +1 with no retain on either side; a throw out of a try-less function leaks its local in rc too; the alias test needs `let` (a folded const emits no retain); the unknown producer is a declared function (.slice needs the default library). Also the verified call-op names, the --di-only variable name, number-typed accumulators, and the isHandOver move spelled out. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * The -mm=own memory model exists: flag, name, tracksOwnership() Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Audit isRefCounted(): what own shares with rc is tracksOwnership() Sixteen of the 21 sites write something a release will read, or emit the ownership ops the inference pass judges, and become tracksOwnership(). Three retain lowerings and the captured-parameter retain are Task 4's; a cell's birth retain stays rc only. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Lowering under own: releases destroy without counting; a surviving retain is an error Under -mm=own emitIfLastReference tests the header against HEAP_BLOCK_IMMORTAL instead of decrementing it, so a release is one load and a compare. A retain that reaches LLVM lowering is an error, never erased; a captured variable holding a heap value is an error (closures are phase 5); and a type descriptor carries no retain routine, since retainViaDescriptor is reached only from retain routines. emitIncRef's other callers (emitRetainCell, buildCaptureBoxRoutine(retaining), buildRetainBody) are all retain paths. --own-skip-inference (hidden) lets the backstop be tested. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * OwnershipInferencePass, phase 0: a fresh value's single acquisition is a move; every other retain is an error Under -mm=own, at the affine level before the verifier: a heap value made by an allocating op (or marked __owned_result) whose births plus retains total one has exactly one owner, so its retains are erased and the release rc already emits becomes the owner's destroy. Any other retain, a RetainCell, a RetainSlot on a hoisted (initializer-less) variable, and a slot released without ever being owned are reported. isHandOver moves to OwnershipOps.h, shared with the ownership verifier. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * The own backstop test writes its .ll to the build tree, not beside the source Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * -mm=own phase-0 tests: positive, negative, and no counting in the output Six programs run under JIT and ahead of time with -mm=own, each checked to emit neither __tslang_inc_ref nor __tslang_dec_ref; five are rejected with the expected error. A literal acquired once is now a birth take too: `let s = "abc"` and `v: number[] = []` were rejected, and a cast constant is either the immortal global (its release skips it) or a copy one owner holds. The "released but never owned" rule is removed: the ownership verifier's stricter form of it is green over the corpus, so no program reaches it, and isHandOver goes back into the verifier. Teeth: with every retain kept, all 15 positive checks fail at the backstop. With the count check loosened to two, own_err_two_owners.ts compiles and segfaults under own (rc runs it: 500000, done.) - the count is what stands between a second owner and a double free. Debug tree: every own test passes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * measure.ps1, and the phase-0 numbers for -mm=own The AOT measuring script, rebuilt: it links the way the test runner does (read from the generated compile.bat) and reads the peak working set from the process handle after exit - sampling while it runs read 0.0 MB for programs that finish in milliseconds. Spec section 11 records the results: own holds flat exactly where rc does (4.2 MB for 1M strings, none 80.9 MB), the teeth, and phase 0's known limits. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * The lowering-error test uses an input that reports its error in a Debug build too The array-to-string cast hits llvm_unreachable (CastLogicHelper.h:856) in a Debug build before it reports anything, so the message test failed there. An array literal lowering cannot build as number[] | null reports its error the same way in Release and Debug. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Implement phase 0 ownership rules and error handling for -mm=own; add tests for various ownership violations --------- Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds a C bindings with tsbindgen section to the README:
simple.cexample and its output, and how to include the result (/// <reference path>, named.tsnot.d.ts)#included headers need--filter;staticfunctions and non-literal macros are skippedextern int f();is "function without a prototype" before C23 — write(void)or pass-- -std=c23.hneeds-- -x c++; only plain structs, enums andextern "C"functions are emitted, classes / namespace members / C++-linkage functions are dropped without a warningEvery output shown was produced by the current build.
Also adds
tslang/tsbindgentotslang.code-workspace.🤖 Generated with Claude Code