Merge pull request #1326 from Enferlain/fix/1287-persisted-coverage-summary

fix(index): preserve persisted coverage summaries

Author: DeusData

src/mcp/mcp.c

@@ -7347,6 +7347,54 @@ static void add_parse_partial_summary(yyjson_mut_doc *doc, yyjson_mut_val *root,
     yyjson_mut_obj_add_val(doc, root, "parse_partial", pp);
 }
 
+/* The pipeline persists the complete current coverage set before this
+ * response is built. Prefer that set over the per-run errors so incremental
+ * runs that do not revisit a flagged file, and artifact bootstraps, do not
+ * make existing gaps appear to have vanished. By-design exclusions have
+ * their own response surface and are not failures. */
+static bool add_persisted_failure_summaries(yyjson_mut_doc *doc, yyjson_mut_val *root,
+                                            cbm_store_t *store, const char *project,
+                                            const char *logfile) {
+    cbm_coverage_row_t *rows = NULL;
+    int row_count = 0;
+    if (cbm_store_coverage_get(store, project, &rows, &row_count) != CBM_STORE_OK) {
+        return false;
+    }
+
+    int failure_count = 0;
+    for (int i = 0; i < row_count; i++) {
+        const char *kind = rows[i].kind ? rows[i].kind : "";
+        if (strcmp(kind, "not_indexed_dir") != 0 && strcmp(kind, "not_indexed_file") != 0) {
+            failure_count++;
+        }
+    }
+
+    cbm_file_error_t *failures =
+        failure_count > 0 ? calloc((size_t)failure_count, sizeof(*failures)) : NULL;
+    if (failure_count > 0 && !failures) {
+        cbm_store_free_coverage(rows, row_count);
+        return false;
+    }
+
+    int n = 0;
+    for (int i = 0; i < row_count; i++) {
+        const char *kind = rows[i].kind ? rows[i].kind : "";
+        if (strcmp(kind, "not_indexed_dir") == 0 || strcmp(kind, "not_indexed_file") == 0) {
+            continue;
+        }
+        failures[n].path = (char *)rows[i].rel_path;
+        failures[n].reason = (char *)rows[i].detail;
+        failures[n].phase = (char *)rows[i].kind;
+        n++;
+    }
+
+    add_skipped_summary(doc, root, failures, failure_count, logfile);
+    add_parse_partial_summary(doc, root, failures, failure_count);
+    free(failures);
+    cbm_store_free_coverage(rows, row_count);
+    return true;
+}
+
 /* Write the FULL (uncapped) skip list to a per-run logfile — ONLY when >=1 file
  * was skipped (no logfile on a clean run). Location:
  *   $CBM_INDEX_LOG (override) else <cache_dir>/logs/<project>-<epoch>.log
@@ -7406,8 +7454,6 @@ static bool build_index_success_response(cbm_mcp_server_t *srv, yyjson_mut_doc *
                                          const cbm_file_error_t *file_errors, int file_error_count,
                                          const char *logfile) {
     add_excluded_summary(doc, root, excluded_dirs, excluded_count);
-    add_skipped_summary(doc, root, file_errors, file_error_count, logfile);
-    add_parse_partial_summary(doc, root, file_errors, file_error_count);
     add_not_indexed_files_summary(doc, root, p);
 
     int exp_nodes = -1;
@@ -7418,6 +7464,10 @@ static bool build_index_success_response(cbm_mcp_server_t *srv, yyjson_mut_doc *
     const int min_floor = CBM_DUMP_VERIFY_MIN_FLOOR;
 
     cbm_store_t *store = resolve_store(srv, project_name);
+    if (!store || !add_persisted_failure_summaries(doc, root, store, project_name, logfile)) {
+        add_skipped_summary(doc, root, file_errors, file_error_count, logfile);
+        add_parse_partial_summary(doc, root, file_errors, file_error_count);
+    }
     int nodes = 0;
     int edges = 0;
     bool degraded = false;

tests/test_index_resilience.c

@@ -439,6 +439,34 @@ TEST(index_parse_partial_reported) {
     /* Clean neighbors still extract. */
     int funcs = rh_count_label(store, lp.project, "Function");
     ASSERT_GTE(funcs, 1);
+    cbm_store_close(store);
+    store = NULL;
+
+    /* An incremental run that changes only a clean neighbor has no new parser
+     * errors, but its response must still describe the persisted current
+     * coverage state for the untouched partial file. */
+    struct timespec delay = {.tv_sec = 0, .tv_nsec = INCR_FIX_SLEEP_NS};
+    nanosleep(&delay, NULL);
+    ri_write_text(lp.tmpdir, "good.py", "def alpha():\n    return 2\n");
+    char iargs[700];
+    snprintf(iargs, sizeof(iargs), "{\"repo_path\":\"%s\"}", lp.tmpdir);
+    char *iresp = cbm_mcp_handle_tool(lp.srv, "index_repository", iargs);
+    ASSERT_NOT_NULL(iresp);
+    yyjson_doc *idoc = yyjson_read(iresp, strlen(iresp), 0);
+    ASSERT_NOT_NULL(idoc);
+    yyjson_val *isc = yyjson_obj_get(yyjson_doc_get_root(idoc), "structuredContent");
+    ASSERT_NOT_NULL(isc);
+    ASSERT_GTE(yyjson_get_int(yyjson_obj_get(isc, "parse_partial_count")), 1);
+    yyjson_val *ipp = yyjson_obj_get(isc, "parse_partial");
+    ASSERT_NOT_NULL(ipp);
+    char *ipp_json = yyjson_val_write(ipp, 0, NULL);
+    ASSERT_NOT_NULL(ipp_json);
+    ASSERT_NOT_NULL(strstr(ipp_json, "split.c"));
+    free(ipp_json);
+    /* The coverage is persisted state, but the logfile remains per-run. */
+    ASSERT_NULL(yyjson_obj_get(isc, "logfile"));
+    yyjson_doc_free(idoc);
+    free(iresp);
 
     yyjson_doc_free(d);
     free(resp);