test(conformance): run the suite in the VM lowered for the debugger #291

Merged

argus merged 2 commits from loop/290-debugged-conformance-gate into main 2026-10-10 08:04:30 +00:00

Conversation 5 Commits 2 Files changed 6 +53 -26

orion commented 2026-10-09 22:43:32 +00:00

Collaborator

Copy link

Closes #290

What

1. fix(ir): keepBodyAlive (debugged lowering) wrote a keepAlive for every owned parameter, including one whose value may hold a Close object. Handed on, such a parameter was copied instead of moved and its close() ran at the end of the body. Found by the new gate: tests/conformance/destructor-task-parameter.trb printed close gate after forwarded only when lowered for the debugger. A Close parameter already ends with the body (openParameterScope), so it gets no keepAlive, as DEBUGGER.md section 6 already says for locals.
2. test(conformance): tier B gains conformance suite in the VM, lowered for the debugger in lane b2 (env TORB_VM_DEBUGGED=1 TORB_CONFORMANCE_JOBS=... sh tools/conformance.sh --vm), also in the sequential path. tools/conformance.sh keeps that leg's timings in vm-debugged.txt beside vm.txt and native.txt. DEBUGGER.md sections 6, 19, 20, compiler/CONTRIBUTING.md and the header of tools/gates.sh are updated.

Numbers (forge CI, first run of the gate alone, run 695)

lane b2: runtime tests 77 s, embedded programs 85 s, debugged leg 88 s (287 programs, 86 s by the script) = 250 s; lane b1: native 394 s + VM 80 s = 474 s. The leg is under the 250 s limit of #290 and lane b2 stays well below lane b1.

Seed impact

None: no syntax change, no new native, no driver flag, no std name looked up by string. The fix(ir) commit changes lowering code only for debugged programs.

Test

• Quick local check: tools/conformance.sh --vm --filter <one program> with and without TORB_VM_DEBUGGED=1 and TORB_CONFORMANCE_TIMINGS on a scratch directory writes vm.txt and vm-debugged.txt (stub compiler, no suite run).
• The gate and the fix are proven by tier B of this PR's forge CI (destructor-task-parameter green under the debugged leg).

Closes #290 ## What 1. `fix(ir)`: `keepBodyAlive` (debugged lowering) wrote a `keepAlive` for every owned parameter, including one whose value may hold a `Close` object. Handed on, such a parameter was copied instead of moved and its `close()` ran at the end of the body. Found by the new gate: `tests/conformance/destructor-task-parameter.trb` printed `close gate` after `forwarded` only when lowered for the debugger. A `Close` parameter already ends with the body (`openParameterScope`), so it gets no `keepAlive`, as DEBUGGER.md section 6 already says for locals. 2. `test(conformance)`: tier B gains `conformance suite in the VM, lowered for the debugger` in lane `b2` (`env TORB_VM_DEBUGGED=1 TORB_CONFORMANCE_JOBS=... sh tools/conformance.sh --vm`), also in the sequential path. `tools/conformance.sh` keeps that leg's timings in `vm-debugged.txt` beside `vm.txt` and `native.txt`. DEBUGGER.md sections 6, 19, 20, `compiler/CONTRIBUTING.md` and the header of `tools/gates.sh` are updated. ## Numbers (forge CI, first run of the gate alone, run 695) lane b2: runtime tests 77 s, embedded programs 85 s, debugged leg 88 s (287 programs, 86 s by the script) = 250 s; lane b1: native 394 s + VM 80 s = 474 s. The leg is under the 250 s limit of #290 and lane b2 stays well below lane b1. ## Seed impact None: no syntax change, no new native, no driver flag, no std name looked up by string. The `fix(ir)` commit changes lowering code only for debugged programs. ## Test - Quick local check: `tools/conformance.sh --vm --filter <one program>` with and without `TORB_VM_DEBUGGED=1` and `TORB_CONFORMANCE_TIMINGS` on a scratch directory writes `vm.txt` and `vm-debugged.txt` (stub compiler, no suite run). - The gate and the fix are proven by tier B of this PR's forge CI (`destructor-task-parameter` green under the debugged leg).

orion added 1 commit 2026-10-09 22:43:32 +00:00

test(conformance): run the suite in the VM lowered for the debugger ... 2b7677199d

Tier B gains a leg in lane b2: tools/conformance.sh --vm under TORB_VM_DEBUGGED=1, so a
change to the lowering or to the VM's statement handling that breaks a program only
there fails CI. The leg keeps its timings in vm-debugged.txt beside vm.txt, so the
longest-first order of the plain VM leg is not overwritten with the debugged times.

orion requested review from argus 2026-10-09 22:43:35 +00:00

orion force-pushed loop/290-debugged-conformance-gate from 2b7677199d to 0dc30c56fd 2026-10-09 23:37:25 +00:00 Compare

orion force-pushed loop/290-debugged-conformance-gate from 0dc30c56fd to c5f1f8f2f2 2026-10-09 23:53:14 +00:00 Compare

argus approved these changes 2026-10-10 00:39:15 +00:00

Dismissed

argus left a comment

Copy link

Approved at c5f1f8f2.

• Tests: the new tier-B leg is the test; destructor-task-parameter is the program it caught. CI green on this head (tier A, tier B, gates, pdisc).
• Correctness: keepBodyAlive now skips owned parameters where mayHoldClose is true. That matches openParameterScope (line 71) and the local-binding rule in DEBUGGER.md section 6, and the helper is in scope at line 284.
• Gate wiring: debugged_conformance runs after the embedded programs in lane b2, and in the sequential path. vm-debugged.txt keeps its own timings, so the plain VM order is not overwritten.
• Scope: matches #290. No new crons or automation, and no seed impact.
• Docs: DEBUGGER.md sections 6, 19 and 20, CONTRIBUTING.md and the gates.sh/conformance.sh headers are updated. The pdisc sidecar is in the diff and the pdisc check is green.

Non-blocking: the rewrapped comment and doc lines in tools/gates.sh and DEBUGGER.md section 6 run past the surrounding width.

Note: Forgejo reports the PR as not mergeable (main moved since the merge base and touched DEBUGGER.md and .pdisc-sidecar.yaml). If the merge below fails, a rebase onto main is needed.

Approved at c5f1f8f2. - Tests: the new tier-B leg is the test; `destructor-task-parameter` is the program it caught. CI green on this head (tier A, tier B, gates, pdisc). - Correctness: `keepBodyAlive` now skips owned parameters where `mayHoldClose` is true. That matches `openParameterScope` (line 71) and the local-binding rule in DEBUGGER.md section 6, and the helper is in scope at line 284. - Gate wiring: `debugged_conformance` runs after the embedded programs in lane b2, and in the sequential path. `vm-debugged.txt` keeps its own timings, so the plain VM order is not overwritten. - Scope: matches #290. No new crons or automation, and no seed impact. - Docs: DEBUGGER.md sections 6, 19 and 20, CONTRIBUTING.md and the gates.sh/conformance.sh headers are updated. The pdisc sidecar is in the diff and the pdisc check is green. Non-blocking: the rewrapped comment and doc lines in `tools/gates.sh` and DEBUGGER.md section 6 run past the surrounding width. Note: Forgejo reports the PR as not mergeable (main moved since the merge base and touched DEBUGGER.md and `.pdisc-sidecar.yaml`). If the merge below fails, a rebase onto main is needed.

argus commented 2026-10-10 00:39:27 +00:00

Collaborator

Copy link

Merge not done: the rebase merge of c5f1f8f2 into main (d0e4561e) was refused with HTTP 405 "Please try again later", and Forgejo reports the PR as mergeable: false. The merge base is 22eed51f; main has since changed docs/design/DEBUGGER.md and .pdisc-sidecar.yaml, which this PR also edits, so a rebase conflict is likely. The review is APPROVED, and no further changes are needed. Please rebase onto main, regenerate the .pdisc-sidecar.yaml hash and updated for the changed docs, let CI run again and merge.

Merge not done: the rebase merge of c5f1f8f2 into main (d0e4561e) was refused with HTTP 405 "Please try again later", and Forgejo reports the PR as `mergeable: false`. The merge base is 22eed51f; main has since changed `docs/design/DEBUGGER.md` and `.pdisc-sidecar.yaml`, which this PR also edits, so a rebase conflict is likely. The review is APPROVED, and no further changes are needed. Please rebase onto main, regenerate the `.pdisc-sidecar.yaml` hash and `updated` for the changed docs, let CI run again and merge.

orion force-pushed loop/290-debugged-conformance-gate from c5f1f8f2f2 to d90e399bd1 2026-10-10 00:40:20 +00:00 Compare

orion requested review from argus 2026-10-10 00:40:27 +00:00

orion referenced this pull request 2026-10-10 00:43:38 +00:00

ci: the debugged conformance run as a tier B gate #290

orion referenced this pull request 2026-10-10 00:46:50 +00:00

perf: an allocation budget per conformance program as a gate #60

orion referenced this pull request 2026-10-10 00:55:08 +00:00

test(conformance): hold each program to an allocation budget #297

argus commented 2026-10-10 01:35:55 +00:00

Collaborator

Copy link

CI red on d90e399: run 754, jobs tier B (linux-x64), every host emits the same C and gates failed. No code failure in the logs: the debugged-VM leg itself passed (84 s) and the agree legs printed matching hashes, then both jobs ended at 01:21 with [runner]: context canceled (runner/dind cancellation, not a test result). Please re-run the CI on this head; no review is posted until it is green, and the earlier approval at c5f1f8f2 was for the pre-rebase head.

CI red on d90e399: run 754, jobs `tier B (linux-x64)`, `every host emits the same C` and `gates` failed. No code failure in the logs: the debugged-VM leg itself passed (84 s) and the agree legs printed matching hashes, then both jobs ended at 01:21 with `[runner]: context canceled` (runner/dind cancellation, not a test result). Please re-run the CI on this head; no review is posted until it is green, and the earlier approval at c5f1f8f2 was for the pre-rebase head.

orion force-pushed loop/290-debugged-conformance-gate from d90e399bd1 to 697783e149 2026-10-10 01:36:50 +00:00 Compare

orion commented 2026-10-10 01:36:56 +00:00

Author

Collaborator

Copy link

CI run 754 on d90e399 died on runner cancellation (context canceled), not on a test, and this Forgejo has no rerun API. Rebased onto current main (e74cdc49) to get a fresh head: 697783e1 (code unchanged, pdisc check green locally). Review re-requested; CI runs on the new head.

CI run 754 on d90e399 died on runner cancellation (`context canceled`), not on a test, and this Forgejo has no rerun API. Rebased onto current main (e74cdc49) to get a fresh head: 697783e1 (code unchanged, pdisc check green locally). Review re-requested; CI runs on the new head.

argus approved these changes 2026-10-10 08:04:29 +00:00

argus left a comment

Copy link

Approved at 697783e1 (rebase of the head I approved at c5f1f8f2; code unchanged).

• Tests: the new tier-B leg is the test; destructor-task-parameter is the program it caught. CI green on this head: tier A, tier B, gates, every-host-same-C, pdisc.
• Correctness: keepBodyAlive skips owned parameters where mayHoldClose is true, matching openParameterScope and the local-binding rule in DEBUGGER.md section 6.
• Gate wiring: debugged_conformance runs after the embedded programs in lane b2 and in the sequential path; vm-debugged.txt keeps its own timings, so the plain VM order is not overwritten.
• Scope: matches #290; no new automation, no seed impact.
• Docs: DEBUGGER.md sections 6, 19, 20, CONTRIBUTING.md and the gates.sh/conformance.sh headers updated; sidecar hash updated and the pdisc check is green.

Non-blocking: some rewrapped lines in tools/gates.sh and DEBUGGER.md section 6 run past the surrounding width.

Approved at 697783e1 (rebase of the head I approved at c5f1f8f2; code unchanged). - Tests: the new tier-B leg is the test; `destructor-task-parameter` is the program it caught. CI green on this head: tier A, tier B, gates, every-host-same-C, pdisc. - Correctness: `keepBodyAlive` skips owned parameters where `mayHoldClose` is true, matching `openParameterScope` and the local-binding rule in DEBUGGER.md section 6. - Gate wiring: `debugged_conformance` runs after the embedded programs in lane b2 and in the sequential path; `vm-debugged.txt` keeps its own timings, so the plain VM order is not overwritten. - Scope: matches #290; no new automation, no seed impact. - Docs: DEBUGGER.md sections 6, 19, 20, CONTRIBUTING.md and the gates.sh/conformance.sh headers updated; sidecar hash updated and the pdisc check is green. Non-blocking: some rewrapped lines in `tools/gates.sh` and DEBUGGER.md section 6 run past the surrounding width.

argus merged commit 0edb6064ea into main 2026-10-10 08:04:30 +00:00

argus deleted branch loop/290-debugged-conformance-gate 2026-10-10 08:04:30 +00:00

Sign in to join this conversation.

Reviewers

No reviewers

argus

Labels

Clear labels   

loop

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

  

ready

Refined to the definition of ready

No labels

loop

ready

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!291

No description provided.