## Summary
- **Restructure `src/server/` into subdirectories** (`service/`,
`compiler/`, `worker/`, `workspace/`, `protocol/`) to separate concerns:
transport/session management, compilation, worker orchestration, and
persistent workspace state.
- **Decouple MasterServer from transport**: MasterServer no longer holds
a `JsonPeer&` reference or registers handlers itself. New `LSPClient`
and `AgentClient` classes own their peer references and register
protocol handlers, accessing MasterServer internals via `friend class`.
- **Add agentic protocol**: A TCP-based side channel
(`agentic/compileCommand`) that lets external tools (AI agents, build
systems) query compile commands from a running clice server. Includes a
CLI client mode (`--mode agentic --port N --path FILE`), server-side
listener when `--port` is specified in pipe mode, and integration tests
for happy path, fallback, concurrency, and connection-refused.
- **Replace fire-and-forget `loop.schedule()` with `kota::task_group`**:
Compiler compile tasks, Indexer background indexing + resource monitor,
WorkerPool worker monitors, and socket accept loops now use structured
concurrency. This eliminates manual `alive_count_`/generation counters
and ensures all spawned tasks are joined on shutdown.
- **Fix flaky integration test**: `CliceClient.initialize()` now always
sets `cache_dir` to a workspace-local `.clice/` directory, preventing
stale PCH artifacts from the global `~/.cache/clice/` from polluting
test runs.
## Details
**Compiler peer lifetime**: `Compiler` and `Indexer` previously took
`JsonPeer&` in their constructors, coupling them to a single connection.
They now store a `JsonPeer*` set via `set_peer()`, with null checks
before sending diagnostics/progress. This supports the multi-connection
model where agentic clients don't need diagnostics.
**Socket mode single-LSP enforcement**: `accept_connections()` takes a
`register_lsp` flag; when true, only the first connection gets an
`LSPClient`. All connections get an `AgentClient`. This prevents
multiple LSP sessions from racing on shared server state.
**Structured shutdown**: `Compiler::stop()` cancels in-flight compile
tasks and joins them. `WorkerPool::stop()` signals workers and joins the
monitor task group. `Indexer` uses a `cancellation_source` to stop its
resource monitor when a background indexing run completes.
**Pin kotatsu**: Changed from `GIT_TAG main` + `GIT_SHALLOW TRUE` to an
exact commit hash for reproducible builds.
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
## Summary
Three pre-existing bugs cause worker processes to crash with SEGV or
SIGABRT. On the main branch these crashes are silent (workers die,
requests fail fast with "transport closed", tests still pass because
null responses are accepted). However when combined with #432's worker
respawn mechanism, the crash-respawn-crash cycle on low-core CI machines
causes request timeouts and smoke test hangs.
### Fixes
- **compilation.cpp**: `ProxyAction::CreateASTConsumer` now checks for
null before passing to `MultiplexConsumer`. When the wrapped action's
`CreateASTConsumer` fails (e.g. missing system headers during PCH
generation), this previously caused a null pointer dereference, SEGV,
ASAN kills the stateless worker.
- **compilation_unit.cpp**: `file_path()` returns empty `StringRef` on
invalid `FileID` instead of asserting. The assert fired when
`IncludeGraph::from()` called `file_path(interested_file())` on an AST
compiled with synthesized default commands (no compile_commands.json,
clang++ -std=c++20 fallback, no system headers, invalid main file ID),
SIGABRT, stateful worker crash.
- **compiler.cpp**: `ensure_pch` now creates the PCH cache directory
before sending the build request. Previously, when `load_workspace()`
exited early (no compile_commands.json), the cache subdirectories were
never created, causing every PCH write to fail with "No such file or
directory".
- **master_server.cpp/h**: `load_workspace()` changed from
`kota::task<>` to plain `void` -- it contains only synchronous
filesystem operations and no co_await, so the coroutine wrapper was
unnecessary. Called directly instead of via `loop.schedule()`.
## Test plan
- [x] Verified zero SEGV/SIGABRT/assertion crashes in worker stderr
after fix
- [x] rapid_edit.jsonl smoke test passes 3/3 runs consistently (34s
each)
- [x] Behavior matches main branch (both return 134 responses, 0
pending)
- [x] Debug build with ASAN (detect_leaks=0) -- clean run, no sanitizer
reports
<!-- codesmith:footer -->
---
<a
href="https://app.blacksmith.sh/clice-io/codesmith/clice/pr/435"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-light.svg"><img
alt="View in Codesmith"
src="https://pr-comments-assets.blacksmith.sh/codesmith/view-in-codesmith-dark.svg"></picture></a>
<sup>Codesmith can help with this PR — just tag <code>@codesmith</code>
or enable autofix.</sup>
- [ ] Autofix CI and bot reviews
<!-- /codesmith:footer -->
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Bug Fixes**
* Improved error handling for AST consumer creation with null checks and
a clear failure path.
* Safer file-path access that returns empty for invalid identifiers
instead of asserting.
* PCH cache handling now validates cache configuration, attempts
directory creation, logs warnings, and aborts PCH builds on failure.
* **Refactor**
* Workspace loading changed from asynchronous to synchronous execution.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
## Summary
- The `eventide` dep was renamed to
[kotatsu](https://github.com/clice-io/kotatsu) with a broad rename of
CMake identifiers, namespaces, header paths, and a few module reorgs
(`serde` → `codec`, `reflection` → `meta`, `common` → `support`). Align
clice to the new names.
- CMake: FetchContent target, option prefix (`ETD_*` → `KOTA_*`,
`ETD_SERDE_*` → `KOTA_CODEC_*`), target names
(`eventide::{ipc::lsp,serde::toml,deco,zest}` →
`kota::{ipc::lsp,codec::toml,deco,zest}`).
- Namespaces: `eventide::` → `kota::`, `eventide::serde::` →
`kota::codec::`, `eventide::refl::` → `kota::meta::`. The short `et`
alias is dropped — all usages now spell `kota::` directly.
- Headers: `eventide/*` → `kota/*`, including special cases
`serde/serde/raw_value.h` → `codec/raw_value.h`, `ipc/json_codec.h` →
`ipc/codec/json.h`, `common/meta.h` → `support/type_traits.h`,
`common/ranges.h` → `support/ranges.h`.
- Kotatsu split `JsonPeer` / `BincodePeer` out of `ipc/peer.h` into the
codec-specific headers; added `kota/ipc/codec/{json,bincode}.h` includes
where those types are used.
- Depends on clice-io/kotatsu#110 (already merged) to prevent `-Wall
-Wextra -Werror` from transitively propagating out of
`kota::project_options`.
## Test plan
- [x] `pixi run unit-test RelWithDebInfo` — 518/518 pass (9 skipped,
unchanged from main)
- [x] `pixi run integration-test RelWithDebInfo` — 119/119 pass
- [x] `pixi run smoke-test RelWithDebInfo` — 2/2 pass
- [x] `pixi run format` clean
## Notes
- `tests/smoke/rapid_edit.jsonl` was intentionally left untouched: the
embedded `#include "eventide/..."` strings are frozen snapshots of file
contents the client sent at record time, not clice source.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Chores**
* Updated internal dependencies from `eventide` to `kota`, including
async runtime, IPC transport, serialization codec, and metadata
libraries.
* Updated build configuration and CMake variables to align with the new
dependency.
* **Refactor**
* Migrated internal implementation to use `kota` namespace and APIs
throughout the codebase.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
## Summary
- **Extract `Indexer` class** — owns all index state (ProjectIndex,
MergedIndex shards, OpenFileIndex) and query methods (definition,
references, call/type hierarchy, workspace symbol search)
- **Extract `Compiler` class** — owns document state, PCH/PCM cache,
compile argument resolution, header context, `ensure_compiled`, and
worker forwarding
- **MasterServer is now a pure LSP handler registration layer** (~700
lines, down from ~3200)
- **`MergedIndexShard`** wraps `index::MergedIndex` with a lazily-cached
PositionMapper; `OpenFileIndex` gains matching
`find_occurrence()`/`find_relations()` APIs — callers get pre-converted
LSP ranges directly
- **Indexer returns typed values** (`vector<Location>`,
`vector<CallHierarchyIncomingCall>`, etc.) instead of pre-serialized
JSON, fixing the references handler from JSON string surgery to simple
vector concatenation
- **Fix**: duplicate `workspace/symbol` loop in the original code
## Test plan
- [x] 465 unit tests pass
- [x] 113 integration tests pass
- [x] 2/2 smoke tests pass
- [x] `clang-format` applied
🤖 Generated with [Claude Code](https://claude.com/claude-code)
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Server-side C++ compilation orchestration (module & precompiled header
builds) with LSP-integrated document handling.
* **Improvements**
* Deterministic, persistent, dependency-aware caching to avoid redundant
rebuilds and speed up incremental work.
* Better cross-file indexing and navigation, improved diagnostics and
more reliable include/import-aware completions.
* **Tests**
* Unit tests updated to the unified worker query/build request shapes.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
## Summary
PCH and PCM artifacts are now cached to disk at
`.clice/cache/{pch/,pcm/}` with content-addressed filenames, so they
survive server restarts. Dependency metadata is persisted in
`cache.json` (using eventide serde) with a shared path table for
deduplication.
### Key changes
- **protocol.h**: `output_path` field on `BuildPCHParams` /
`BuildPCMParams` so master specifies where workers write
- **stateless_worker.cpp**: Atomic write via `.tmp` + `fs::rename`;
`CompilationUnit` destroyed before rename to flush the file to disk;
fallback to temp file when `output_path` is empty (unit tests)
- **master_server.h**: `PCMState` struct, `pcm_states` map,
`load_cache()` / `save_cache()` / `cleanup_cache()` methods
- **master_server.cpp**: Cache lifecycle — load from `cache.json` on
startup, save after each PCH/PCM build and on exit; deterministic path
computation (`xxh3` preamble hash for PCH, module name + source path
hash for PCM); stale files (>7 days) cleaned on startup; `cache.json`
uses shared path table to avoid redundant storage of header paths across
entries
- **filesystem.h**: `fs::rename()` helper; `ThreadSafeFS` broadened to
match `.pch` extension instead of `preamble-` prefix
- **tests**: 11 new integration tests covering PCH/PCM persistence,
cross-session reuse, staleness detection, shared preamble dedup, and
restart survival; unit tests updated with `output_path`
### Naming scheme
- **PCH**: `.clice/cache/pch/<016x(xxh3(preamble))>.pch`
- **PCM**:
`.clice/cache/pcm/<module_name>-<016x(xxh3(source_path))>.pcm`
## Test plan
- [x] Unit tests — 448 passed
- [x] Integration tests — 92 passed (including 11 new persistent cache
tests)
- [x] Smoke tests — 1 passed
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
## Summary
- Add `ensure_pch()` helper to MasterServer that builds/reuses
precompiled headers via stateless workers, with preamble hash-based
staleness detection (xxh3_64bits)
- Fix `BuildPCHParams` to carry `preamble_bound` so the stateless worker
truncates content at the preamble boundary (fixes redefinition errors
when PCH included full file)
- Wire PCH into both `run_build_drain` (stateful compile path) and
`forward_stateless` (completion/signatureHelp path)
- Add PCH state cleanup on `didClose` and hash invalidation on `didSave`
## Test plan
- [x] 398 unit tests pass (including 6 new PCH tests: PreambleHash x3,
PCHWorker x2, BuildPCHRequest assertion)
- [x] 5 new integration tests pass (`test_pch.py`: diagnostics on open,
body edit recompile, no-include file, hover with PCH, completion with
PCH)
- [x] 21 existing integration tests pass unchanged
- [x] Build succeeds with 0 errors
🤖 Generated with [Claude Code](https://claude.com/claude-code)
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Precompiled header (PCH) caching to speed compilations and reduce edit
latency
* Automatic attachment of cached PCH to compile requests, improving
hover and completion responsiveness
* Module-aware completions expanded to include available module
artifacts from other files
* **Bug Fixes**
* PCH cache cleared on file close; saving now triggers broader PCH
invalidation to prevent stale PCH use
* **Tests**
* Added unit and integration tests exercising PCH build, reuse, and
editor interactions
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
## Summary
- **CMake-based CDB generation for module tests**: Replace hand-written
compile_commands.json with CMakeLists.txt (CMake 3.28 `FILE_SET
CXX_MODULES`) in all 26 `tests/data/modules/*/` directories. CDB is
generated on-the-fly via `cmake -G Ninja` during test setup.
- **`@pytest.mark.workspace()` decorator**: Introduce a marker + fixture
pattern so tests declare their workspace via decorator and receive a
resolved `workspace` path. The fixture auto-generates CDB when a
CMakeLists.txt is present.
- **`CliceClient` helper methods**: Add `initialize()`, `open()`,
`wait_diagnostics()`, and `open_and_wait()` to reduce boilerplate across
all test files.
- **Use `asyncio_mode = "auto"`**: Switch from `@pytest_asyncio.fixture`
+ `@pytest.mark.asyncio` to `@pytest.fixture` + auto mode for proper
Pylance type inference on fixtures.
- **Test cleanup**: Remove redundant section separators and docstrings,
delete `tests/pyproject.toml` (config moved to `pytest.ini`).
- **Format task**: Add `.cppm` to `format-cpp` glob pattern.
- **CI fix**: Disable `CMAKE_CXX_SCAN_FOR_MODULES` and prefer pixi
clang++ to fix macOS CI where CMake rejects module scanning.
## Test plan
- [x] All 26 module test directories have CMakeLists.txt with FILE_SET
CXX_MODULES
- [x] generate_cdb() produces valid compile_commands.json with module
flags
- [x] Integration tests pass locally
- [ ] CI passes on all platforms (Linux, macOS, Windows)
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Tests**
* Unified fixtures and client workflow: new init/open/wait helpers,
workspace marker support, bounded diagnostics waiting, CMake-based
compilation-database generation, and directory-backed temp-file
workflows; enabled asyncio test mode.
* **Chores**
* Added many C++20 module test projects and test data; removed prior
test pyproject in favor of pytest config; updated formatter to include
.cppm files.
* **Style**
* Reformatted many module/source implementations to consistent
multi-line function bodies.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
