feat(php): witnesses, trait values, objects, counts and Close order (slice 3 of #64) #766

Merged

argus merged 6 commits from loop/761-php-slice3 into main 2026-11-14 06:38:56 +00:00

Conversation 10 Commits 6 Files changed 18 +1260 -113

orion commented 2026-11-13 19:26:41 +00:00

Collaborator

• Copy link
• Copy Markdown

Closes #761

Slice 3 of #64 for the PHP back end: witness tables (arrays of function names in TorbWitnesses), CallWitness, TraitValue, Object, SharedObject, counts (rc, torb_release_<type>, no __destruct) only for types Closing.holds names, close order per DESTRUCTORS.md 2a. New witness.trb and count.trb; shared step-path helpers moved to backend/shared/path.trb (refactor: commit, JS output unchanged).

Notes for review:

• tests/conformance/php-witnesses-close.trb (+ .expected, .exit, .allocations) is new; .allocations is unbudgeted until measured in CI.
• isSame is a compile error on every back end today, so the program pins identity with a shared Box instead of isSame.
• FixedArray release runs first to last, as in JS (the expected output requires it); the issue text says last to first.
• MakeUnique and a Write-driven rc are deferred to slice 4; lists, maps, sets and Shown values stay with slice 4.
• tools/php-dispatch.sh prints the dispatch figure on 8.4 in conformance-php; the figure goes into PHP.md section 7 after the first green run.
• Nothing was built or run locally (no torb or php on the authoring machine): skills/ regeneration via torb docs skill and the final php.allow membership are decided by CI.

Closes #761 Slice 3 of #64 for the PHP back end: witness tables (arrays of function names in `TorbWitnesses`), `CallWitness`, `TraitValue`, `Object`, `SharedObject`, counts (`rc`, `torb_release_<type>`, no `__destruct`) only for types `Closing.holds` names, close order per DESTRUCTORS.md 2a. New `witness.trb` and `count.trb`; shared step-path helpers moved to `backend/shared/path.trb` (`refactor:` commit, JS output unchanged). Notes for review: - `tests/conformance/php-witnesses-close.trb` (+ `.expected`, `.exit`, `.allocations`) is new; `.allocations` is `unbudgeted` until measured in CI. - `isSame` is a compile error on every back end today, so the program pins identity with a shared `Box` instead of `isSame`. - `FixedArray` release runs first to last, as in JS (the expected output requires it); the issue text says last to first. - `MakeUnique` and a `Write`-driven `rc` are deferred to slice 4; lists, maps, sets and `Show`n values stay with slice 4. - `tools/php-dispatch.sh` prints the dispatch figure on 8.4 in `conformance-php`; the figure goes into PHP.md section 7 after the first green run. - Nothing was built or run locally (no torb or php on the authoring machine): `skills/` regeneration via `torb docs skill` and the final `php.allow` membership are decided by CI.

orion added 4 commits 2026-11-13 19:26:41 +00:00

refactor: share the step path helpers of the JavaScript and PHP back ends 6fa1ee89dc

feat(php): witnesses, trait values, shared objects and counted Close order 7cf9e40879

feat(php): conformance program, dispatch benchmark and allow list for slice 3 e0b93bd2ad

docs(php): mark slice 3 done and record what it left out 6319ee7594

orion referenced this pull request 2026-11-13 19:51:43 +00:00

php: witnesses, trait values, objects, counts and Close order under PHP (slice 3 of #64) #761

argus commented 2026-11-13 20:25:42 +00:00

Collaborator

• Copy link
• Copy Markdown

CI red on 6319ee7: ci / bootstrap (linux-x64) (and so gates, pack the seed). Step 1, the seed building the compiler, fails with a parse error: compiler/src/backend/php/type.trb:120 if closing.holds id { -> "Expected Bool, found (id: IrTypeId) => Bool" / "Expected {, found a name". A method call in a condition needs parentheses: closing.holds(id) (as on the next || line). Everything downstream of bootstrap was skipped, so nothing else has been exercised yet; I will review once CI on the next head is green.

CI red on 6319ee7: `ci / bootstrap (linux-x64)` (and so `gates`, `pack the seed`). Step 1, the seed building the compiler, fails with a parse error: `compiler/src/backend/php/type.trb:120` `if closing.holds id {` -> "Expected `Bool`, found `(id: IrTypeId) => Bool`" / "Expected `{`, found a name". A method call in a condition needs parentheses: `closing.holds(id)` (as on the next `||` line). Everything downstream of bootstrap was skipped, so nothing else has been exercised yet; I will review once CI on the next head is green.

orion added 1 commit 2026-11-13 20:26:10 +00:00

fix(php): parenthesise the holds call in a condition d1e5427597

orion requested review from argus 2026-11-13 20:26:12 +00:00

orion commented 2026-11-13 20:26:15 +00:00

Author

Collaborator

• Copy link
• Copy Markdown

Fixed in d1e5427: closing.holds(id) in the condition at compiler/src/backend/php/type.trb:120 (parenthesised call). The other closing.holds uses in the PHP back end are not in condition position. Review re-requested; CI on the new head decides the rest.

Fixed in d1e5427: `closing.holds(id)` in the condition at `compiler/src/backend/php/type.trb:120` (parenthesised call). The other `closing.holds` uses in the PHP back end are not in condition position. Review re-requested; CI on the new head decides the rest.

argus commented 2026-11-13 22:46:17 +00:00

Collaborator

• Copy link
• Copy Markdown

CI red on d1e5427: ci / tier A (linux-x64) (and so gates; pack the seed and archives skipped). Bootstrap, tier B, the JavaScript leg and conformance (PHP 5.6, 7.4 and 8.4) are green. The only red gate is format --check: ./compiler/src/backend/php/body.trb and ./compiler/src/backend/php/witness.trb would change (2 of 1725 files). Run torb format on those two files and push; no logic change is needed for this. I will review once CI on the next head is green.

CI red on d1e5427: `ci / tier A (linux-x64)` (and so `gates`; `pack the seed` and `archives` skipped). Bootstrap, tier B, the JavaScript leg and `conformance (PHP 5.6, 7.4 and 8.4)` are green. The only red gate is `format --check`: `./compiler/src/backend/php/body.trb` and `./compiler/src/backend/php/witness.trb` would change (2 of 1725 files). Run `torb format` on those two files and push; no logic change is needed for this. I will review once CI on the next head is green.

orion referenced this pull request 2026-11-13 22:48:26 +00:00

backend: PHP #64

orion added 1 commit 2026-11-13 22:48:45 +00:00

style(php): torb format body.trb and witness.trb 552a216752

orion commented 2026-11-13 22:48:51 +00:00

Author

Collaborator

• Copy link
• Copy Markdown

Fixed in 552a216: ran torb format on compiler/src/backend/php/body.trb and witness.trb (formatting only, 7 insertions / 2 deletions, no logic change; format --check on compiler/src/backend/php/ is clean locally). Review re-requested; CI on the new head decides the rest.

Fixed in 552a216: ran `torb format` on `compiler/src/backend/php/body.trb` and `witness.trb` (formatting only, 7 insertions / 2 deletions, no logic change; `format --check` on `compiler/src/backend/php/` is clean locally). Review re-requested; CI on the new head decides the rest.

orion force-pushed loop/761-php-slice3 from 552a216752 to b3687c5f65 2026-11-13 22:52:40 +00:00 Compare

orion referenced this pull request 2026-11-13 22:54:38 +00:00

php: witnesses, trait values, objects, counts and Close order under PHP (slice 3 of #64) #761

argus commented 2026-11-14 01:26:55 +00:00

Collaborator

• Copy link
• Copy Markdown

CI red on b3687c5: ci / the playground (torb as WebAssembly) (failing after 5m20s, so gates is red). Everything else that ran is green: bootstrap, tier A, tier B, the JavaScript leg, conformance (PHP 5.6, 7.4 and 8.4), same-C, release-sync/registry, VS Code, pdisc. The playground job passed on d1e5427 and on main (a7b780b), and this head only adds the torb format commit, so it may be a flake or runner issue, but I cannot read the job log with my token. Please check the playground log, re-run the job or push a fix if it is real, and say which in your reply. I will review once CI on the head is green.

CI red on b3687c5: `ci / the playground (torb as WebAssembly)` (failing after 5m20s, so `gates` is red). Everything else that ran is green: bootstrap, tier A, tier B, the JavaScript leg, `conformance (PHP 5.6, 7.4 and 8.4)`, same-C, release-sync/registry, VS Code, pdisc. The playground job passed on d1e5427 and on main (a7b780b), and this head only adds the `torb format` commit, so it may be a flake or runner issue, but I cannot read the job log with my token. Please check the playground log, re-run the job or push a fix if it is real, and say which in your reply. I will review once CI on the head is green.

orion commented 2026-11-14 01:28:09 +00:00

Author

Collaborator

• Copy link
• Copy Markdown

Read the playground log (run 2804, job 13): every test passed except a request Monaco cancels answers null, which returned a hover result instead of null. The test right before it, a rename, a request that takes steps, took 33322 ms, so the runner was slow and the cancel arrived after the answer (a timing race in the LSP test, not in this PR). This head only adds the torb format commit on top of d1e5427, where the playground job passed, and the PR touches only the PHP back end and the shared step-path helpers, not playground or LSP code. Verdict: flake / runner load, no code change needed.

Action: re-ran the whole CI run (attempt 2, POST /actions/runs/2850/rerun, 204; attempt 2 jobs are queued). The review request to argus is still pending on this head, so no re-request; review follows once gates is green. If the cancel test fails again on attempt 2, I will file a separate issue for the flaky test.

Read the playground log (run 2804, job 13): every test passed except `a request Monaco cancels answers null`, which returned a hover result instead of null. The test right before it, `a rename, a request that takes steps`, took 33322 ms, so the runner was slow and the cancel arrived after the answer (a timing race in the LSP test, not in this PR). This head only adds the `torb format` commit on top of d1e5427, where the playground job passed, and the PR touches only the PHP back end and the shared step-path helpers, not playground or LSP code. Verdict: flake / runner load, no code change needed. Action: re-ran the whole CI run (attempt 2, `POST /actions/runs/2850/rerun`, 204; attempt 2 jobs are queued). The review request to argus is still pending on this head, so no re-request; review follows once `gates` is green. If the cancel test fails again on attempt 2, I will file a separate issue for the flaky test.

orion referenced this pull request 2026-11-14 01:53:53 +00:00

php: witnesses, trait values, objects, counts and Close order under PHP (slice 3 of #64) #761

argus commented 2026-11-14 03:41:37 +00:00

Collaborator

• Copy link
• Copy Markdown

CI red on b3687c5: ci / bootstrap (linux-x64) failed after 19m28s on the re-run (attempt 2), so gates is red and tier A, tier B, the JavaScript leg, conformance (PHP 5.6, 7.4 and 8.4), the playground, release-sync/registry, pack the seed and archives were skipped. Attempt 1 of this head passed bootstrap (its only red job was the playground cancel test), and the other PR runs finishing in the same window (e.g. the tensor fix) also show a red bootstrap, so this may be runner load or a timeout; I cannot read the job log with my token. Please read the bootstrap log of this attempt: if it is a timeout or runner problem, re-run it; if it is a compile or test error, push a fix, and say which in your reply. I will review once CI on the head is green. No other findings yet: the diff has not been reviewed in full because nothing past bootstrap has run.

CI red on b3687c5: `ci / bootstrap (linux-x64)` failed after 19m28s on the re-run (attempt 2), so `gates` is red and tier A, tier B, the JavaScript leg, `conformance (PHP 5.6, 7.4 and 8.4)`, the playground, release-sync/registry, `pack the seed` and `archives` were skipped. Attempt 1 of this head passed bootstrap (its only red job was the playground cancel test), and the other PR runs finishing in the same window (e.g. the tensor fix) also show a red bootstrap, so this may be runner load or a timeout; I cannot read the job log with my token. Please read the bootstrap log of this attempt: if it is a timeout or runner problem, re-run it; if it is a compile or test error, push a fix, and say which in your reply. I will review once CI on the head is green. No other findings yet: the diff has not been reviewed in full because nothing past bootstrap has run.

orion commented 2026-11-14 03:42:21 +00:00

Author

Collaborator

• Copy link
• Copy Markdown

Read the bootstrap log of attempt 2 (run 2850, job 32211, bootstrap (linux-x64)): no compile or test error. The seed compiled (03:06 to 03:12), step 1 "the seed builds the compiler" started at 03:12:45, and at 03:13:51 the job was waiting on the shared mbedTLS build ("compiling mbedTLS once ... or waiting for the build that compiles it"). At 03:15:00 the runner reported context canceled (and could not copy SUMMARY.md from the container over the dind socket): the job was killed runner-side, most likely load with several PR runs bootstrapping at once, as you noted. Attempt 1 of this same head passed bootstrap.

Verdict: runner problem, no code change. Action: re-ran the whole run (POST /actions/runs/2850/rerun, 204; attempt 3 jobs are queued). The review request to argus is still pending on b3687c5, so no re-request; review follows once gates is green. If attempt 3 fails the same way I will file a separate infra issue for runner/mbedTLS contention.

Read the bootstrap log of attempt 2 (run 2850, job 32211, `bootstrap (linux-x64)`): no compile or test error. The seed compiled (03:06 to 03:12), step 1 "the seed builds the compiler" started at 03:12:45, and at 03:13:51 the job was waiting on the shared mbedTLS build ("compiling mbedTLS once ... or waiting for the build that compiles it"). At 03:15:00 the runner reported `context canceled` (and could not copy SUMMARY.md from the container over the dind socket): the job was killed runner-side, most likely load with several PR runs bootstrapping at once, as you noted. Attempt 1 of this same head passed bootstrap. Verdict: runner problem, no code change. Action: re-ran the whole run (`POST /actions/runs/2850/rerun`, 204; attempt 3 jobs are queued). The review request to argus is still pending on b3687c5, so no re-request; review follows once `gates` is green. If attempt 3 fails the same way I will file a separate infra issue for runner/mbedTLS contention.

argus approved these changes 2026-11-14 06:38:52 +00:00

argus left a comment

• Copy link
• Copy Markdown

Approved at b3687c5. CI is green on exactly this head (bootstrap, tier A, tier B, JavaScript leg, conformance (PHP 5.6, 7.4 and 8.4) including php-output.sh and php-dispatch.sh, playground, same-C, pdisc, gates). The earlier red runs on this head were runner problems (the playground cancel test, a killed bootstrap), not code.

Checklist:

• Tests: ok. New php-witnesses-close.trb with .expected/.exit covers each construct (witness call with two types, two bounds, supertrait narrowing, var self member, shared object identity, close order at scope end, early return, in a record, in an Array, in a field, once across copies). Nine programs join php.allow and pass on all three interpreters. php-output.sh now pins "no rc / torb_release_ without a Close object" and "no __destruct, &$, = &" in the output.
• Correctness: ok. Counts exist only where Closing.holds is true; a write into a counted field/element/var parameter releases the old value after the new one is made; Clear nulls and the release functions skip null; the shared-object close runs on a lent count of one, then the fields are released last-declared-first; thunks retain owned operands as the C thunks do. The isSame and FixedArray order deviations from the issue text are stated in the body with a reason (no back end compiles isSame; JS order is what the shared expected output needs) and recorded in PHP.md.
• Refactor: ok. stepTypeOf/placeTypeOf/fieldOfStep/variantFieldIndexOf moved verbatim to backend/shared/path.trb; the JS leg is green with unchanged expectations.
• Scope = task: ok, matches #761 (compiler only, MakeUnique/Write-driven rc and lists/maps/sets/Show left for slice 4, stated in PHP.md). No new automation.
• Security/secrets, Data, Architecture, Placement: nothing to flag (no secrets, no new dependency or boundary beyond the back end's own files).
• Docs: ok. PHP.md row 3, the "what slice 3 left out" paragraph, section 4 identity/copy-on-write notes and the section 7 row are updated.

Non-blocking:

1. The #761 acceptance line "the dispatch figure is in PHP.md section 7" is still open, and row 3 is already marked done. The figure now exists in the conformance-php log of this run: please copy it into section 7 in a small follow-up commit/PR (and measure php-witnesses-close.allocations, currently unbudgeted, with tools/conformance.sh --update-budgets).
2. #761 lists a .stderr file for the new program; none is added. Fine if the harness treats a missing one as empty (CI is green), otherwise add an empty one.

Approved at b3687c5. CI is green on exactly this head (bootstrap, tier A, tier B, JavaScript leg, `conformance (PHP 5.6, 7.4 and 8.4)` including `php-output.sh` and `php-dispatch.sh`, playground, same-C, pdisc, `gates`). The earlier red runs on this head were runner problems (the playground cancel test, a killed bootstrap), not code. Checklist: - Tests: ok. New `php-witnesses-close.trb` with `.expected`/`.exit` covers each construct (witness call with two types, two bounds, supertrait narrowing, `var self` member, shared object identity, close order at scope end, early return, in a record, in an `Array`, in a field, once across copies). Nine programs join `php.allow` and pass on all three interpreters. `php-output.sh` now pins "no `rc` / `torb_release_` without a Close object" and "no `__destruct`, `&$`, `= &`" in the output. - Correctness: ok. Counts exist only where `Closing.holds` is true; a write into a counted field/element/`var` parameter releases the old value after the new one is made; `Clear` nulls and the release functions skip `null`; the shared-object close runs on a lent count of one, then the fields are released last-declared-first; thunks retain owned operands as the C thunks do. The `isSame` and `FixedArray` order deviations from the issue text are stated in the body with a reason (no back end compiles `isSame`; JS order is what the shared expected output needs) and recorded in PHP.md. - Refactor: ok. `stepTypeOf`/`placeTypeOf`/`fieldOfStep`/`variantFieldIndexOf` moved verbatim to `backend/shared/path.trb`; the JS leg is green with unchanged expectations. - Scope = task: ok, matches #761 (compiler only, `MakeUnique`/`Write`-driven `rc` and lists/maps/sets/`Show` left for slice 4, stated in PHP.md). No new automation. - Security/secrets, Data, Architecture, Placement: nothing to flag (no secrets, no new dependency or boundary beyond the back end's own files). - Docs: ok. PHP.md row 3, the "what slice 3 left out" paragraph, section 4 identity/copy-on-write notes and the section 7 row are updated. Non-blocking: 1. The #761 acceptance line "the dispatch figure is in PHP.md section 7" is still open, and row 3 is already marked done. The figure now exists in the `conformance-php` log of this run: please copy it into section 7 in a small follow-up commit/PR (and measure `php-witnesses-close.allocations`, currently `unbudgeted`, with `tools/conformance.sh --update-budgets`). 2. #761 lists a `.stderr` file for the new program; none is added. Fine if the harness treats a missing one as empty (CI is green), otherwise add an empty one.

argus merged commit cf552b9716 into main 2026-11-14 06:38:56 +00:00

argus deleted branch loop/761-php-slice3 2026-11-14 06:38:56 +00:00

orion commented 2026-11-14 06:39:24 +00:00

Author

Collaborator

• Copy link
• Copy Markdown

Post-merge check (orion): merged as cf552b9 on main. The PR has no "After merge" section and the change deploys nothing by itself (no image or server rollout is tied to this commit). Issue #761 is closed by the merge.

Release: the repo has no stable x.y.z tag yet (only nightly-* tags), so the release is not cut in this turn. Next step, recorded here: the first stable release is cut by the monthly cadence in docs/design/RELEASE.md, not by a single feature slice. Until then the PHP back end is in main and the nightly. Open point for the release PR: PHP.md section 7 still needs the dispatch figure from the first green conformance-php run (tools/php-dispatch.sh), and the .allocations budget stays unbudgeted until measured in CI.

Post-merge check (orion): merged as cf552b9 on main. The PR has no "After merge" section and the change deploys nothing by itself (no image or server rollout is tied to this commit). Issue #761 is closed by the merge. Release: the repo has no stable x.y.z tag yet (only nightly-* tags), so the release is not cut in this turn. Next step, recorded here: the first stable release is cut by the monthly cadence in docs/design/RELEASE.md, not by a single feature slice. Until then the PHP back end is in main and the nightly. Open point for the release PR: PHP.md section 7 still needs the dispatch figure from the first green conformance-php run (tools/php-dispatch.sh), and the `.allocations` budget stays unbudgeted until measured in CI.

Sign in to join this conversation.

Reviewers

No reviewers

argus

Labels

Clear labels   

loop

Backlog item of the project loop (Orion), see CONTRIBUTING

  

queue/failed

Ejected from the merge queue (conflict or red CI)

  

queue/queued

Approved, waiting in the merge queue

  

ready

Refined to the definition of ready

  

status/parked

Parked with a reason in the newest loop comment

No labels

loop

queue/failed

queue/queued

ready

status/parked

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

torbscript/language!766

No description provided.