Repository navigation
Finetune to reduce LLM token - #5
Merged
Merged
Conversation
WalkthroughThe change adds deadline-aware resumable graph searches with server-side session storage. MCP sessions now support configurable symbol detail and output formats, normalized paths, omitted default fields, and resumable search responses. Context Markdown rendering supports optional root-prefix removal. ChangesResumable search and session storage
MCP output and path handling
Estimated code review effort: 4 (Complex) | ~60 minutes Sequence Diagram(s)sequenceDiagram
participant Client
participant MCPTools
participant GraphApi
participant SearchSessionStore
participant GraphIndex
Client->>MCPTools: codegraph_search_symbol(query, timeout, resume)
MCPTools->>GraphApi: search_symbol_paged_resumable(...)
GraphApi->>SearchSessionStore: get(resume)
GraphApi->>GraphIndex: search_symbol_paged_resumable(cursor, deadline)
GraphIndex-->>GraphApi: page, progress, continuation
GraphApi->>SearchSessionStore: put or remove session
GraphApi-->>MCPTools: ResumeSearchOutcome
MCPTools-->>Client: formatted response with resume metadata
Possibly related PRs
Suggested reviewers: Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches 💡 1🛠️ Fix failing CI checks 💡
📝 Generate docstrings
🧪 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 |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 8
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
crates/codegraph-mcp/src/usage.rs (1)
120-128: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick winDeduplicate on the normalized path, not the raw path.
The lookup normalizes with
strip_root_prefix, butseenstores the raw answer pathp. If the same file appears once as an absolute path and once already relativized, both forms map to the same entry and its bytes are counted twice.🐛 Proposed fix
let mut seen = std::collections::HashSet::new(); let mut total = 0u64; for p in paths { - if let Some(b) = bytes_by_path.get(crate::tools::strip_root_prefix(&p, root)) { - if seen.insert(p) { + let key = crate::tools::strip_root_prefix(&p, root).to_string(); + if let Some(b) = bytes_by_path.get(key.as_str()) { + if seen.insert(key) { total += b; } } }🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-mcp/src/usage.rs` around lines 120 - 128, Update the deduplication in the path-totaling loop to use the normalized path produced by strip_root_prefix as the HashSet key, while using that same normalized value for bytes_by_path lookup. Preserve counting each file only once when absolute and relative representations resolve to the same path.crates/codegraph-mcp/src/session.rs (1)
184-213: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick winA re-run of
codegraph_initwithoutdetailresets the session detail level.Session::initacceptsdetailas a requiredDetailLeveland always writes it, while it acceptsformatas anOption<OutputStyle>and writes it only when present. The caller converts an absentdetailargument intoDetailLevel::default(), so the two settings behave differently for the same "argument omitted" case.
crates/codegraph-mcp/src/session.rs#L184-L213: change the parameter todetail: Option<DetailLevel>and writeself.detailonly when it isSome, matching the existingformathandling.crates/codegraph-mcp/src/lib.rs#L131-L153: drop.unwrap_or_default()so the parsed value staysOption<DetailLevel>, pass it through toinit, and reportself.session.detail().await.as_str()in the response instead of the localdetail.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-mcp/src/session.rs` around lines 184 - 213, Update Session::init in crates/codegraph-mcp/src/session.rs:184-213 to accept detail: Option<DetailLevel> and only write self.detail when it is Some, matching format handling. Update crates/codegraph-mcp/src/lib.rs:131-153 to remove unwrap_or_default(), pass the optional detail through to init, and report self.session.detail().await.as_str() rather than the local value.
🧹 Nitpick comments (5)
crates/codegraph-context/src/lib.rs (1)
146-146: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winAdd regression coverage for prefix stripping.
The supplied
crates/codegraph-api/tests/api.rstest passesstrip_prefix: None. It checks only the unchanged behavior. Add a Markdown test withSome(root)and assert relative paths for the primary symbol, callers, and callees. Include a boundary case such as/repo-old/src/a.tswith/repo.Also applies to: 158-176
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-context/src/lib.rs` at line 146, Add regression coverage for prefix stripping in the Markdown/API test flow by using strip_prefix: Some(root). Assert that relative paths are correct for the primary symbol, callers, and callees, and add a boundary case such as /repo-old/src/a.ts with /repo to ensure only a complete path prefix is stripped.crates/codegraph-graph/src/lib.rs (3)
1759-1793: 🚀 Performance & Scalability | 🔵 Trivial | ⚡ Quick win
ExactandPrefixscan every distinct name.The
ScanNamesphase walkssorted_name_keyselement by element for all non-Containsmodes.Exactcan resolve throughname_indexin constant time.Prefixcan usepartition_pointon the sorted vector and stop at the first non-matching key. OnlySuffixneeds a full scan.The current form makes an exact-name lookup O(number of distinct names) per call, and it also creates unnecessary resume round-trips on large indexes.
♻️ Suggested fast paths
SearchCursorPhase::ScanNames { name_pos } => { + // Exact: hash lookup — không cần quét. + if mode == SymbolMatch::Exact { + let matched = if self.name_index.contains_key(&q) { + vec![q.clone()] + } else { + Vec::new() + }; + phase = SearchCursorPhase::Expand { + names: matched, + name_idx: 0, + id_idx: 0, + collected: Vec::new(), + }; + } else { let mut matched: Vec<String> = Vec::new(); let mut pos = *name_pos;For
Prefix, startposatself.sorted_name_keys.partition_point(|n| n.as_str() < q.as_str())and stop as soon as a key no longer starts withq.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-graph/src/lib.rs` around lines 1759 - 1793, Optimize the ScanNames branch in the search flow: resolve SymbolMatch::Exact directly through name_index instead of scanning sorted_name_keys, and for SymbolMatch::Prefix initialize pos with partition_point and stop when keys no longer start with q. Preserve the existing resumable scan behavior for timeouts and retain the full scan only for SymbolMatch::Suffix.
141-143: 🚀 Performance & Scalability | 🔵 Trivial | ⚡ Quick win
sorted_name_keysduplicatesname_records.
rebuild_name_enginebuildssorted_name_keysfromdistinctand then pushes the same strings, in the same order, intoname_records. Both vectors hold identical content after the loop. On a large index this doubles the memory used for distinct names.The scan phase in
search_symbol_paged_resumablecan indexname_recordsdirectly, because it is already sorted by construction.♻️ Suggested consolidation
- let mut record = 0usize; - self.sorted_name_keys = distinct.iter().map(|s| s.to_string()).collect(); + let mut record = 0usize;Then replace
self.sorted_name_keysreads insearch_symbol_paged_resumablewithself.name_records, and drop the field plus its reset at line 650.Also applies to: 597-597
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-graph/src/lib.rs` around lines 141 - 143, Remove the redundant sorted_name_keys field and its reset, since rebuild_name_engine already populates name_records with the sorted distinct names. Update search_symbol_paged_resumable to use name_records for all existing sorted-name scan and indexing reads, preserving the current ordering and pagination behavior.
1798-1839: 🚀 Performance & Scalability | 🔵 Trivial | ⚡ Quick winCheck the deadline in batches, not per symbol id.
The phase B loop calls
Instant::now()for every single id it appends. On a large expansion this clock read dominates the useful work. A counter that checks the deadline every N iterations (for example 1024) keeps cancellation responsive and removes most clock reads.♻️ Suggested batching
loop { - if let Some(dl) = deadline - && Instant::now() >= dl - { - timed_out = true; - break; - } + steps += 1; + if steps % 1024 == 0 + && let Some(dl) = deadline + && Instant::now() >= dl + { + timed_out = true; + break; + }Declare
let mut steps = 0usize;before the loop. Apply the same pattern to theScanNamesloop.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-graph/src/lib.rs` around lines 1798 - 1839, Update the Phase B expansion loop in the SearchCursorPhase::Expand branch to track a steps counter and check the deadline only once per batch, such as every 1024 iterations, while preserving timeout behavior. Apply the same batched deadline-check pattern to the ScanNames loop, using a counter initialized before each loop and incremented for processed entries.crates/codegraph-api/tests/api.rs (1)
262-316: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winAdd coverage for repeated paging through the session store.
This test performs three calls, so the store never reaches its 512-entry cap. Each successful call with remaining pages stores a new session entry (see
crates/codegraph-api/src/lib.rslines 261-271). A longer paging loop would expose the eviction of the still-usedresume_id.Add a test that pages through the full result set with the same
resume_idand asserts that the id stays valid until completion.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-api/tests/api.rs` around lines 262 - 316, Extend search_symbol_paged_resume_paging into a loop that requests every page using the original resume_id, advancing the offset by the page size and asserting each call succeeds with the expected page size until the final page. Verify the same resume_id remains valid throughout paging and that the final response has no resume token, covering session-store eviction beyond 512 entries.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@crates/codegraph-api/src/lib.rs`:
- Around line 90-94: Update CodegraphCursorStore::get to enforce the configured
TTL before returning an entry, removing expired cursors and returning None when
their age exceeds the TTL; preserve the existing tuple and clone behavior for
valid entries. Keep the documentation consistent with this behavior.
- Around line 261-271: Update the session handling around resume_id so continued
pages refresh the existing session identified by resume in place when it is
valid, rather than always calling sessions.put to create a new id. Preserve
creation of a new session when no resume id was supplied, and continue removing
the session when out.cursor is None.
In `@crates/codegraph-context/src/lib.rs`:
- Around line 121-133: Update rel_path to use camino::Utf8Path::strip_prefix for
platform-aware prefix matching, handling both slash styles and trailing
separators while returning the original p when stripping fails. Add focused
tests covering trailing `/`, Windows `\` separators, and unmatched prefixes.
In `@crates/codegraph-graph/src/lib.rs`:
- Around line 1729-1758: Update the search flow around codegraph_search_symbol
and the SearchCursorPhase::Engine branch to detect an empty query when the
search mode is Contains, after validating the resume state, and return an empty
PagedSearchOutcome without invoking search_resumable. Preserve normal engine
behavior for non-empty queries and other search modes; alternatively, map the
resulting NotFound error to an empty result consistently with the legacy helper.
In `@crates/codegraph-graph/src/radix.rs`:
- Around line 837-855: Update the DfsState::Collect handling in
Radix::search_dfs so each node’s record is appended only on its initial stack
visit, while subsequent child-index resumes process children without rereading
or pushing the record. Preserve pre-order traversal, MAX_RESULTS accounting, and
checkpoint contents without duplicate records.
In `@crates/codegraph-graph/src/search.rs`:
- Around line 572-613: Fix the depth-filter resume flow in the search resolve
block so resume state preserves the original unfiltered record_ids while also
retaining the already-filtered output. Add a dedicated resolved-results field to
SearchResume and restore/use it across retries, or restart filtering from
resolve_idx zero; ensure resumed searches do not discard previously accepted
records or report premature completion.
In `@crates/codegraph-mcp/src/server-instructions.md`:
- Around line 85-89: Update the fenced code block in the instructions text to
declare the text language, using a text fence for the plain error message and
preserving its contents unchanged.
In `@crates/codegraph-mcp/src/tools.rs`:
- Around line 658-693: Normalize the not-found response in the
codegraph_function_scope handler so the "function" key has the same shape in
both branches. In the None branch, emit the resolved symbol through symbol_json
using the existing detail and format values, or use a distinct key for the plain
name instead; prefer preserving "function" with symbol_json.
---
Outside diff comments:
In `@crates/codegraph-mcp/src/session.rs`:
- Around line 184-213: Update Session::init in
crates/codegraph-mcp/src/session.rs:184-213 to accept detail:
Option<DetailLevel> and only write self.detail when it is Some, matching format
handling. Update crates/codegraph-mcp/src/lib.rs:131-153 to remove
unwrap_or_default(), pass the optional detail through to init, and report
self.session.detail().await.as_str() rather than the local value.
In `@crates/codegraph-mcp/src/usage.rs`:
- Around line 120-128: Update the deduplication in the path-totaling loop to use
the normalized path produced by strip_root_prefix as the HashSet key, while
using that same normalized value for bytes_by_path lookup. Preserve counting
each file only once when absolute and relative representations resolve to the
same path.
---
Nitpick comments:
In `@crates/codegraph-api/tests/api.rs`:
- Around line 262-316: Extend search_symbol_paged_resume_paging into a loop that
requests every page using the original resume_id, advancing the offset by the
page size and asserting each call succeeds with the expected page size until the
final page. Verify the same resume_id remains valid throughout paging and that
the final response has no resume token, covering session-store eviction beyond
512 entries.
In `@crates/codegraph-context/src/lib.rs`:
- Line 146: Add regression coverage for prefix stripping in the Markdown/API
test flow by using strip_prefix: Some(root). Assert that relative paths are
correct for the primary symbol, callers, and callees, and add a boundary case
such as /repo-old/src/a.ts with /repo to ensure only a complete path prefix is
stripped.
In `@crates/codegraph-graph/src/lib.rs`:
- Around line 1759-1793: Optimize the ScanNames branch in the search flow:
resolve SymbolMatch::Exact directly through name_index instead of scanning
sorted_name_keys, and for SymbolMatch::Prefix initialize pos with
partition_point and stop when keys no longer start with q. Preserve the existing
resumable scan behavior for timeouts and retain the full scan only for
SymbolMatch::Suffix.
- Around line 141-143: Remove the redundant sorted_name_keys field and its
reset, since rebuild_name_engine already populates name_records with the sorted
distinct names. Update search_symbol_paged_resumable to use name_records for all
existing sorted-name scan and indexing reads, preserving the current ordering
and pagination behavior.
- Around line 1798-1839: Update the Phase B expansion loop in the
SearchCursorPhase::Expand branch to track a steps counter and check the deadline
only once per batch, such as every 1024 iterations, while preserving timeout
behavior. Apply the same batched deadline-check pattern to the ScanNames loop,
using a counter initialized before each loop and incremented for processed
entries.
🪄 Autofix
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: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: b81d5972-e9f1-43c3-8d25-031d64a0efa4
📒 Files selected for processing (16)
Cargo.tomlcrates/codegraph-api/src/lib.rscrates/codegraph-api/tests/api.rscrates/codegraph-context/src/lib.rscrates/codegraph-core/src/semgraph.rscrates/codegraph-graph/src/lib.rscrates/codegraph-graph/src/radix.rscrates/codegraph-graph/src/search.rscrates/codegraph-graph/src/shared.rscrates/codegraph-mcp/src/lib.rscrates/codegraph-mcp/src/server-instructions.mdcrates/codegraph-mcp/src/session.rscrates/codegraph-mcp/src/stdio.rscrates/codegraph-mcp/src/tools.rscrates/codegraph-mcp/src/usage.rscrates/codegraph/src/main.rs
Comment on lines
+90
to
+94
| /// Đọc cursor theo id — `None` nếu không có / quá TTL. | ||
| pub fn get(&self, id: &str) -> Option<(u64, SearchCursor)> { | ||
| let map = self.inner.lock().unwrap(); | ||
| map.get(id).map(|s| (s.index_version, s.cursor.clone())) | ||
| } |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
get does not apply the TTL.
The doc comment states that get returns None when the entry is older than the TTL. The body only performs a map lookup. Expiry happens only inside put. If no put runs, an expired cursor stays reachable.
Either enforce the TTL in get, or correct the doc comment.
🔧 Proposed fix
/// Đọc cursor theo id — `None` nếu không có / quá TTL.
pub fn get(&self, id: &str) -> Option<(u64, SearchCursor)> {
let map = self.inner.lock().unwrap();
- map.get(id).map(|s| (s.index_version, s.cursor.clone()))
+ let s = map.get(id)?;
+ if Instant::now().duration_since(s.created) >= self.ttl {
+ return None;
+ }
+ Some((s.index_version, s.cursor.clone()))
}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| /// Đọc cursor theo id — `None` nếu không có / quá TTL. | |
| pub fn get(&self, id: &str) -> Option<(u64, SearchCursor)> { | |
| let map = self.inner.lock().unwrap(); | |
| map.get(id).map(|s| (s.index_version, s.cursor.clone())) | |
| } | |
| /// Đọc cursor theo id — `None` nếu không có / quá TTL. | |
| pub fn get(&self, id: &str) -> Option<(u64, SearchCursor)> { | |
| let map = self.inner.lock().unwrap(); | |
| let s = map.get(id)?; | |
| if Instant::now().duration_since(s.created) >= self.ttl { | |
| return None; | |
| } | |
| Some((s.index_version, s.cursor.clone())) | |
| } |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@crates/codegraph-api/src/lib.rs` around lines 90 - 94, Update
CodegraphCursorStore::get to enforce the configured TTL before returning an
entry, removing expired cursors and returning None when their age exceeds the
TTL; preserve the existing tuple and clone behavior for valid entries. Keep the
documentation consistent with this behavior.
Comment on lines
+261
to
+271
| // ── Quản lý session: lưu khi còn tiếp tục (timeout / còn page), xoá | ||
| // khi xong hẳn. ── | ||
| let resume_id = match &out.cursor { | ||
| Some(c) => Some(self.sessions.put(c.clone(), version)), | ||
| None => { | ||
| if let Some(id) = &resume { | ||
| self.sessions.remove(id); | ||
| } | ||
| None | ||
| } | ||
| }; |
There was a problem hiding this comment.
🩺 Stability & Availability | 🟠 Major | ⚡ Quick win
Each continued page stores a new session and never releases the previous one.
When out.cursor is Some, this code always calls put and returns a fresh id. The id supplied in resume is removed only when the search completes. The integration test crates/codegraph-api/tests/api.rs (lines 283-316) reuses the same resume_id across pages, so the caller keeps using the original id while the store fills with unused clones.
Two consequences:
- Every paged call consumes one slot of the 512-slot store.
- Eviction removes the oldest entry first. The original id that the caller still uses is the oldest, so it can be evicted by its own clones. The next retry then fails with "resume id expired or unknown".
Refresh the existing session in place instead of creating a new one when the caller passed a valid resume.
🔧 Proposed direction
let resume_id = match &out.cursor {
- Some(c) => Some(self.sessions.put(c.clone(), version)),
+ // Cùng một search đang tiếp tục → cập nhật session cũ, giữ nguyên id.
+ Some(c) => match &resume {
+ Some(id) => {
+ self.sessions.update(id, c.clone(), version);
+ Some(id.clone())
+ }
+ None => Some(self.sessions.put(c.clone(), version)),
+ },
None => {
if let Some(id) = &resume {
self.sessions.remove(id);
}
None
}
};🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@crates/codegraph-api/src/lib.rs` around lines 261 - 271, Update the session
handling around resume_id so continued pages refresh the existing session
identified by resume in place when it is valid, rather than always calling
sessions.put to create a new id. Preserve creation of a new session when no
resume id was supplied, and continue removing the session when out.cursor is
None.
Comment on lines
+121
to
+133
| /// Strip `root/` prefix khỏi path (boundary-aware) — `None` giữ nguyên. | ||
| fn rel_path<'a>(p: &'a str, strip: Option<&str>) -> &'a str { | ||
| if let Some(root) = strip { | ||
| if let Some(rest) = p.strip_prefix(root) { | ||
| if let Some(rest) = rest.strip_prefix('/') { | ||
| return rest; | ||
| } | ||
| } | ||
| } | ||
| p | ||
| } | ||
|
|
||
| fn render_markdown(resp: &ContextResponse, strip: Option<&str>) -> String { |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '--- candidate files ---'
git ls-files | rg '(^|/)(crates/codegraph-context|.*context.*|.*path.*|.*test.*)$' | head -200
printf '%s\n' '--- rel_path and related symbols ---'
rg -n -C 8 'fn rel_path|render_markdown|strip_prefix|ContextRequest|ContextResponse' crates/codegraph-context 2>/dev/null || true
printf '%s\n' '--- file outline ---'
if command -v ast-grep >/dev/null 2>&1; then
ast-grep outline crates/codegraph-context/src/lib.rs
else
wc -l crates/codegraph-context/src/lib.rs
fiRepository: hungpham10/codegraph-rs
Length of output: 9201
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '--- rendering implementation ---'
sed -n '133,240p' crates/codegraph-context/src/lib.rs
printf '%s\n' '--- context request usages ---'
rg -n -C 5 'ContextRequest\s*\{|strip_prefix|file:' crates --glob '*.rs' --glob '*.md' --glob '*.toml' | head -300
printf '%s\n' '--- path conventions and targets ---'
rg -n -C 3 'target|cfg\(target|PathBuf|std::path::Path|replace\(.*\\\\|replace\(.*\/|canonical|workspace root|workspace_root' Cargo.toml crates docs --glob '*.rs' --glob '*.toml' --glob '*.md' | head -300
printf '%s\n' '--- package metadata ---'
cat crates/codegraph-context/Cargo.tomlRepository: hungpham10/codegraph-rs
Length of output: 44828
🏁 Script executed:
#!/bin/bash
set -eu
python3 - <<'PY'
from pathlib import PurePosixPath, PureWindowsPath
def current(p, root):
rest = p[len(root):] if p.startswith(root) else None
if rest is not None and rest.startswith('/'):
return rest[1:]
return p
cases = [
("/workspace/src/a.ts", "/workspace/"),
("/workspace/src/a.ts", "/workspace"),
("/src/a.ts", "/"),
(r"C:\\workspace\\src\\a.ts", r"C:\\workspace\\"),
(r"C:\\workspace\\src\\a.ts", r"C:\\workspace"),
]
for p, root in cases:
print(f"current({p!r}, {root!r}) -> {current(p, root)!r}")
print(f"posix.relative_to -> ", end="")
try:
print(repr(str(PurePosixPath(p).relative_to(PurePosixPath(root)))))
except ValueError as e:
print(type(e).__name__)
print(f"windows.relative_to -> ", end="")
try:
print(repr(str(PureWindowsPath(p).relative_to(PureWindowsPath(root)))))
except ValueError as e:
print(type(e).__name__)
PYRepository: hungpham10/codegraph-rs
Length of output: 865
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '--- MCP root state and initialization ---'
rg -n -C 8 'struct .*root|root:|Utf8PathBuf|codegraph_context|strip_root_prefix' crates/codegraph-mcp/src --glob '*.rs' | head -400
printf '%s\n' '--- extraction path flow ---'
sed -n '1,120p' crates/codegraph-extract/src/walker.rs
sed -n '1,120p' crates/codegraph-extract/src/orchestrator.rs
rg -n -C 5 'FileMatch|\.path|parse_file|parse_project|to_string\(\)|as_str\(\)' crates/codegraph-extract/src --glob '*.rs' | head -400
printf '%s\n' '--- workspace camino dependency ---'
rg -n -C 3 'camino' Cargo.toml crates --glob 'Cargo.toml'Repository: hungpham10/codegraph-rs
Length of output: 50379
🏁 Script executed:
#!/bin/bash
set -eu
if command -v rustc >/dev/null 2>&1; then
cat >/tmp/rel_path_probe.rs <<'RS'
use std::path::Path;
fn main() {
let cases = [
("/workspace/src/a.ts", "/workspace/"),
("/workspace/src/a.ts", "/workspace"),
("/src/a.ts", "/"),
("C:\\workspace\\src\\a.ts", "C:\\workspace\\"),
("C:\\workspace\\src\\a.ts", "C:\\workspace"),
];
for (p, root) in cases {
let result = Path::new(p)
.strip_prefix(Path::new(root))
.and_then(|rest| rest.to_str())
.map(str::to_owned);
println!("{p:?} / {root:?} => {result:?}");
}
}
RS
rustc /tmp/rel_path_probe.rs -o /tmp/rel_path_probe
/tmp/rel_path_probe
else
printf '%s\n' 'rustc unavailable; exact Rust probe skipped'
fiRepository: hungpham10/codegraph-rs
Length of output: 752
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '--- focused root initialization ---'
rg -n -C 12 'pub async fn init|fn init|with_root_and_format|Session::with_root|root =|root:' crates/codegraph-mcp/src/session.rs crates/codegraph-mcp/src/lib.rs
printf '%s\n' '--- focused parse path handoff ---'
rg -n -C 10 'fn parse_one|parse_file\(' crates/codegraph-extract/src/orchestrator.rs crates/codegraph-extract/src/lib.rs
printf '%s\n' '--- context tests and workspace test declarations ---'
rg -n -C 5 'mod tests|#\[test\]|render_markdown|rel_path' crates/codegraph-context/src/lib.rs crates/codegraph-context/tests 2>/dev/null || true
printf '%s\n' '--- root normalization calls ---'
rg -n -C 4 'canonicalize|normalize|clean|trim_end_matches|Utf8PathBuf::from\(path\)|init\(' crates/codegraph-mcp/src/session.rs crates/codegraph-mcp/src/lib.rs crates/codegraph-mcp/src/main.rs 2>/dev/null || trueRepository: hungpham10/codegraph-rs
Length of output: 36694
🏁 Script executed:
#!/bin/bash
set -eu
if command -v rustc >/dev/null 2>&1; then
cat >/tmp/rel_path_probe.rs <<'RS'
use std::path::Path;
fn main() {
let cases = [
("/workspace/src/a.ts", "/workspace/"),
("/workspace/src/a.ts", "/workspace"),
("/src/a.ts", "/"),
("C:\\workspace\\src\\a.ts", "C:\\workspace\\"),
("C:\\workspace\\src\\a.ts", "C:\\workspace"),
];
for (p, root) in cases {
let result = Path::new(p)
.strip_prefix(Path::new(root))
.ok()
.and_then(|rest| rest.to_str())
.map(str::to_owned);
println!("{p:?} / {root:?} => {result:?}");
}
}
RS
rustc /tmp/rel_path_probe.rs -o /tmp/rel_path_probe
/tmp/rel_path_probe
else
printf '%s\n' 'rustc unavailable; exact Rust probe skipped'
fiRepository: hungpham10/codegraph-rs
Length of output: 422
Use platform-aware path prefix matching.
rel_path removes only /. It fails for a trailing separator, /, and Windows \ separators. Use camino::Utf8Path::strip_prefix and preserve p when stripping fails. Add tests for these cases.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@crates/codegraph-context/src/lib.rs` around lines 121 - 133, Update rel_path
to use camino::Utf8Path::strip_prefix for platform-aware prefix matching,
handling both slash styles and trailing separators while returning the original
p when stripping fails. Add focused tests covering trailing `/`, Windows `\`
separators, and unmatched prefixes.
Comment on lines
+1729
to
1758
| match &mut phase { | ||
| SearchCursorPhase::Engine(sr) => { | ||
| let page = self | ||
| .names | ||
| .search_resumable(q.as_bytes(), None, Some(sr.clone()), deadline) | ||
| .await?; | ||
| if page.timed_out { | ||
| phase = SearchCursorPhase::Engine(page.resume.unwrap_or_default()); | ||
| timed_out = true; | ||
| } else { | ||
| // record → tên, sort → Expand. | ||
| let mut names: Vec<String> = page | ||
| .record_ids | ||
| .iter() | ||
| .filter_map(|&r| { | ||
| if r == 0 { | ||
| return None; | ||
| } | ||
| self.name_records.get(r - 1).cloned() | ||
| }) | ||
| .collect(); | ||
| names.sort(); | ||
| phase = SearchCursorPhase::Expand { | ||
| names, | ||
| name_idx: 0, | ||
| id_idx: 0, | ||
| collected: Vec::new(), | ||
| }; | ||
| for &id in name_ids { | ||
| if seen.insert(id) { | ||
| ids.push(id); | ||
| } | ||
| } | ||
| } | ||
| } |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Find callers of the paged/resumable search and check query validation upstream.
rg -nP -C4 '\bsearch_symbol_paged(_resumable)?\s*\(' --type=rust
rg -nP -C4 '\bsearch_resumable\s*\(' --type=rust crates/codegraph-api crates/codegraph-mcpRepository: hungpham10/codegraph-rs
Length of output: 161
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '== candidate files =='
fd -t f 'lib.rs|.*\.rs$' crates/codegraph-graph crates/codegraph-api crates/codegraph-mcp 2>/dev/null | head -200
printf '%s\n' '== paged search references =='
rg -n -C5 'search_symbol_paged|search_symbol_filtered|search_resumable|struct SearchCursor|PagedSearchOutcome' crates --glob '*.rs' || true
printf '%s\n' '== target outline =='
ast-grep outline crates/codegraph-graph/src/lib.rs --match 'search_symbol_paged_resumable' --view expanded 2>/dev/null || trueRepository: hungpham10/codegraph-rs
Length of output: 39514
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '== generic search implementation =='
sed -n '430,590p' crates/codegraph-graph/src/search.rs
printf '%s\n' '== old helper =='
sed -n '1128,1170p' crates/codegraph-graph/src/lib.rs
printf '%s\n' '== paged implementation setup/finalization =='
sed -n '1660,1900p' crates/codegraph-graph/src/lib.rs
printf '%s\n' '== API forwarding =='
sed -n '150,270p' crates/codegraph-api/src/lib.rs
printf '%s\n' '== MCP query extraction =='
sed -n '285,340p' crates/codegraph-mcp/src/tools.rs
sed -n '465,515p' crates/codegraph-mcp/src/tools.rsRepository: hungpham10/codegraph-rs
Length of output: 26800
🏁 Script executed:
#!/bin/bash
set -eu
python3 - <<'PY'
from pathlib import Path
files = [
Path("crates/codegraph-graph/src/search.rs"),
Path("crates/codegraph-graph/src/lib.rs"),
Path("crates/codegraph-api/src/lib.rs"),
Path("crates/codegraph-mcp/src/tools.rs"),
]
terms = ("is_empty", "trim", "query", "search_symbol_paged_resumable", "search_resumable")
for path in files:
text = path.read_text()
print(f"== {path} ==")
for i, line in enumerate(text.splitlines(), 1):
if any(term in line for term in terms):
print(f"{i}: {line}")
PYRepository: hungpham10/codegraph-rs
Length of output: 8671
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '== argument parser =='
sed -n '930,975p' crates/codegraph-mcp/src/tools.rs
printf '%s\n' '== match-mode parsing =='
rg -n -C8 'impl.*SymbolMatch|enum SymbolMatch|SymbolMatch::parse|fn parse' crates/codegraph-graph/src/lib.rs crates/codegraph-api/src/lib.rs crates/codegraph-mcp/src/tools.rs
printf '%s\n' '== focused behavioral verifier =='
python3 - <<'PY'
from pathlib import Path
search = Path("crates/codegraph-graph/src/search.rs").read_text()
graph = Path("crates/codegraph-graph/src/lib.rs").read_text()
mcp = Path("crates/codegraph-mcp/src/tools.rs").read_text()
api = Path("crates/codegraph-api/src/lib.rs").read_text()
assert "if pattern.is_empty() {\n return Err(Error::NotFound);\n }" in search
assert "let page = self.search_resumable(q.as_bytes(), None, Some(sr.clone()), deadline)" in graph
assert "let q = query.to_lowercase();" in graph
paged = graph[graph.index("pub async fn search_symbol_paged_resumable"):graph.index("/// Resumable + deadline-aware", graph.index("pub async fn search_symbol_paged_resumable"))]
assert "if q.is_empty()" not in paged
old = graph[graph.index("async fn search_symbol_filtered"):graph.index("/// Search symbol nâng cao", graph.index("async fn search_symbol_filtered"))]
assert "Err(_) => return Ok(Vec::new())" in old
assert "let q = arg_str(&args, \"query\")?;" in mcp
assert ".search_symbol_paged_resumable(q, kind, mode" in mcp
assert "SymbolMatch::Contains" in api and ".search_symbol_paged_resumable(" in api
print("Contains + empty query: Err(Error::NotFound)")
print("Prefix + empty query: scan path; starts_with(\"\") is true")
print("Suffix + empty query: scan path; ends_with(\"\") is true")
print("Exact + empty query: scan path; only name == \"\" can match")
print("Legacy helper: engine errors map to Ok(Vec::new())")
print("MCP/API callers: no empty-query validation found in the forwarding paths")
PYRepository: hungpham10/codegraph-rs
Length of output: 5198
🏁 Script executed:
#!/bin/bash
set -u
printf '%s\n' '== arg_str definition and calls =='
rg -n -C8 'fn arg_str|arg_str\s*\(' crates/codegraph-mcp/src/tools.rs || true
printf '%s\n' '== exact generic call lines =='
nl -ba crates/codegraph-graph/src/lib.rs | sed -n '1728,1736p'
printf '%s\n' '== exact source assertions =='
python3 - <<'PY'
from pathlib import Path
import re
search = Path("crates/codegraph-graph/src/search.rs").read_text()
graph = Path("crates/codegraph-graph/src/lib.rs").read_text()
mcp = Path("crates/codegraph-mcp/src/tools.rs").read_text()
api = Path("crates/codegraph-api/src/lib.rs").read_text()
checks = [
("generic empty pattern returns NotFound",
"pattern.is_empty()" in search and "return Err(Error::NotFound);" in search[search.index("pub async fn search_resumable"):]),
("paged calls generic search with q bytes",
re.search(r"\.search_resumable\(\s*q\.as_bytes\(\)", graph) is not None),
("paged has no empty guard",
"if q.is_empty()" not in graph[graph.index("pub async fn search_symbol_paged_resumable"):graph.index("/// Resumable + deadline-aware", graph.index("pub async fn search_symbol_paged_resumable"))]),
("legacy helper maps engine errors to empty",
"Err(_) => return Ok(Vec::new())" in graph[graph.index("async fn search_symbol_filtered"):]),
("MCP extracts query directly",
'let q = arg_str(&args, "query")?;' in mcp),
("MCP forwards paged query",
".search_symbol_paged_resumable(q, kind, mode" in mcp),
("API defaults resumable search to Contains",
"SymbolMatch::Contains" in api[api.index("pub async fn search_resumable"):api.index("pub async fn search_symbol_paged"):]),
]
for name, result in checks:
print(f"{'PASS' if result else 'FAIL'}: {name}")
assert result, name
PYRepository: hungpham10/codegraph-rs
Length of output: 8647
Handle empty Contains queries before calling the name engine.
arg_str accepts "", and codegraph_search_symbol defaults to Contains. The empty byte pattern reaches Search::search_resumable at line 1734, which returns Error::NotFound; the legacy helper maps this error to an empty result. Return an empty PagedSearchOutcome for empty Contains queries after resume validation, or map this error to an empty result.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@crates/codegraph-graph/src/lib.rs` around lines 1729 - 1758, Update the
search flow around codegraph_search_symbol and the SearchCursorPhase::Engine
branch to detect an empty query when the search mode is Contains, after
validating the resume state, and return an empty PagedSearchOutcome without
invoking search_resumable. Preserve normal engine behavior for non-empty queries
and other search modes; alternatively, map the resulting NotFound error to an
empty result consistently with the legacy helper.
Comment on lines
+837
to
855
| DfsState::Collect { root, mut stack } => { | ||
| // Collect subtree theo pre-order (record của node trước, sau | ||
| // đó mới children — giống bản đệ quy cũ). | ||
| if let Some((node_id, child_idx)) = stack.pop() { | ||
| let (_prefix_bytes, record) = | ||
| { self.storage.read().await.get_node(node_id).await? }; | ||
| if record != EMPTY { | ||
| records.push(record); | ||
| } | ||
| let children = { self.storage.read().await.get_children(node_id).await? }; | ||
| if child_idx < children.len() { | ||
| stack.push((node_id, child_idx + 1)); | ||
| stack.push((children[child_idx], 0)); | ||
| } | ||
| Some(DfsState::Collect { root, stack }) | ||
| } else { | ||
| None // Collect xong — candidate đã có records, dừng. | ||
| } | ||
| } |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Collect pushes a node record once per child visit.
Each stack entry (node_id, child_idx) is re-pushed as (node_id, child_idx + 1) before descending into a child. When that entry pops again, the code re-reads the node and pushes record a second time. A node with k children therefore contributes its record k + 1 times (k times when child_idx < children.len(), plus the final pop).
The recursive implementation pushed the record exactly once. Search::search_resumable hides the effect because it deduplicates through seen, but Radix::search_dfs returns duplicates directly, MAX_RESULTS accounting inflates, and each duplicate is carried inside the checkpoint.
Push the record only on the first visit to the node.
🐛 Proposed fix
DfsState::Collect { root, mut stack } => {
// Collect subtree theo pre-order (record của node trước, sau
// đó mới children — giống bản đệ quy cũ).
if let Some((node_id, child_idx)) = stack.pop() {
let (_prefix_bytes, record) =
{ self.storage.read().await.get_node(node_id).await? };
- if record != EMPTY {
+ // Chỉ lần thăm đầu tiên (child_idx == 0) mới ghi record —
+ // các lần pop sau chỉ để duyệt tiếp children.
+ if child_idx == 0 && record != EMPTY {
records.push(record);
}📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| DfsState::Collect { root, mut stack } => { | |
| // Collect subtree theo pre-order (record của node trước, sau | |
| // đó mới children — giống bản đệ quy cũ). | |
| if let Some((node_id, child_idx)) = stack.pop() { | |
| let (_prefix_bytes, record) = | |
| { self.storage.read().await.get_node(node_id).await? }; | |
| if record != EMPTY { | |
| records.push(record); | |
| } | |
| let children = { self.storage.read().await.get_children(node_id).await? }; | |
| if child_idx < children.len() { | |
| stack.push((node_id, child_idx + 1)); | |
| stack.push((children[child_idx], 0)); | |
| } | |
| Some(DfsState::Collect { root, stack }) | |
| } else { | |
| None // Collect xong — candidate đã có records, dừng. | |
| } | |
| } | |
| DfsState::Collect { root, mut stack } => { | |
| // Collect subtree theo pre-order (record của node trước, sau | |
| // đó mới children — giống bản đệ quy cũ). | |
| if let Some((node_id, child_idx)) = stack.pop() { | |
| let (_prefix_bytes, record) = | |
| { self.storage.read().await.get_node(node_id).await? }; | |
| // Chỉ lần thăm đầu tiên (child_idx == 0) mới ghi record — | |
| // các lần pop sau chỉ để duyệt tiếp children. | |
| if child_idx == 0 && record != EMPTY { | |
| records.push(record); | |
| } | |
| let children = { self.storage.read().await.get_children(node_id).await? }; | |
| if child_idx < children.len() { | |
| stack.push((node_id, child_idx + 1)); | |
| stack.push((children[child_idx], 0)); | |
| } | |
| Some(DfsState::Collect { root, stack }) | |
| } else { | |
| None // Collect xong — candidate đã có records, dừng. | |
| } | |
| } |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@crates/codegraph-graph/src/radix.rs` around lines 837 - 855, Update the
DfsState::Collect handling in Radix::search_dfs so each node’s record is
appended only on its initial stack visit, while subsequent child-index resumes
process children without rereading or pushing the record. Preserve pre-order
traversal, MAX_RESULTS accounting, and checkpoint contents without duplicate
records.
Comment on lines
+572
to
+613
| // ── Resolve: filter `depth` (key length trong storage) — deadline-aware. | ||
| // Bỏ qua nếu không giới hạn depth (name engine — chiếm đa số query). | ||
| if let Some(m) = max_len { | ||
| let mut out = Vec::new(); | ||
| let mut ridx = resolve_idx; | ||
| let storage = self.storage.read().await; | ||
| for &rid in &record_ids { | ||
| loop { | ||
| if let Some(dl) = deadline | ||
| && Instant::now() >= dl | ||
| { | ||
| return Ok(SearchPage { | ||
| record_ids: out.clone(), | ||
| resume: Some(SearchResume { | ||
| cand_idx: candidates.len(), | ||
| dfs: None, | ||
| record_ids: out.clone(), | ||
| resolve_idx: ridx, | ||
| }), | ||
| timed_out: true, | ||
| }); | ||
| } | ||
| if ridx >= record_ids.len() { | ||
| break; | ||
| } | ||
| let rid = record_ids[ridx]; | ||
| ridx += 1; | ||
| if rid == EMPTY { | ||
| continue; | ||
| } | ||
| if let Some(m) = max_len | ||
| && storage.get_key_len(rid).await?.unwrap_or(usize::MAX) > m | ||
| { | ||
| if storage.get_key_len(rid).await?.unwrap_or(usize::MAX) > m { | ||
| continue; | ||
| } | ||
| let meta = storage.get_meta(rid).await?; | ||
| results.push((rid, meta)); | ||
| if results.len() >= MAX_RESULTS { | ||
| break; | ||
| } | ||
| out.push(rid); | ||
| } | ||
| record_ids = out; | ||
| } | ||
|
|
||
| if results.is_empty() { | ||
| Err(Error::NotFound) | ||
| } else { | ||
| Ok(results) | ||
| } | ||
| Ok(SearchPage { | ||
| record_ids, | ||
| resume: None, | ||
| timed_out: false, | ||
| }) |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
The depth-resolve resume state loses every record.
At the resolve timeout the code stores record_ids: out.clone() together with resolve_idx: ridx. out holds only the records that passed the filter, so its length is at most ridx, while ridx indexes the original unfiltered record_ids.
On resume, record_ids is restored from out and ridx from resolve_idx. The condition ridx >= record_ids.len() is then true on the first iteration, the loop breaks immediately, and record_ids = out assigns a fresh empty vector. The call returns zero results and reports completion.
Today no caller combines depth = Some(..) with a deadline, so the path is latent. The state machine is still incorrect and will silently drop results as soon as such a caller appears.
Keep the unfiltered list in the resume state and carry the filtered prefix separately.
🐛 Proposed fix
if let Some(m) = max_len {
- let mut out = Vec::new();
+ // `out` = phần đã filter; `record_ids` giữ nguyên danh sách gốc để
+ // `resolve_idx` luôn trỏ đúng vị trí khi resume.
+ let mut out: Vec<usize> = Vec::new();
let mut ridx = resolve_idx;
let storage = self.storage.read().await;
loop {
if let Some(dl) = deadline
&& Instant::now() >= dl
{
return Ok(SearchPage {
record_ids: out.clone(),
resume: Some(SearchResume {
cand_idx: candidates.len(),
dfs: None,
- record_ids: out.clone(),
+ record_ids: record_ids.clone(),
resolve_idx: ridx,
}),
timed_out: true,
});
}This alone still discards the already-filtered out across the resume boundary. Either add a dedicated resolved: Vec<usize> field to SearchResume, or drop the incremental out and re-filter from resolve_idx = 0 on resume.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@crates/codegraph-graph/src/search.rs` around lines 572 - 613, Fix the
depth-filter resume flow in the search resolve block so resume state preserves
the original unfiltered record_ids while also retaining the already-filtered
output. Add a dedicated resolved-results field to SearchResume and restore/use
it across retries, or restart filtering from resolve_idx zero; ensure resumed
searches do not discard previously accepted records or report premature
completion.
Comment on lines
+85
to
+89
| ``` | ||
| codegraph_search_symbol timed out after 2000ms (collected 134 symbols so far). | ||
| Retry the same call with the same arguments plus "resume": "<id>" to continue | ||
| the search from where it stopped. | ||
| ``` |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win
Add a language to the fenced code block.
markdownlint reports MD040 for this block. The block holds plain error text, so text is appropriate.
📝 Proposed fix
-```
+```text
codegraph_search_symbol timed out after 2000ms (collected 134 symbols so far).
Retry the same call with the same arguments plus "resume": "<id>" to continue
the search from where it stopped.📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| ``` | |
| codegraph_search_symbol timed out after 2000ms (collected 134 symbols so far). | |
| Retry the same call with the same arguments plus "resume": "<id>" to continue | |
| the search from where it stopped. | |
| ``` |
🧰 Tools
🪛 markdownlint-cli2 (0.23.2)
[warning] 85-85: Fenced code blocks should have a language specified
(MD040, fenced-code-language)
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@crates/codegraph-mcp/src/server-instructions.md` around lines 85 - 89, Update
the fenced code block in the instructions text to declare the text language,
using a text fence for the plain error message and preserving its contents
unchanged.
Source: Linters/SAST tools
Comment on lines
658
to
+693
| "codegraph_function_scope" => { | ||
| let target = resolve_target(api, &args, "id", "func_name", &[]).await?; | ||
| match target { | ||
| Target::Ambiguous(v) => Ok(json_str(v)), | ||
| Target::Ambiguous(v) => emit_value(root.as_str(), v), | ||
| Target::Symbol(sym) => match api.function_scope(sym.id).await { | ||
| Some(scope) => serde_json::to_string_pretty(&scope) | ||
| .map_err(|e| Error::Invalid(e.to_string())), | ||
| None => Ok(serde_json::to_string_pretty(&json!({ | ||
| "function": sym.name, | ||
| "parameters": [], | ||
| "locals": [], | ||
| "total": 0, | ||
| })) | ||
| .map_err(|e| Error::Invalid(e.to_string()))?), | ||
| Some(scope) => { | ||
| let detail = detail_from_args(&args, session_detail); | ||
| let format = format_from_args(&args, session_format); | ||
| let parameters: Vec<Value> = scope | ||
| .parameters | ||
| .iter() | ||
| .map(|s| symbol_json(root.as_str(), s, detail, format)) | ||
| .collect(); | ||
| let locals: Vec<Value> = scope | ||
| .locals | ||
| .iter() | ||
| .map(|s| symbol_json(root.as_str(), s, detail, format)) | ||
| .collect(); | ||
| emit_value( | ||
| root.as_str(), | ||
| json!({ | ||
| "function": symbol_json(root.as_str(), &scope.function, detail, format), | ||
| "parameters": parameters, | ||
| "locals": locals, | ||
| }), | ||
| ) | ||
| } | ||
| None => emit_value( | ||
| root.as_str(), | ||
| json!({ | ||
| "function": sym.name, | ||
| "parameters": [], | ||
| "locals": [], | ||
| "total": 0, | ||
| }), | ||
| ), |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win
codegraph_function_scope returns two different shapes for the function key.
The Some(scope) branch sets "function" to the output of symbol_json, which is an array in minimize style or an object in medium style. The None branch sets "function" to a plain name string. A consumer that parses the response must handle three shapes for one key.
Use a distinct key for the not-found case, or emit the resolved symbol through symbol_json in both branches.
🔧 Proposed fix
None => emit_value(
root.as_str(),
json!({
- "function": sym.name,
+ "function": symbol_json(
+ root.as_str(),
+ &sym,
+ detail_from_args(&args, session_detail),
+ format_from_args(&args, session_format),
+ ),
"parameters": [],
"locals": [],
"total": 0,
}),
),📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| "codegraph_function_scope" => { | |
| let target = resolve_target(api, &args, "id", "func_name", &[]).await?; | |
| match target { | |
| Target::Ambiguous(v) => Ok(json_str(v)), | |
| Target::Ambiguous(v) => emit_value(root.as_str(), v), | |
| Target::Symbol(sym) => match api.function_scope(sym.id).await { | |
| Some(scope) => serde_json::to_string_pretty(&scope) | |
| .map_err(|e| Error::Invalid(e.to_string())), | |
| None => Ok(serde_json::to_string_pretty(&json!({ | |
| "function": sym.name, | |
| "parameters": [], | |
| "locals": [], | |
| "total": 0, | |
| })) | |
| .map_err(|e| Error::Invalid(e.to_string()))?), | |
| Some(scope) => { | |
| let detail = detail_from_args(&args, session_detail); | |
| let format = format_from_args(&args, session_format); | |
| let parameters: Vec<Value> = scope | |
| .parameters | |
| .iter() | |
| .map(|s| symbol_json(root.as_str(), s, detail, format)) | |
| .collect(); | |
| let locals: Vec<Value> = scope | |
| .locals | |
| .iter() | |
| .map(|s| symbol_json(root.as_str(), s, detail, format)) | |
| .collect(); | |
| emit_value( | |
| root.as_str(), | |
| json!({ | |
| "function": symbol_json(root.as_str(), &scope.function, detail, format), | |
| "parameters": parameters, | |
| "locals": locals, | |
| }), | |
| ) | |
| } | |
| None => emit_value( | |
| root.as_str(), | |
| json!({ | |
| "function": sym.name, | |
| "parameters": [], | |
| "locals": [], | |
| "total": 0, | |
| }), | |
| ), | |
| "codegraph_function_scope" => { | |
| let target = resolve_target(api, &args, "id", "func_name", &[]).await?; | |
| match target { | |
| Target::Ambiguous(v) => emit_value(root.as_str(), v), | |
| Target::Symbol(sym) => match api.function_scope(sym.id).await { | |
| Some(scope) => { | |
| let detail = detail_from_args(&args, session_detail); | |
| let format = format_from_args(&args, session_format); | |
| let parameters: Vec<Value> = scope | |
| .parameters | |
| .iter() | |
| .map(|s| symbol_json(root.as_str(), s, detail, format)) | |
| .collect(); | |
| let locals: Vec<Value> = scope | |
| .locals | |
| .iter() | |
| .map(|s| symbol_json(root.as_str(), s, detail, format)) | |
| .collect(); | |
| emit_value( | |
| root.as_str(), | |
| json!({ | |
| "function": symbol_json(root.as_str(), &scope.function, detail, format), | |
| "parameters": parameters, | |
| "locals": locals, | |
| }), | |
| ) | |
| } | |
| None => emit_value( | |
| root.as_str(), | |
| json!({ | |
| "function": symbol_json( | |
| root.as_str(), | |
| &sym, | |
| detail_from_args(&args, session_detail), | |
| format_from_args(&args, session_format), | |
| ), | |
| "parameters": [], | |
| "locals": [], | |
| "total": 0, | |
| }), | |
| ), |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@crates/codegraph-mcp/src/tools.rs` around lines 658 - 693, Normalize the
not-found response in the codegraph_function_scope handler so the "function" key
has the same shape in both branches. In the None branch, emit the resolved
symbol through symbol_json using the existing detail and format values, or use a
distinct key for the plain name instead; prefer preserving "function" with
symbol_json.
There was a problem hiding this comment.
Actionable comments posted: 1
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (6)
crates/codegraph-mcp/src/tools.rs (6)
734-744: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick winDo not expose the limited result count as
total.
GraphApi::references(call_name, limit)truncates the result vector before this handler computeshits.len(). The field therefore reports the returned count, not the total number of matching callers. Return a real total from the API, or rename the field tocount. (raw.githubusercontent.com)🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-mcp/src/tools.rs` around lines 734 - 744, Update the references handler around GraphApi::references and the emitted JSON so total does not use the truncated hits.len() value. Prefer extending the API result to provide the true total number of matching callers and emit that value; otherwise rename the field to count to accurately represent the limited result count.
459-475: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick winMake path-prefix filtering boundary-aware.
starts_with(prefix)matches sibling names. For example,path="src"also matchessrc_old/..., and/workspace/srcmatches/workspace/src2/.... Match the exact path or a path followed by/for both absolute and relative candidates.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-mcp/src/tools.rs` around lines 459 - 475, Update the filtering closure in the "codegraph_files" branch to use boundary-aware prefix matching for both f.path and strip_root_prefix(&f.path, root.as_str()). Accept only an exact path match or a candidate beginning with the prefix followed by "/", preventing sibling names such as src_old or src2 from matching.
551-552: 🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick winApply the active serializer to ambiguous target responses.
Target::Ambiguouscontains rawSymbolvalues, and these branches pass them directly toemit_value. Aformat="minimize"request therefore returns object-shaped matches, while resolved responses return minimized arrays. Serializematcheswithsymbol_jsonbefore emitting the ambiguous response. Thecodegraph_symbolpath already follows this pattern. (raw.githubusercontent.com)Also applies to: 602-603, 667-668
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-mcp/src/tools.rs` around lines 551 - 552, Update the Target::Ambiguous response branches to serialize each raw Symbol in matches with symbol_json before passing the result to emit_value, including the corresponding branches near the other referenced locations. Match the existing codegraph_symbol serialization pattern so format="minimize" produces the same minimized array shape as resolved responses.
325-345: 🗄️ Data Integrity & Integration | 🟠 Major | 🏗️ Heavy liftReturn the continuation token after a non-timeout page.
GraphApi::search_resumablereturnsresumewhen a search times out or when more pages remain. This branch uses the token only in the timeout error and discards it on the normal path. Queries with more matches thanlimittherefore lose their continuation token, and the server-side cursor becomes unreachable. Return a stable envelope withresults,total, andresume, and update the tool description. (raw.githubusercontent.com)🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-mcp/src/tools.rs` around lines 325 - 345, Update the normal completion path of the search handler around GraphApi::search_resumable to return a stable envelope containing the mapped results, total, and the optional resume token instead of emitting the bare result array. Preserve the timeout error’s continuation token, include the non-timeout out.resume value in the response, and revise the tool description to document the new response shape and continuation behavior.
949-958: 🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick winPreserve semantic
falsefields during default omission.
is_default_valueremoves every false boolean. This removescompact=false, even though the class-method schema defaultscompacttotrue. It can also removepresent=false, which identifies an absent simulation result. Restrict omission to fields with an explicit false default, or preserve protocol and status fields explicitly. (raw.githubusercontent.com)🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-mcp/src/tools.rs` around lines 949 - 958, Update is_default_value and its callers to preserve semantic false booleans such as compact=false and present=false; only omit false values for fields with an explicit false default. Use the existing field/key context to distinguish defaulted fields, while retaining omission of other default values and existing protocol/status semantics.
563-589: 🎯 Functional Correctness | 🟠 Major | 🏗️ Heavy liftHonor the declared
formatparameter incodegraph_class_methods.The tool schema exposes
format, but this handler only readscompactand always emits object values.format="minimize"and the sessionOutputStyle::Minimizeare ignored. Add format-awareMemberInfoserialization, or removeformatfrom the schema and documentation. (raw.githubusercontent.com)🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@crates/codegraph-mcp/src/tools.rs` around lines 563 - 589, Update the codegraph_class_methods handler to honor the declared format parameter, including session OutputStyle::Minimize, when serializing MemberInfo results. Add format-aware minimized serialization alongside the existing compact/full behavior, or remove format from the tool schema and documentation if it is not intended to be supported; ensure the emitted methods match the selected format.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@crates/codegraph-graph/src/lib.rs`:
- Around line 1714-1716: Update the ScanNames cursor flow to persist the
accumulated matched-name vector alongside name_pos in
SearchCursorPhase::ScanNames, restore it when handling resume, and continue
scanning without discarding prior matches. Add partial-scan deadline tests
covering Prefix, Suffix, and Exact modes, verifying resumed pages and totals
include matches found before timeout.
---
Outside diff comments:
In `@crates/codegraph-mcp/src/tools.rs`:
- Around line 734-744: Update the references handler around GraphApi::references
and the emitted JSON so total does not use the truncated hits.len() value.
Prefer extending the API result to provide the true total number of matching
callers and emit that value; otherwise rename the field to count to accurately
represent the limited result count.
- Around line 459-475: Update the filtering closure in the "codegraph_files"
branch to use boundary-aware prefix matching for both f.path and
strip_root_prefix(&f.path, root.as_str()). Accept only an exact path match or a
candidate beginning with the prefix followed by "/", preventing sibling names
such as src_old or src2 from matching.
- Around line 551-552: Update the Target::Ambiguous response branches to
serialize each raw Symbol in matches with symbol_json before passing the result
to emit_value, including the corresponding branches near the other referenced
locations. Match the existing codegraph_symbol serialization pattern so
format="minimize" produces the same minimized array shape as resolved responses.
- Around line 325-345: Update the normal completion path of the search handler
around GraphApi::search_resumable to return a stable envelope containing the
mapped results, total, and the optional resume token instead of emitting the
bare result array. Preserve the timeout error’s continuation token, include the
non-timeout out.resume value in the response, and revise the tool description to
document the new response shape and continuation behavior.
- Around line 949-958: Update is_default_value and its callers to preserve
semantic false booleans such as compact=false and present=false; only omit false
values for fields with an explicit false default. Use the existing field/key
context to distinguish defaulted fields, while retaining omission of other
default values and existing protocol/status semantics.
- Around line 563-589: Update the codegraph_class_methods handler to honor the
declared format parameter, including session OutputStyle::Minimize, when
serializing MemberInfo results. Add format-aware minimized serialization
alongside the existing compact/full behavior, or remove format from the tool
schema and documentation if it is not intended to be supported; ensure the
emitted methods match the selected format.
🪄 Autofix
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: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: a498c66a-fa34-42ac-b7c9-2ee2af7da7a0
📒 Files selected for processing (6)
crates/codegraph-api/src/lib.rscrates/codegraph-api/tests/api.rscrates/codegraph-graph/src/lib.rscrates/codegraph-graph/src/radix.rscrates/codegraph-mcp/src/lib.rscrates/codegraph-mcp/src/tools.rs
🚧 Files skipped from review as they are similar to previous changes (4)
- crates/codegraph-api/src/lib.rs
- crates/codegraph-graph/src/radix.rs
- crates/codegraph-mcp/src/lib.rs
- crates/codegraph-api/tests/api.rs
Comment on lines
+1714
to
+1716
| pagination: Pagination, | ||
| resume: Option<SearchCursor>, | ||
| deadline: Option<Instant>, |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Persist matched names in the ScanNames cursor.
When a Prefix, Suffix, or Exact scan reaches its deadline after finding matches, the cursor retains only name_pos. The next call creates a new matched vector and expands only names after that position. This drops earlier matches and returns an incorrect page and total.
Store the accumulated matched names in SearchCursorPhase::ScanNames, restore them on resume, and add a partial-scan timeout test for each scan-based mode.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@crates/codegraph-graph/src/lib.rs` around lines 1714 - 1716, Update the
ScanNames cursor flow to persist the accumulated matched-name vector alongside
name_pos in SearchCursorPhase::ScanNames, restore it when handling resume, and
continue scanning without discarding prior matches. Add partial-scan deadline
tests covering Prefix, Suffix, and Exact modes, verifying resumed pages and
totals include matches found before timeout.
hungpham10
added a commit
that referenced
this pull request
Oct 8, 2026
* Finetune to reduce LLM token * style: apply rustfmt * Fix lint
hungpham10
added a commit
that referenced
this pull request
Oct 25, 2026
…eaning magic features (#51) * Implement dynamic dispatch flag + fan-out * Enrich Go semantics: method-set matching, function values, imports, concurrency markers - Method-set matching cho implicit interface: receiver_fn hook lưu receiver base type vào Symbol.type_name (`func (f *File) Write` → "File"); khai báo method trong interface body là node `method_elem` (không phải method_declaration) — giờ được extract với scope đúng vào interface symbol. build_go_method_sets tính structural typing: T implements I ⟺ method names(I) ⊆ method names(T); w.Write() fan-out chỉ tới các type thỏa thay vì naive CHA (fallback giữ nguyên khi method-set chưa đủ dữ liệu). - Function value tracking: go_fn_ptr_assign bắt `handler := process` / `handler = process` → PointerAssign fan-out (dùng chung machinery C). - Package/import: FileDeps { package, imports } mới trong ParseResult; go_file_deps đọc package_clause + import_spec (alias→path); alias_qualified_name resolve `orders.Get` qua import alias cho in-repo packages (match segment cuối path). - Concurrency markers: MARKER_GO/MARKER_DEFER (Marker::Go/Defer) vào chain cho `go f()`/`defer f()`; select_statement wired vào switch machinery (communication_case); EffectType::Panic cho panic/log.Fatal/os.Exit. Tests: 5 extractor (receiver type, interface method scope, fn value, file deps, markers) + 2 e2e graph (method-set precision, import alias). Clippy sạch, fmt xong. * Go method-set: match signature, không chỉ tên + gom LangSpec thành nhóm - Symbol.sig_key: signature key chuẩn hóa `(param_types)(result_types)` (bỏ tên param) qua sig_key_fn hook; go_signature_key dùng chung cho method_declaration/method_elem/function_declaration. build_go_method_sets thỏa interface ⟺ (name, sig_key) khớp — `Wrong.Read(string) string` không còn bị coi là implement `Reader.Read([]byte) (int, error)`. Thiếu key nào → fallback name-only (lenient); interface 0 method → skip. - LangSpec refactor: gom 3 nhóm base / hooks (HookSpec — các Option<fn>, HookSpec::none()) / markers (MarkerSpec — control-flow kinds + field names, MarkerSpec::dflt()); LangSpec::dflt() const + FRU hợp lệ trong static. 14 language spec chỉ khai field KHÁC default — thêm hook mới không còn phải sửa 14 file. * Rust semantic enrichment: field/param typing, use indexing, cross-file impl linking, sig_key - decls += field_declaration/parameter/let_declaration: struct field, tham số và local giờ là symbol có type_name (từ field `type`) — tie-break field_declared_type resolve `db.query()` theo type của `db`. Parameter lấy tên từ field `pattern` TRƯỚC name_type_fallback (tránh lấy nhầm tên kiểu). - rust_file_deps: extract `use` (alias = `as` alias hoặc segment cuối); import_alias_candidates hỗ trợ `Type::method` — `use crate::storage::Database` + `Database::new()` resolve về method của class Database (Alias). - link_impls_globally: impl linking chuyển lên global ingest — `impl Worker` khác file với `struct Worker` vẫn nối (type_ref → def, re-parent methods). - rust_signature_key: `(params)(return)` cho method/fn (self → "self") — nguyên liệu trait matching theo signature. * Python semantic enrichment + persist dynamic indexes qua reopen Python (theo review gaps): - param_container_kinds: `parameters` container → Parameter symbols (identifier / typed_parameter / default_parameter, type từ type hint) — field/param typing cho tie-break resolution. - field_from_self_attr: `self.attr = Database()` → Field symbol scope vào class, type từ callee text — instance-attr typing. - py_file_deps: import/import-as/from-import → FileDeps (alias→path), ăn vào import_alias_candidates sẵn có. - py_superclasses: `class File(Writer)` → impl_trait "Writer" — inheritance fan-out qua machinery impl_index (giống C++ base class). Persistence (fix: dynamic indexes chỉ ở RAM, MCP restart là mất): - Storage trait + sqlite (sg_file_deps/sg_fn_ptr_assigns) + in_memory: persist ĐẦU VÀO thô FileDeps + FnPtrAssigns theo file path. - rebuild() khi open: rebuild file_imports/fn_ptr_index từ blob + link_impls_globally/impl_index/go_impls từ sg_symbols — resolution quality giữ nguyên sau restart. E2e: dynamic_indexes_survive_reopen. * Rust config review fixes: trait-declared methods + macro invocations Review bằng fixture "thật" (trait + impl + macro + closure + async) phát hiện: - function_signature_item (method KHAI BÁO trong trait) không được extract — không có symbol scope vào Interface nên fan-out Rust không bao giờ fire trên code thật (các test trước dùng synthetic symbols). Thêm vào decls → trigger Interface như Go method_elem. - macro_invocation không phải call (println!/vec! mất hoàn toàn) — thêm CallRule với rust_macro_call_name (`println!`). - E2E extraction thật: trait Repo{ping} + 2 impl → r.ping() fan-out cả hai (HierarchyFanOut) — chứng minh path thật chạy, không chỉ synthetic. - Cập nhật chain expectation test cũ: cg_t!/cg_d! giờ là call thật. Còn lại (defer): closure/callable tracking (cb()), generic instantiation, associated type semantics, UFCS. * Persist file_deps/fn_ptr_assigns cho MỌI storage backend Trước đó chỉ sqlite + in_memory implement — lmdb/postgres/mysql/redis rơi về no-op default = mất dynamic indexes khi reopen trên các backend đó. - lmdb: 2 named DBI (sg_file_deps/sg_fn_ptr_assigns) + set/all (cursor iter) + clear_entities. - postgres/mysql: bảng sg_file_deps/sg_fn_ptr_assigns (repo_id, path) PK, CREATE TABLE IF NOT EXISTS idempotent khi mở repo (không cần migration tay); mysql dùng ON DUPLICATE KEY UPDATE; clear_entities thêm 2 bảng. - redis: 2 hash keys (filedeps/fnptpassigns) + DEL trong clear_entities. - cached.rs: forward 4 methods xuống inner storage (trước đây rơi về default no-op → bypass backend thật!). - Test roundtrip in_memory (set/all/clear); build + test --all-features. * callers()/impact xuyên dyn-trait: fan-out fire theo receiver_text + dyn_callers index Review qua MCP index thật (codegraph-rs tự index chính nó) phát hiện: - `callers(set_file_deps)` trả rỗng dù ingest_with_progress gọi nó qua `dyn Storage` — call qua trait object bị resolve Single vì is_method_dispatch check `call_name.contains('.')` nhưng extractor đã rút call_name còn bare, trong khi receiver_text vẫn giữ receiver. - Fix 1: is_method_dispatch = receiver_text.is_some() || call_name.contains('.') → fan-out fire cho dyn-trait call. - Fix 2: dyn_callers (callee → callers) build từ call_meta khi ingest, scan call records khi rebuild; callers_resumable merge vào pending (dedup qua visited) → callers()/impact() thấy caller xuyên dyn-trait/interface/fn-ptr. - E2E: callers() của từng impl method thấy dyn caller (trước đây rỗng). Note vận hành: MCP server đang chạy bị MDB_BAD_RSLOT sau re-index in-process — reader locktable stale, cần restart MCP server (issue riêng, chưa xử lý). * Java + JavaScript/TypeScript semantic enrichment Review setup vs machinery Go/Rust/Python đã có — lấp các gap: Java (typing đã tốt nhất: field/param/local + annotations + implements): - java_supers: gộp extends (superclass) + implements vào impl_trait — inheritance fan-out qua machinery C++-base (superclass đứng trước). - java_signature_key: (params)(return) — tên param bỏ. - java_file_deps: import (bỏ static/wildcard) → FileDeps; strip keyword "extends" khỏi superclass text. JavaScript/TypeScript: - param_container_kinds formal_parameters: js identifier/rest_pattern, ts required/optional_parameter (name từ field pattern, type từ type_annotation — strip ':' + '?'). Trước đây params KHÔNG được extract. - js_file_deps (share TS): import default/named-as/namespace → FileDeps. Bug kèm theo: class_type_name cũ không bao giờ match JS thuần (class_heritage không có extends_clause — chỉ TS có); js_extends xử lý cả 2 grammar. - js_extends → impl_trait: inheritance fan-out cho class extends. - js_fn_value_assign: `handler = process` / `const h = process` (identifier thuần) → PointerAssign machinery. Tests: 3 fixture mới (java supers/sig/imports; js params/imports/extends/ fn-value; ts typed params + interface method scope). 20/21 suite (Redis env). * PHP/Ruby/Lua/Scala semantic enrichment — hoàn thiện fan-out toàn ngôn ngữ PHP (typing đã tốt: params/property/variable): - php_supers: base_clause (extends) + class_interface_clause (implements) — LÀ NODE KIND không phải field name (child_by_field_name trả None!) — inheritance fan-out. - php_signature_key + php_use_deps (namespace_use_declaration, bỏ function/const/wildcard). Ruby (mỏng nhất — params chưa từng extract): - param_container_kinds method_parameters: identifier/optional_parameter/ keyword_parameter/splat_parameter. - ruby_superclass → impl_trait (bỏ dynamic superclass `< Struct.new(..)`). - ruby_require_deps: require/require_relative → FileDeps. Lua: - param_container_kinds "parameters" (kind thật, không phải parameter_list). - lua_require_deps: `local mod = require('mod')` ghép variable_list với expression_list theo vị trí. - lua_fn_value_assign: `handler = process` → PointerAssign. Scala: - scala_supers: field "extend" (extends_clause là node kind, field tên "extend") — extends B with T → "B,T". - scala_signature_key + scala_file_deps (import_declaration, selector `.{a, b => c}`); token keyword "import" có cùng kind — guard child_count. Bài học chung: field name vs node kind khác nhau từng grammar — luôn dump AST trước khi wire hook. * Swift/C# semantic enrichment — đủ 14/14 ngôn ngữ Swift (trước đây gần trống — không hooks): - decls += protocol_function_declaration: method KHAI BÁO trong protocol scope vào Interface → fan-out (đúng bug Rust từng mắc). - swift_supers: inheritance_specifier (`class FileRepo: BaseRepo, Repo`) → impl_trait — protocol conformance đi cùng cơ chế fan-out. - swift_signature_key: Swift dùng TRÙNG tên field "name" cho tên param VÀ type (lấy field "name" cuối); match cả protocol_function_declaration; return = user_type sau token "->". - swift_import_deps: import_declaration → FileDeps. C# (typing/annotations sẵn tốt): - csharp_bases: base_list (node kind, unnamed child) → impl_trait (base class + interfaces, strip generic). - csharp_signature_key: return field là "returns" (không phải "type"). - csharp_using_deps: `using App.Repos` + alias form `using Db = App.Data`. Tests: 2 fixture mới (swift protocol/supers/sig/imports; csharp bases/sig/ using). 25/25 dynamic_dispatch, 20/21 workspace (Redis env-only). * FunctionValue + member assignment + MARKER_AWAIT (priority 1/3/6) A. FunctionValue cho 4 ngôn ngữ còn thiếu: - Python: `handler = process` (identifier=identifier) + `self.handler = process` (member, var = "self.handler"). - Ruby: CHỈ `handler = method(:process)` — plain `x = process` là LỜI GỌI trong Ruby, không track. - Java: `Runnable r = this::run` — method_reference, target = identifier cuối (không phụ thuộc field name grammar). - C#: `Action handler = Process` — value là identifier UNNAMED (không có field "value") — lấy identifier con cuối khác name. B. Member fn assignment `obj.handler = fn`: - Mở guard fn_ptr_index (bỏ `!contains('.')`) — lookup theo call_name bare lẫn dotted. - js/ts: member_expression left → key full dotted text (`Foo.prototype.run`). - lua: text chứa '.' → full text; bracket `handlers["read"]` skip. - python: self.attr. C. MARKER_AWAIT (Marker::Await = 27): MarkerSpec.await_kinds — python "await", js/ts/csharp/swift "await_expression" (dump AST xác nhận); walk_chain push AWAIT trước call như GO/DEFER → edge mang ngữ cảnh async boundary. E2E: member_fn_assignment_resolves_dotted_call — `obj.handler = process` + `obj.handler()` → PointerAssign (trước đây guard chặn). 25/25 dynamic_dispatch (2 test mới), 20/21 workspace (Redis env-only). * C/C++ enrichment batch 1-4: include graph, fn-ptr mở rộng, sig_key, virtual precision 1. c_include_deps: #include "foo.h" / <openssl/ssl.h> → FileDeps (path nguyên văn, alias = tên file không extension) — include graph qua machinery file_imports; C + C++ share. 2. c_fn_ptr_assign mở rộng: LHS field_expression (`obj.handler` — dotted key, ăn guard fn_ptr_index mới) + pointer_expression (`*handler` → "handler"); RHS `&f` address-of + `&ns::cb` qualified. 3. cpp_signature_key: params từ function_declarator (function_definition 1 tầng declarator; method inline field_declaration cũng 1 tầng — dump mới thấy, không phải 2); return field "type". 4. Virtual precision trong collect_fanout (C++ path): nếu CÓ candidate mang keyword virtual/override (signature chứa — extractor lưu dòng khai báo) → chỉ giữ virtual candidates + base decl; không keyword nào → giữ hết (soundness fallback). Loại non-virtual cùng tên trong hierarchy. E2E: cpp_virtual_filter_prefers_override_candidates (D2 non-virtual bị loại). Tests: 26/26 dynamic_dispatch (2 mới), 121/121 graph lib, 20/21 workspace (Redis env-only). * Web UI: dynamic-dispatch badges, Call sites/Pattern/Trace tabs, search pagination + annotation Explore UI (SvelteKit + GraphQL) lộ đầy đủ query surface mới: - Flow tab: mỗi call hiện badge resolvedVia ([dyn]/[fn-ptr]/[unresolved]/…) + candidate chips click-to-navigate (searchExact). GraphQL enum là SCREAMING_SNAKE_CASE — smoke query bắt được bug casing lowercase ban đầu. - Tab Call sites (references): tìm callers theo call name — thấy cả unresolved/library + dynamic calls; prefill theo symbol name. - Tab Pattern (searchFlow): tìm hàm theo chain pattern (markers/tên/id). - Tab Trace (traceFullFlow — targetId arg): debug-style paths caller → target, indent theo stackDepth, icon theo action (CALL/RETURN/THROW/BRANCH/LOOP), màu theo PathEnding; chọn caller depth. - SymbolSearch: Load more (offset pagination), "continue" khi timed_out (resumable cursor), source by name | by annotation (graphcodeSearchByAnnotation). Smoke qua server thật: flow(ingest_with_progress) có 11 HIERARCHY_FAN_OUT calls với 7 candidates (dyn Storage backends), trace 3 paths, annotation "derive" 196 kết quả, references set_file_deps 2 callers. svelte-check 0 errors, build-web assets embed, workspace tests 20/21 (Redis env-only). * C/C++ batch #6 + #8: declaration–definition linking + namespace-qualified call #6 link_cpp_decl_def (global, chạy cả ingest lẫn rebuild): - `void Service::run() {}` trong .cpp (definition, có chain) nối với method khai báo trong .hpp (declaration, không chain): cùng tên + cùng sig_key (khi cả hai có) + khác file → def re-parent về scope của decl (class/ namespace) + reclassify theo kind của decl; decl.type_ref → def. - Hiệu ứng: fan-out/hierarchy thấy definition (có body/flow), class_info theo type_ref tới nơi có chain. #8 qualified call `orders::run()`: - Short-name fallback split cả '::' (rsplit ['.', ':'] bỏ segment rỗng) — trước đây "::" không split → qualified call unresolved hoàn toàn. - Namespace preference: prefix trước '::' lọc candidate theo tổ tiên Module (namespace) cùng tên hoặc file chứa prefix — `foo::run` không còn nhầm `bar::run`; không subset khớp → giữ tất cả (soundness). E2E: cpp_decl_def_linked_across_files (link đúng, pick_best ưu tiên def has-chain); cpp_qualified_call_prefers_namespace (orders::run chọn đúng namespace, không chọn bar). 123/123 graph lib, 20/21 workspace (Redis env). * C/C++ batch cuối: lambda có tên (#5), macro theo thiết kế ghi-nhận (#9-lite), typedef signature (#4-lite) #5 C++ lambda thành Function symbol: - decls/func_kinds += lambda_expression → chain riêng cho calls trong lambda body (attribution chính xác, không gộp vào enclosing fn). - anonymous_name_fn cpp_lambda_name: `auto f = [](){}` mượn tên biến — LƯU Ý init_declarator đặt tên ở field `declarator`, không phải `name` như JS variable_declarator (grammar khác nhau). #9-lite macro THEO THIẾT KẾ GHI-NHẬN (không resolve): - preproc_function_def + preproc_def → Constant symbol, signature = định nghĩa nguyên văn (`CALL(x) x()`, `MAX 10`) — macro KHÔNG phải function, invocation giữ raw call → usage tra qua call-name index (references). - Hành vi semantic do config khai báo ("LLM tự định nghĩa"): `[[effect_rules]]` trong .codegraph/config.toml (matcher + EffectType, xét trước bảng default) — test chứng minh rule Contains "CALL" → Panic (exception-like). - Không làm (đúng defer): macro expansion, conditional compilation. #4-lite typedef signature: - HookSpec thêm typedef_sig_fn; c_typedef_sig (C + C++ share): typedef + macro → signature = text định nghĩa (strip typedef/#define + ;). Hook ƯU TIÊN extract_signature (nó trả full text trước khiến hook không chạy). - fn-ptr typedef giờ giữ underlying type: `int (*handler_t)(int)` thay vì chỉ `int` từ field type. Tests: cpp_lambda_macro_typedef_batch (lambda symbol + chain + calls, macro Constant + signature + raw invocation + effect-rule config, typedef sig). 20/21 workspace (Redis env-only), clippy sạch. * MCP + GraphQL tool cấu hình effect rules runtime (LLM tự định nghĩa macro semantics) Trước đây `[[effect_rules]]` chỉ đọc được từ .codegraph/config.toml — LLM phải sửa file ngoài band rồi re-index. Giờ có tool chính tắc: - config.rs CRUD API: list_effect_rules (raw — kể cả rule lỗi) / add_effect_rule (validate matcher + EffectType, toml_edit preserve mọi section khác) / remove_effect_rule (theo index). Sau ghi: install_current cho thread hiện tại; orchestrator reload config mỗi index nên rule áp dụng từ lần re-index kế tiếp (re-classify call records). - MCP tool `codegraph_config_effects`: op list|add|remove; mô tả tool kèm ví dụ macro exception-like (contains CHECK → panic) + ghi chú áp dụng. - GraphQL mutation `configEffects(op, matcher, text, effect, index)` mirror cho web UI. Test: effect_rule_crud_roundtrip — add 2 rules → list đúng, classifier load từ file classify CHECK_OK → Panic / db.Query → SqlQuery, effect sai bị chặn (file không đổi), remove theo index, section khác (storage) nguyên vẹn. Fix bug bắt được bởi test: or_insert tạo ArrayOfTables với table rỗng dummy → rule rỗng đầu danh sách. * Bump version to v2.2.11 * Dọn debug eprintln sót (DBG by_name/DBG/DBG3/PHPDBG/SCALADBG) sau các batch debug
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.
Summary by CodeRabbit
New Features
Bug Fixes
Documentation