Repository navigation
0.23.0: a file that ends clangd costs the session that file, the session keeps a reserve of engine starts (#86), and code actions reach the editor (#89) - #88
Merged
Conversation
…session keeps a reserve of engine starts (#86) The reports of #86 after 0.22.0: clangd.okb ended by 0xC000001D on Windows 11 24H2 on every module unit that includes a library header (fixed in the engine, mcppls-clangd 0.0.22, tested here as a candidate), and the server turned that into a loop that spent the sandbox's grant table. - the exit keeps its number and its name (ExitCondition; no more -1); - the suspect is the file whose build was in flight (ast-build events), set aside at the first exit when it is the only one; an end from outside names nobody; - a crash ledger in the workspace's cache: one retry per term, whatever the reason; unchanged is not eligible after the second exit; a new session reads it; Restart clangd tries again; - a start budget: automatic restarts stop at a reserve of three starts (paused); the refused start is logged, counted truly and offers Restart Server; - the runner's fault report (okrun on Linux and macOS) is read into the exit; - Workspace::serve_kept finishes the job before detaching engine requests (a use after free the engine-trap fixture found); - D: lastExit.reproduce, fileIdentity, the payload's dirty stamp, the kit line; - tests: exits unit cases, two engine cases with a process that ends by a signal, conformance engine-trap, clangd-crash-closure under the ledger's rule, a windows-2025 conformance row; - series 0.23; packaging/engine-candidate.json names the engine PR's run. Design: .agents/docs/2026-11-12-issue-86-engine-trap-and-start-budget-design.md Plan: .agents/docs/2026-11-12-issue-86-implementation-plan.md Co-authored-by: speak-agent <248744407+speak-agent@users.noreply.github.com>
…ap's second session says so in its status A file set aside on the guess (asked about or edited lately, and clangd ended twice with nothing in flight) keeps the short term and the way back on an edit it had before: the ledger's long terms across sessions are for a file clangd named or was building. A new session with a ledger-held file open is degraded at once, with file-ended-engine: the fixture asserts that instead of ready. Co-authored-by: speak-agent <248744407+speak-agent@users.noreply.github.com>
…sks its CMake for its import std gate; the candidate is the engine PR's head - tests/unit: the engine trap case expected SIGILL on every host; on Windows the same fault is 0xC000001D. - ci: windows-2025 ships CMake 4.4 where windows-2022 has 3.31.6, and cmake-msvc-std pins the gate of the latter's experimental import std. The row asks the image's CMake which known gate is its own. - engine: a build is over when its diagnostics come, whoever they are for (a prime unit too). - packaging/engine-candidate.json: mcppls-clangd run 43198727766, commit 9c5b935 (the PR's head). - plan: the budget counts a workspace's own starts; windows-2025's CMake.
… is another toolchain windows-2025 has CMake 4.4.5 and MSVC 14.51. The fixture's own project is written for CMake 3.31.6's experimental import std: the gate is another UUID there, and with the right one CMake 4.4.5 builds the standard module's library and leaves it off the program's link line (LNK2001). That is the fixture's prepare, before any server runs. The probe of the gate goes again; 74 of the row's 75 cases run.
… is red The engine's run has one artifact since mcppls-clangd 0.0.21 (clangd-openkal), and download-artifact puts one match straight into its path: the payload job looked for the archive in candidate/clangd-openkal, found no part, and assembled the payload from the lock's engine under --candidate, without a word. This pull request's runs so far tested 23.1.0-mcppls.10 (the status of a conformance case said so: 15 features). - the archive is looked for wherever the artifact put it; no part with a candidate named stops the job; the job prints the engine of every platform and checks the payload carries the candidate's image; - mcppls-devtools payload --candidate refuses to run with no --clangd; - the candidate is the engine pull request's head (run 43213696250, c3669bd); - the changelog says which conformance case the 26100 row leaves out.
…release doc in Chinese says what the payload job checks
…ppls-clangd's PR 23, whose image links clangd's refactorings; conformance code-actions carries each one out; mtest 0.6.3 The bundled engine's image was linked from clangd's libraries alone and offered no refactoring; the server passed on the empty list. The runner answers workspace/applyEdit and its check code-action does what an editor does with an action (the literal's edit, or its command and the edit the server then asks for). The fixture asks on a class's name, a selected expression and auto, in a source file and a module unit.
…gine candidate is PR 23's present head stability (linux-x64, 1 of 2) failed twice at the first round of mcpp-split: query symbol hello::count answered in about 3 s without the definition, where the two later rounds and the run before answered in 4.5 s with it. The check reported the expectation alone. It now reports the command's output (provenance, snapshot, the symbol) and what it wrote to standard error, so a run that fails there says which answer was short. packaging/engine-candidate.json names mcppls-clangd run 43288717890 (e6b4756): the same engine inputs, with the code-actions case that finds a refactoring by its identifier. Co-authored-by: speak-agent <248744407+speak-agent@users.noreply.github.com>
…n progress report query symbol answered without the definition, and query refs with "no symbol named", when the query's own start was slow enough for the workspace to report it ($/progress, token mcppls/<root>): the kernel's sink counted every progress report as the engine's, so settle_index took the server's "loading" for the index begun and ended, skipped opening a unit, and asked workspace/symbol of a clangd that had not read its database. The answer was the module index's alone: the declaration. Reproduced here by a first query beside a session that had just begun (3 of 3 without the definition in 2.7 s; 4 of 4 with it in 6 s after the change), and it is what stability's first round on a fresh runner showed twice (3 s instead of 4.5 s, no definition). - orchestrator.client: status_progress_token / is_status_progress_token, the one definition the workspace writes its token by and the kernel tells it by; unit test in test_server. - kernel: the sink counts an engine's reports only. Co-authored-by: speak-agent <248744407+speak-agent@users.noreply.github.com>
…-mcppls.12); the candidate file is gone
Sunrisepeak
changed the title
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
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.
Server side of #86 after 0.22.0 — implements
.agents/docs/2026-11-12-issue-86-engine-trap-and-start-budget-design.md; the split, the order across repositories and what changed while implementing are in.agents/docs/2026-11-12-issue-86-implementation-plan.md.Companions: Sunrisepeak/mcppls-clangd#22 (the engine: the trap itself,
23.1.0-mcppls.11) and #23 (issue 89: the image links clangd's refactorings), tested here as the candidate and released asv0.0.23(23.1.0-mcppls.12), which the lock now pins (the candidate file is gone); mcpplibs/openkal-windows#47 (merged, 0.22.0: one identity for a file, grant slots returned).What the reports showed, and what changes
exitCode: -1, "clangd exited unexpectedly"clangd ended by an illegal instruction (0xC000001D) while building src/ui/monitor_view.cppm;exitCode,exitCondition,endedByEnvironment; the runner's fault block (Linux, macOS) asexception+stackmain.cpp)engine-paused); the refusal is logged, says the true count and offers Restart ServerAlso:
Workspace::serve_keptfinished its job after detaching the engines' requests — a clangd between an exit and its restart answered the detachment at once as cancelled, the editor gotcancelledinstead of the kept answer and the job was written to after it was erased (SIGSEGV under gdb; found by the new fixture).build.dirtyno longer countsmcpp.lock; the kit line says when its target is the host's default.Tests
tests/unit/test_exits.cpp: conditions (NTSTATUS, signals), builds in flight, suspects, the ledger's rules and its file, the budget's arithmetic (the report's 16×4 and CI's 21×3), the runner's fault lines.tests/unit/test_server.cpp: the engine with a process that ends by a signal while one file's build is in flight — set aside at the first exit, across an edit and a new session, Restart clangd tries again and the term doubles; and the reserve — an automatic restart above it, paused inside it, the person's restart made.engine-trap(Linux, macOS): the payload's clangd ended by SIGTRAP whenever it buildssrc/trap.cppm;clangd-crash-closureunder the ledger's rule.windows-2025(build 26100), where [engine-start-failed] clangd could not be started: granting C:/Program Files: out of memory #86's fault occurs and no row ran before: 74 of the 75 Windows cases.cmake-msvc-stdstays onwindows-2022alone -- its own project does not build with that image's CMake 4.4.5 (plan §2.14).Found while getting CI green (plan §2.13–2.19)
download-artifactput it and assembled the payload from the lock's engine, silently; three runs here were green on23.1.0-mcppls.10. The job now finds the part, prints the engine of every platform and fails when the payload does not carry the candidate's image;mcppls-devtools payload --candidaterefuses to run without an engine. The runs since tested the candidates, and the last the released23.1.0-mcppls.12.proot,clean machine), macOS runners not assigned, and one failure each ofmcpp-polling-watch-inputs@polling(windows-2025),mcpp-target-live(windows-2022) andverify-changesV5 in a stability round (darwin), each passing when run again.Local:
mcpp test625 cases green;engine-trapgreen against the 0.22.0 payload;mcppls-devtools check allgreen.Not in this PR (plan §2, §5)
The platform pin (openkal-platform with openkal-windows ≥ 0.22.0), the runner's trap report on Windows (E2 layer 1), WA-CLANGD-015 (the stopgap; the engine ships with this release),
windows-2025rows for vscode-e2e / clean-machine / bench.