trending/codebase-memory-mcp
1
0
Fork 0
mirror of https://github.com/DeusData/codebase-memory-mcp.git synced 2026-11-09 09:11:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: honor file_pattern in bm25 graph search (#555)

Browse Source 
Signed-off-by: pcristin <xxxokzxxx@protonmail.com>
This commit is contained in:
Pcristin
2026-06-22 22:05:29 +02:00
committed by GitHub
parent df8bc9254d
commit 0618eaa99a
2 changed files with 99 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/mcp/mcp.c
+42 -3
View File
@@ -1208,6 +1208,7 @@ enum {
BM25_BIND_LIMIT = 3,
BM25_BIND_OFFSET = 4,
BM25_BIND_INNER = 5,
BM25_BIND_FILE = 6,
BM25_SQL_AUTO_LEN = -1,
/* Inner FTS5 candidate cap. SQLite can early-terminate a plain FTS5 query
* (no JOIN/WHERE on outer table) of the form:
@@ -1267,11 +1268,31 @@ static int bm25_build_match(const char *query, char *out, size_t out_size) {
return tokens;
}
static char *bm25_file_pattern_like(const char *file_pattern) {
if (!file_pattern) {
return NULL;
}
char *like = cbm_glob_to_like(file_pattern);
if (like && !strchr(file_pattern, '*') && !strchr(file_pattern, '?')) {
size_t len = strlen(like);
char *contains = malloc(len + MCP_SEPARATOR + SKIP_ONE);
if (contains) {
contains[0] = '%';
memcpy(contains + SKIP_ONE, like, len);
contains[len + SKIP_ONE] = '%';
contains[len + MCP_SEPARATOR] = '\0';
free(like);
like = contains;
}
}
return like;
}
/* Run the BM25 full-text search path and return the JSON result string.
* Returns NULL if FTS5 is unavailable or the query produced no usable tokens,
* in which case the caller falls back to the regex-based search path. */
static char *bm25_search(cbm_store_t *store, const char *project, const char *query, int limit,
int offset) {
static char *bm25_search(cbm_store_t *store, const char *project, const char *query,
const char *file_pattern, int limit, int offset) {
sqlite3 *db = cbm_store_get_db(store);
if (!db) {
return NULL;
@@ -1281,6 +1302,7 @@ static char *bm25_search(cbm_store_t *store, const char *project, const char *qu
if (tok_count == 0) {
return NULL;
}
char *file_like = bm25_file_pattern_like(file_pattern);
/* BM25 ranked query using a two-step approach to enable FTS5 early termination.
*
@@ -1311,11 +1333,13 @@ static char *bm25_search(cbm_store_t *store, const char *project, const char *qu
"JOIN nodes n ON n.id = fts.rowid "
"WHERE n.project = ?2 "
" AND n.label NOT IN ('File','Folder','Module','Section','Variable','Project') "
" AND (?6 IS NULL OR n.file_path LIKE ?6) "
"ORDER BY rank "
"LIMIT ?3 OFFSET ?4";
sqlite3_stmt *stmt = NULL;
if (sqlite3_prepare_v2(db, sql, BM25_SQL_AUTO_LEN, &stmt, NULL) != SQLITE_OK) {
free(file_like);
return NULL;
}
sqlite3_bind_text(stmt, BM25_BIND_QUERY, fts_query, BM25_SQL_AUTO_LEN, MCP_SQLITE_TRANSIENT);
@@ -1323,6 +1347,11 @@ static char *bm25_search(cbm_store_t *store, const char *project, const char *qu
sqlite3_bind_int(stmt, BM25_BIND_LIMIT, limit > 0 ? limit : BM25_DEFAULT_LIMIT);
sqlite3_bind_int(stmt, BM25_BIND_OFFSET, offset > 0 ? offset : 0);
sqlite3_bind_int(stmt, BM25_BIND_INNER, BM25_INNER_LIMIT);
if (file_like) {
sqlite3_bind_text(stmt, BM25_BIND_FILE, file_like, BM25_SQL_AUTO_LEN, MCP_SQLITE_TRANSIENT);
} else {
sqlite3_bind_null(stmt, BM25_BIND_FILE);
}
/* Count hits within the same inner-limit window — capped at BM25_INNER_LIMIT.
* Uses the identical subquery structure so the FTS5 early-exit applies here too. */
@@ -1337,6 +1366,7 @@ static char *bm25_search(cbm_store_t *store, const char *project, const char *qu
" JOIN nodes n ON n.id = fts.rowid "
" WHERE n.project = ?2 "
" AND n.label NOT IN ('File','Folder','Module','Section','Variable','Project')"
" AND (?6 IS NULL OR n.file_path LIKE ?6)"
")";
sqlite3_stmt *cs = NULL;
if (sqlite3_prepare_v2(db, count_sql, BM25_SQL_AUTO_LEN, &cs, NULL) == SQLITE_OK) {
@@ -1345,6 +1375,12 @@ static char *bm25_search(cbm_store_t *store, const char *project, const char *qu
sqlite3_bind_text(cs, BM25_BIND_PROJECT, project, BM25_SQL_AUTO_LEN,
MCP_SQLITE_TRANSIENT);
sqlite3_bind_int(cs, BM25_BIND_LIMIT, BM25_INNER_LIMIT);
if (file_like) {
sqlite3_bind_text(cs, BM25_BIND_FILE, file_like, BM25_SQL_AUTO_LEN,
MCP_SQLITE_TRANSIENT);
} else {
sqlite3_bind_null(cs, BM25_BIND_FILE);
}
if (sqlite3_step(cs) == SQLITE_ROW) {
total = sqlite3_column_int(cs, 0);
}
@@ -1377,6 +1413,7 @@ static char *bm25_search(cbm_store_t *store, const char *project, const char *qu
emitted++;
}
sqlite3_finalize(stmt);
free(file_like);
yyjson_mut_obj_add_val(doc, root, "results", results);
yyjson_mut_obj_add_bool(doc, root, "has_more", total > offset + emitted);
@@ -1517,7 +1554,9 @@ static char *handle_search_graph(cbm_mcp_server_t *srv, const char *args) {
if (query && query[0]) {
int q_limit = cbm_mcp_get_int_arg(args, "limit", BM25_DEFAULT_LIMIT);
int q_offset = cbm_mcp_get_int_arg(args, "offset", 0);
char *bm25_json = bm25_search(store, project, query, q_limit, q_offset);
char *q_file_pattern = cbm_mcp_get_string_arg(args, "file_pattern");
char *bm25_json = bm25_search(store, project, query, q_file_pattern, q_limit, q_offset);
free(q_file_pattern);
if (bm25_json) {
free(query);
free(project);
tests/test_mcp.c
+57
View File
@@ -466,6 +466,62 @@ TEST(tool_search_graph_includes_node_properties) {
PASS();
}
TEST(tool_search_graph_query_honors_file_pattern_issue552) {
cbm_mcp_server_t *srv = cbm_mcp_server_new(NULL);
ASSERT_NOT_NULL(srv);
cbm_store_t *st = cbm_mcp_server_store(srv);
ASSERT_NOT_NULL(st);
const char *proj = "issue-552";
cbm_mcp_server_set_project(srv, proj);
cbm_store_upsert_project(st, proj, "/tmp/issue-552");
cbm_node_t lib_status = {0};
lib_status.project = proj;
lib_status.label = "Function";
lib_status.name = "status";
lib_status.qualified_name = "issue-552.src.lib.status";
lib_status.file_path = "src/lib/status.c";
lib_status.start_line = 1;
lib_status.end_line = 3;
ASSERT_GT(cbm_store_upsert_node(st, &lib_status), 0);
cbm_node_t component_status = {0};
component_status.project = proj;
component_status.label = "Function";
component_status.name = "status";
component_status.qualified_name = "issue-552.src.components.status";
component_status.file_path = "src/components/status.c";
component_status.start_line = 1;
component_status.end_line = 3;
ASSERT_GT(cbm_store_upsert_node(st, &component_status), 0);
cbm_store_exec(st, "INSERT INTO nodes_fts(nodes_fts) VALUES('delete-all');");
ASSERT_EQ(cbm_store_exec(st,
"INSERT INTO nodes_fts(rowid, name, qualified_name, label, "
"file_path) "
"SELECT id, cbm_camel_split(name), qualified_name, label, file_path "
"FROM nodes;"),
CBM_STORE_OK);
char *resp = cbm_mcp_server_handle(
srv, "{\"jsonrpc\":\"2.0\",\"id\":552,\"method\":\"tools/call\","
"\"params\":{\"name\":\"search_graph\","
"\"arguments\":{\"project\":\"issue-552\",\"query\":\"status\","
"\"file_pattern\":\"src/lib/*\",\"limit\":10}}}");
ASSERT_NOT_NULL(resp);
char *inner = extract_text_content(resp);
ASSERT_NOT_NULL(inner);
ASSERT_NOT_NULL(strstr(inner, "\"search_mode\":\"bm25\""));
ASSERT_NOT_NULL(strstr(inner, "\"file_path\":\"src/lib/status.c\""));
ASSERT_NULL(strstr(inner, "src/components/status.c"));
free(inner);
free(resp);
cbm_mcp_server_free(srv);
PASS();
}
TEST(tool_query_graph_basic) {
cbm_mcp_server_t *srv = setup_mcp_with_data();
@@ -2062,6 +2118,7 @@ SUITE(mcp) {
RUN_TEST(tool_unknown_tool);
RUN_TEST(tool_search_graph_basic);
RUN_TEST(tool_search_graph_includes_node_properties);
RUN_TEST(tool_search_graph_query_honors_file_pattern_issue552);
RUN_TEST(tool_query_graph_basic);
RUN_TEST(tool_index_status_no_project);
RUN_TEST(tool_index_status_includes_git_metadata);