Repository navigation
refactor: pull-based compilation for document lifecycle - #385
Merged
Merged
Conversation
Replace the push-based compilation model (schedule_build/run_build_drain with debounce timers) with a pull-based model where compilation is triggered lazily by feature requests via ensure_compiled(). - didOpen/didChange only update DocumentState and mark ast_dirty - didSave marks dependent documents dirty and queues indexing - ensure_compiled() handles PCM deps, PCH, and AST compilation on demand - Add CompileGraph::compile_deps() to compile module dependencies without compiling the file itself - Update integration tests for pull-based model - Fix URI encoding mismatch in test infrastructure Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughReplaces debounce-driven push builds with a pull-based on-demand compilation flow. Adds Changes
Sequence DiagramsequenceDiagram
participant Client as LSP Client
participant MS as MasterServer
participant CG as CompileGraph
participant Pool as CompilationPool
Client->>MS: hover/completion request
MS->>MS: ensure_compiled(path_id)
Note over MS: check DocumentState.ast_dirty
alt doc dirty
MS->>CG: compile_deps(path_id)
CG->>CG: compute ancestors & deps list
loop per dependency
CG->>Pool: dispatch compilation for dep (compile_impl dep, dispatch_self=false)
Pool-->>CG: compile result
end
CG-->>MS: deps succeeded/failed
MS->>MS: ensure_pch(path_id)
MS->>Pool: dispatch compilation for path_id
Pool-->>MS: compilation result
MS->>MS: publish diagnostics / set ast_dirty=false (on success)
else not dirty
Note over MS: fast-path (no compile)
end
MS-->>Client: respond to original LSP request
Estimated Code Review Effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
…iled - Extract finish() lambda to deduplicate compiling/completion cleanup - Use early return to reduce nesting in compile_deps and ensure_compiled - Flatten if/else into early-return style for result handling Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 7
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@src/server/master_server.cpp`:
- Around line 768-769: Stateless, module-aware requests are skipping the
preflight compilation, so update forward_stateless() to call
ensure_compiled(path_id) (and co_await its result) before handling requests like
completion/signatureHelp when the request is module-aware; if
ensure_compiled(...) returns false, return serde_raw{"null"} just like the
stateful path does. Also apply the same guard where similar stateless branches
exist (the other occurrence referenced around the second diff) so that pcm_paths
cleared by didSave() are rebuilt before servicing stateless module requests.
- Around line 376-379: The fast-path in ensure_compiled() incorrectly returns
early on !doc.ast_dirty even though didSave() can clear pch_hashes and thereby
invalidate stateful AST for headers; update the fast-path to conservatively
bypass the early return when pch_hashes were cleared/changed by a save (i.e.,
treat the TU as dirty if pch_hashes is empty/has been modified) so
hover/definition/semantic tokens rebuild using a fresh AST; locate references to
didSave(), pch_hashes and ensure_compiled()/doc.ast_dirty to implement this
change, and apply the same conservative check at the other occurrence around the
1379-1384 region.
- Around line 438-442: ensure_pch(...) may fail but the code always re-attaches
the previous PCH from pch_paths[path_id]; capture the boolean result of co_await
ensure_pch(path_id, params.path, params.text, params.directory,
params.arguments) and only set params.pch from pch_paths and pch_bounds when
that result is true; if ensure_pch returns false, clear or avoid using the
existing pch_paths[path_id] (e.g. do not assign params.pch or explicitly reset
it) so a failed rebuild cannot leave a stale PCH attached.
- Around line 469-489: The code currently returns true for a generation-mismatch
or failed compile which lets forward_stateful() run against a stale AST and
clears/marks the document clean incorrectly; change the logic so that only a
successful compile for the current generation returns true: when doc2.generation
!= gen, return false (do not clear diagnostics or set doc2.ast_dirty=false);
when result.has_value() is false, do not clear diagnostics or set
ast_dirty=false and return false so the compile loop can retry; only after
publish_diagnostics(...) for the matching generation set doc2.ast_dirty = false,
schedule_indexing(), and return true; reference ensure_compiled(),
forward_stateful(), publish_diagnostics(), clear_diagnostics(), and
doc2.ast_dirty to locate and update the branches.
In `@tests/conftest.py`:
- Around line 75-78: The test client currently unquotes URIs before sending and
only stores diagnostics under a single key; change sends the percent-encoded URI
produced by Path.as_uri() directly when calling didOpen/other wire requests (do
not unquote before sending) and update the diagnostics handling in
_normalize_uri / diagnostics storage so that when receiving diagnostics you
store them under both the raw server-provided URI and the normalized form (i.e.,
keep diagnostics[uri] and also diagnostics[self._normalize_uri(uri)]) so lookups
succeed for either representation; ensure diagnostics_events uses the same
dual-key logic (references: _normalize_uri, diagnostics, diagnostics_events,
didOpen).
In `@tests/integration/test_modules.py`:
- Around line 192-201: The test reads client.diagnostics for leaf_uri but may
see stale diagnostics from the first compile; before calling
client.text_document_hover_async()/assert on client.diagnostics,
subscribe/capture the next diagnostics publish for leaf_uri and await it so you
only read the fresh diagnostics produced after reopening leaf.cppm—e.g.,
register a diagnostics listener or use the test harness's "next diagnostics"
helper for leaf_uri, await that event, then call
client.text_document_hover_async(...) and assert
len(client.diagnostics.get(leaf_uri, [])) == 0.
In `@tests/integration/test_pch.py`:
- Around line 44-49: The test erroneously asserts uri in client.diagnostics
which was already set by open_and_wait(), so change the test to record the prior
diagnostics (e.g., prev = client.diagnostics.get(uri)) before sending the
hover/didChange and then wait for the next diagnostics notification for that uri
(use the client's diagnostics-wait helper or loop/await until
client.diagnostics[uri] != prev) before asserting; update the code around
client.text_document_hover_async / client.diagnostics to wait for the new
diagnostics publication rather than relying on the existing entry and then close
the document as before.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: 7dfbc840-4d34-4137-b818-3921d739be39
📒 Files selected for processing (11)
src/server/compile_graph.cppsrc/server/compile_graph.hsrc/server/config.hsrc/server/master_server.cppsrc/server/master_server.htests/conftest.pytests/integration/test_file_operation.pytests/integration/test_index.pytests/integration/test_modules.pytests/integration/test_pch.pytests/unit/server/compile_graph_tests.cpp
💤 Files with no reviewable changes (1)
- src/server/config.h
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@src/server/master_server.cpp`:
- Around line 1365-1373: The loop that marks ast_dirty for documents currently
skips the saved file by checking if(dirty_id == path_id) which prevents the
saved/open file from being marked; to fix, treat the saved path like other
dirtied entries by removing or changing the conditional so that when iterating
over dirtied (variable dirtied) you still lookup in documents and set
doc_it->second.ast_dirty = true for dirty_id equal to path_id as well; update
the loop around documents.find(dirty_id) and the check involving path_id so the
saved file's ast_dirty is set when it is present in documents.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: f526dff5-1823-4df4-bf47-a0eeb467fbf9
📒 Files selected for processing (2)
src/server/compile_graph.cppsrc/server/master_server.cpp
Extract event_loop/schedule/run pattern into execute() helper. Use std::optional<CompileGraph> and setup()/teardown() for shared state. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Remove redundant setup() since TEST_SUITE creates fresh instances per case. Remove unnecessary inline on file-scope helper functions. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (1)
tests/unit/server/compile_graph_tests.cpp (1)
416-418: Minor: Inconsistent use ofllvm::findvsranges::find.The file uses
ranges::findelsewhere (lines 101-102, 123-125, etc.), but this test usesllvm::find. Consider usingranges::findfor consistency.♻️ Suggested fix
- EXPECT_TRUE(llvm::find(dirtied, 1u) != dirtied.end()); - EXPECT_TRUE(llvm::find(dirtied, 2u) != dirtied.end()); - EXPECT_TRUE(llvm::find(dirtied, 3u) != dirtied.end()); + EXPECT_TRUE(ranges::find(dirtied, 1u) != dirtied.end()); + EXPECT_TRUE(ranges::find(dirtied, 2u) != dirtied.end()); + EXPECT_TRUE(ranges::find(dirtied, 3u) != dirtied.end());🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@tests/unit/server/compile_graph_tests.cpp` around lines 416 - 418, The three EXPECT_TRUE assertions use llvm::find but the test suite otherwise uses ranges::find; update the assertions in tests/unit/server/compile_graph_tests.cpp to use ranges::find(dirtied, Xu) instead of llvm::find so the checks remain identical but consistent with other tests (update the three lines that call EXPECT_TRUE(llvm::find(dirtied, 1u) != dirtied.end()) / for 2u and 3u to use ranges::find and compare against dirtied.end()); keep the existing EXPECT_TRUE and the dirtied variable unchanged.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@tests/unit/server/compile_graph_tests.cpp`:
- Around line 59-73: The tests share the global vector compiled and never call
setup(), causing state leakage; fix by either (A) invoking setup() at the top of
every TEST_CASE that uses compiled/graph (e.g., add a call to setup(); as the
first statement in those tests that also use execute()/event loop), or (B)
convert to a doctest fixture: create a struct (e.g., CompileGraphFixture) with
setup() as its constructor that clears compiled and resets graph, then change
TEST_CASE to TEST_CASE_FIXTURE(CompileGraphFixture, "name") so setup runs before
each test. Ensure references to compiled, graph, setup(), and execute() remain
valid after the change.
---
Nitpick comments:
In `@tests/unit/server/compile_graph_tests.cpp`:
- Around line 416-418: The three EXPECT_TRUE assertions use llvm::find but the
test suite otherwise uses ranges::find; update the assertions in
tests/unit/server/compile_graph_tests.cpp to use ranges::find(dirtied, Xu)
instead of llvm::find so the checks remain identical but consistent with other
tests (update the three lines that call EXPECT_TRUE(llvm::find(dirtied, 1u) !=
dirtied.end()) / for 2u and 3u to use ranges::find and compare against
dirtied.end()); keep the existing EXPECT_TRUE and the dirtied variable
unchanged.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: 72f449d3-7d7b-4cb5-8a42-42dff6ae2eb1
📒 Files selected for processing (1)
tests/unit/server/compile_graph_tests.cpp
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
tests/unit/server/compile_graph_tests.cpp (1)
494-516:⚠️ Potential issue | 🟡 MinorSynchronize
update(1)with dispatch actually starting.Line 513 currently relies on
compilerrunning far enough that unit1is already in-flight beforeupdaterexecutes. If scheduling orcompile()changes,graph->update(1)can run first and this test stops exercising cancellation at all. Gateupdateron adispatch_startedevent fromgated_dispatch().Possible hardening
TEST_CASE(UpdateDuringCompile) { et::event_loop loop; et::event gate; + et::event dispatch_started; - auto gated_dispatch = [&gate](std::uint32_t) -> et::task<bool> { + auto gated_dispatch = [&gate, &dispatch_started](std::uint32_t) -> et::task<bool> { + dispatch_started.set(); co_await gate.wait(); co_return true; }; @@ // Coroutine 2: update(1) while dispatch is in flight, then unblock gate. auto updater = [&]() -> et::task<> { + co_await dispatch_started.wait(); graph->update(1); gate.set(); co_return; };Also applies to: 518-522
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@tests/unit/server/compile_graph_tests.cpp` around lines 494 - 516, The test races because updater calls graph->update(1) before the dispatched coroutine is actually started; modify gated_dispatch (the lambda used as the dispatch) to signal a "dispatch_started" synchronization primitive (e.g., a second et::gate or promise/flag) immediately when it begins, then co_await the existing gate; change updater to wait on that dispatch_started signal before calling graph->update(1) (then unblock the original gate as before). Reference: gated_dispatch, updater, compiler, graph->update(1), gate.set().
🧹 Nitpick comments (1)
tests/unit/server/compile_graph_tests.cpp (1)
680-708: Force the shared dependency to stay in-flight before asserting concurrent dedup.Lines 693-695 launch both requests, but
tracking_dispatch()returns immediately. That means this still passes in a schedule where request A fully compiles dep3before request B reaches it, so the test only proves clean-skip reuse, not the new in-progress dedup path. Hold path3inside dispatch and queue the second request while it is blocked.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@tests/unit/server/compile_graph_tests.cpp` around lines 680 - 708, The test currently races both compile_deps calls but tracking_dispatch(compiled) completes immediately, so it doesn't exercise the in-flight dedup path; modify the test (CompileDepsConcurrentDedup) so tracking_dispatch blocks when invoked for dependency 3 (e.g., use a latch/promise/future or condition_variable inside tracking_dispatch to keep dep 3 "in-flight"), start both graph->compile_deps(1) and graph->compile_deps(2) while dep 3 is blocked, then release the blocker for dep 3 only after both calls have been launched and waiting; ensure you still record compiled entries into the compiled vector and then assert the three unique deps were compiled exactly once.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Outside diff comments:
In `@tests/unit/server/compile_graph_tests.cpp`:
- Around line 494-516: The test races because updater calls graph->update(1)
before the dispatched coroutine is actually started; modify gated_dispatch (the
lambda used as the dispatch) to signal a "dispatch_started" synchronization
primitive (e.g., a second et::gate or promise/flag) immediately when it begins,
then co_await the existing gate; change updater to wait on that dispatch_started
signal before calling graph->update(1) (then unblock the original gate as
before). Reference: gated_dispatch, updater, compiler, graph->update(1),
gate.set().
---
Nitpick comments:
In `@tests/unit/server/compile_graph_tests.cpp`:
- Around line 680-708: The test currently races both compile_deps calls but
tracking_dispatch(compiled) completes immediately, so it doesn't exercise the
in-flight dedup path; modify the test (CompileDepsConcurrentDedup) so
tracking_dispatch blocks when invoked for dependency 3 (e.g., use a
latch/promise/future or condition_variable inside tracking_dispatch to keep dep
3 "in-flight"), start both graph->compile_deps(1) and graph->compile_deps(2)
while dep 3 is blocked, then release the blocker for dep 3 only after both calls
have been launched and waiting; ensure you still record compiled entries into
the compiled vector and then assert the three unique deps were compiled exactly
once.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: 726ed29f-344c-49d5-b7fe-716051b3dc54
📒 Files selected for processing (1)
tests/unit/server/compile_graph_tests.cpp
- didSave: mark saved file itself as dirty (not just dependents) - didSave: mark ALL open docs dirty after pch_hashes.clear() for header saves not tracked by CompileGraph - ensure_compiled: return false on generation mismatch (stale AST) - ensure_pch: only attach PCH when build succeeds - forward_stateless: compile module deps before stateless requests - Fix URI encoding: send encoded URI on wire, normalize for lookups - Tests: wait for fresh diagnostics after reopen/hover Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
src/server/master_server.cpp (1)
830-834:⚠️ Potential issue | 🟠 MajorGuard stateless PCH attachment on rebuild success.
Line 831 awaits
ensure_pch(...)but ignores its boolean result, and Lines 832-834 still attach any cached PCH. If rebuild fails, completion/signatureHelp can run with stale preamble state.Proposed fix
- // Ensure PCH is available for stateless compilation (completion/signatureHelp). - co_await ensure_pch(path_id, path, wp.text, wp.directory, wp.arguments); - if(auto pch_it = pch_paths.find(path_id); pch_it != pch_paths.end()) { - wp.pch = {pch_it->second, pch_bounds[path_id]}; - } + // Ensure PCH is available for stateless compilation (completion/signatureHelp). + auto pch_ok = co_await ensure_pch(path_id, path, wp.text, wp.directory, wp.arguments); + if(pch_ok) { + if(auto pch_it = pch_paths.find(path_id); pch_it != pch_paths.end()) { + wp.pch = {pch_it->second, pch_bounds[path_id]}; + } + }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/server/master_server.cpp` around lines 830 - 834, ensure_pch(...) can fail but its boolean result is ignored; only attach cached PCH to wp.pch when ensure_pch(path_id, path, wp.text, wp.directory, wp.arguments) returns true and the entry exists. Modify the code around the call to ensure_pch (the ensure_pch invocation and the subsequent use of pch_paths, pch_bounds, and wp.pch) so you capture the returned bool and conditionally set wp.pch = {pch_it->second, pch_bounds[path_id]} only when the rebuild succeeded and pch_it != pch_paths.end().
♻️ Duplicate comments (1)
tests/conftest.py (1)
167-171:⚠️ Potential issue | 🟠 MajorUse encoded URI for hover wire request as well.
open()correctly sends encodedPath.as_uri()on the wire, butopen_and_wait()sends the normalized URI inHoverParams(Line 169). For files with reserved characters, this can break URI round-tripping and request targeting.🔧 Suggested fix
- uri, content = self.open(filepath) + uri, content = self.open(filepath) + wire_uri = filepath.as_uri() event = self.wait_for_diagnostics(uri) @@ await self.text_document_hover_async( HoverParams( - text_document=TextDocumentIdentifier(uri=uri), + text_document=TextDocumentIdentifier(uri=wire_uri), position=Position(line=0, character=0), ) )🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@tests/conftest.py` around lines 167 - 171, The hover request is using the unencoded/native `uri` so URIs with reserved characters don't round-trip; update the `HoverParams` call in `text_document_hover_async` usage to pass an encoded URI (the same form used by `open()`), e.g. replace `TextDocumentIdentifier(uri=uri)` with an encoded form such as `TextDocumentIdentifier(uri=Path(uri).as_uri())` or reuse the existing `encoded_uri` variable so `text_document_hover_async(... HoverParams(...))` sends the same encoded `Path.as_uri()` value as `open()`.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Outside diff comments:
In `@src/server/master_server.cpp`:
- Around line 830-834: ensure_pch(...) can fail but its boolean result is
ignored; only attach cached PCH to wp.pch when ensure_pch(path_id, path,
wp.text, wp.directory, wp.arguments) returns true and the entry exists. Modify
the code around the call to ensure_pch (the ensure_pch invocation and the
subsequent use of pch_paths, pch_bounds, and wp.pch) so you capture the returned
bool and conditionally set wp.pch = {pch_it->second, pch_bounds[path_id]} only
when the rebuild succeeded and pch_it != pch_paths.end().
---
Duplicate comments:
In `@tests/conftest.py`:
- Around line 167-171: The hover request is using the unencoded/native `uri` so
URIs with reserved characters don't round-trip; update the `HoverParams` call in
`text_document_hover_async` usage to pass an encoded URI (the same form used by
`open()`), e.g. replace `TextDocumentIdentifier(uri=uri)` with an encoded form
such as `TextDocumentIdentifier(uri=Path(uri).as_uri())` or reuse the existing
`encoded_uri` variable so `text_document_hover_async(... HoverParams(...))`
sends the same encoded `Path.as_uri()` value as `open()`.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
Run ID: f42a1c5d-1256-47f6-ac65-2680d79b7a37
📒 Files selected for processing (4)
src/server/master_server.cpptests/conftest.pytests/integration/test_modules.pytests/integration/test_pch.py
🚧 Files skipped from review as they are similar to previous changes (2)
- tests/integration/test_pch.py
- tests/integration/test_modules.py
- Extract ensure_deps() to share module deps / PCH / PCM filling between ensure_compiled() and forward_stateless() - Remove setup() since TEST_SUITE members are fresh per test case - Remove redundant `inline` on file-scope test helpers Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This was referenced Apr 3, 2026
Merged
This was referenced Apr 17, 2026
5 tasks done
This was referenced Jun 11, 2026
This was referenced Jul 5, 2026
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.
Summary
Replace the push-based compilation model with a pull-based (lazy) model where compilation is driven entirely by feature requests.
Server core (
master_server.cpp/h)schedule_build(),run_build_drain(), debounce timers, andDocumentStateflags (build_running,build_requested,drain_scheduled)debounce_msconfig fielddidOpen/didChangeonly updateDocumentStateand markast_dirty— no compilation triggereddidSavemarks dependent docs dirty viaCompileGraph::update(), invalidates PCH hashes, marks all open documentsast_dirty(header saves), and queues background indexingensure_compiled(path_id)— the pull-based entry point called byforward_stateful()/forward_stateless()before every feature request:!ast_dirtycompile_graph->compile_deps()ensure_pch()(only attach on success)CompileParamsto stateful workerfalse, keep dirty for retryforward_stateless()now also callscompile_graph->compile_deps()before stateless requests (completion/signatureHelp)resolve_fn(was duplicated inrun_build_drainandensure_compiled)CompileGraph (
compile_graph.cpp/h)compile_deps(path_id)— compiles all transitive module dependencies but NOT the file itself (used for plain .cpp files thatimportmodules)compile/compile_depsviacompile_impl(path_id, ancestors, dispatch_self)parametercompile_depscompiles dependencies concurrently viawhen_allfinish()lambda to deduplicatecompiling=false; completion->set()cleanup across all exit pathsstd::ranges::removeinstead of legacystd::removeTest infrastructure (
conftest.py)open_and_wait()now sends a hover request to triggerensure_compiled()(pull-based model requires a feature request to compile)_normalize_uri()helper usingurllib.parse.unquoteIntegration tests
didOpendiagnostics_wait_for_index()sends hover to trigger compilation before pollingworkspace/symboltest_hover_save_closesimplified — hover directly triggers compilationtest_save_recompileandtest_pch_*wait for fresh diagnostics after hover-triggered recompilationUnit tests (
compile_graph_tests.cpp)compiled/graphas TEST_SUITE members withstd::optional<CompileGraph>execute(callback)helper to deduplicate event_loop boilerplatecompile_depstests: no-deps, single dep, chain, diamond, failure, plain-cpp, concurrent dedup, resolve-onceinlineon file-scope helpersTest plan
🤖 Generated with Claude Code