Memory Leak Fixes - #258
Open
L1l1thLY wants to merge 1 commit into
Open
Conversation
Author
Issue: The function did not free partition column names, auxiliary column names, metadata column names, and metadata chunk shadow table names. Fix: Added three loops to free this memory:
Issue: In the error path, only vec0_free(pNew) was called to free member variables, but the pNew structure itself was not freed. Fix: Added sqlite3_free(pNew) after vec0_free(pNew).
Issue: zSql allocated via sqlite3_mprintf was never freed. Fix: Added sqlite3_free((void *)zSql) immediately after the sqlite3_prepare_v2 call.
Issue: zSql was not freed in two places:
Fix: Added sqlite3_free((void *)zSql) after each sqlite3_prepare_v2 call.
Issue:
Fix:
|
mceachen
added a commit
to photostructure/sqlite-vec
that referenced
this pull request
Feb 10, 2026
- vec_eachFilter: free pzErrMsg on error to prevent leak - vec_slice: use goto done instead of return in INT8 and BIT cases to ensure vector cleanup is called on malloc failure These were the last 3 unfixed memory leaks from upstream PR asg017#258. The other 4 issues from that PR were already addressed in previous commits. Upstream-PR: asg017#258
mceachen
added a commit
to photostructure/sqlite-vec
that referenced
this pull request
Feb 10, 2026
Add test-error-paths.py with 30 tests specifically targeting error-handling code paths that previously had memory leaks (fixed in PR asg017#258). These tests ensure that error paths are exercised by the test suite so that memory leaks would be caught by sanitizers (ASan/LSan) if reintroduced. Tests cover: - vec_each with invalid inputs (NULL, wrong types, malformed JSON, empty arrays) - vec_slice error conditions (NULL, invalid types, bad indices) - vector_from_value error paths (mismatched types/dimensions in distance/add/sub) - vec0 INSERT/KNN errors (NULL, dimension mismatches, type mismatches) - Metadata operations with invalid inputs - Repeated error operations to stress-test cleanup paths Note: malloc failure paths (OOM conditions) are not tested as they require fault injection (SQLITE_TESTCTRL_FAULT_INSTALL) which is not available in the standard build. The fixes ensure proper cleanup via 'goto done' instead of early 'return' statements. Related: Upstream PR asg017#258
stevemurr
added a commit
to stevemurr/sqlite-vec
that referenced
this pull request
Feb 18, 2026
…ries) This commit applies fixes from PR asg017#258 and additional memory leak fixes identified in issues asg017#245, asg017#259, and asg017#265. Memory leak fixes include: 1. vec_slice: Early return without calling cleanup(vector) - now uses goto done 2. vec_eachFilter: pzErrMsg allocated but not freed on error 3. vec0_free: Missing cleanup for partition, auxiliary, and metadata column names 4. vec0_init: zSql not freed before goto error, pNew not freed on error path 5. vec0_metadata_filter_text: rowids not freed on error after allocation 6. vec0Filter_knn: knn_data not freed on cleanup when error occurs, array cleanup missing on error paths for metadata in-list processing 7. vec0_write_metadata_value: zSql not freed after sqlite3_prepare_v2 8. vec0Update_Insert: zSql not freed after sqlite3_prepare_v2 9. vec0Update_Delete_ClearMetadata: zSql not freed, stmt not finalized on error 10. vec0Update_UpdateAuxColumn: zSql not freed after sqlite3_prepare_v2 11. vec_static_blob_entriesFilter: Complete rewrite to properly handle all error paths - knn_data, queryVector, topk_rowids, distances, candidates, taken not properly freed on errors These fixes address memory corruption and database corruption issues that occurred during vec0 INSERT/DELETE operations and KNN queries. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
stevemurr
added a commit
to stevemurr/sqlite-vec
that referenced
this pull request
Feb 18, 2026
Add comprehensive test suite that exercises code paths fixed in the memory leak fixes from PR asg017#258: - vec0_free: Tests create/destroy cycles with partition, auxiliary, and metadata columns to verify proper cleanup of column names - vec0_init: Tests error conditions (too many partition keys, too many metadata columns, invalid syntax) to verify cleanup on error paths - vec0Update_Insert: Tests batch inserts with various column types - vec0Update_Delete_ClearMetadata: Tests DELETE operations with metadata - vec0Filter_knn: Tests KNN queries with metadata filters - vec0_metadata_filter_text: Tests KNN queries with text filters - vec0_write_metadata_value: Tests inserts with metadata values The tests include stress tests for rapid create/insert/drop cycles and mixed operations that would surface memory leaks when run under memory checkers like valgrind or ASAN. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Oldes
pushed a commit
to Siskin-framework/sqlite-vec
that referenced
this pull request
Feb 23, 2026
…ries) This commit applies fixes from PR asg017#258 and additional memory leak fixes identified in issues asg017#245, asg017#259, and asg017#265. Memory leak fixes include: 1. vec_slice: Early return without calling cleanup(vector) - now uses goto done 2. vec_eachFilter: pzErrMsg allocated but not freed on error 3. vec0_free: Missing cleanup for partition, auxiliary, and metadata column names 4. vec0_init: zSql not freed before goto error, pNew not freed on error path 5. vec0_metadata_filter_text: rowids not freed on error after allocation 6. vec0Filter_knn: knn_data not freed on cleanup when error occurs, array cleanup missing on error paths for metadata in-list processing 7. vec0_write_metadata_value: zSql not freed after sqlite3_prepare_v2 8. vec0Update_Insert: zSql not freed after sqlite3_prepare_v2 9. vec0Update_Delete_ClearMetadata: zSql not freed, stmt not finalized on error 10. vec0Update_UpdateAuxColumn: zSql not freed after sqlite3_prepare_v2 11. vec_static_blob_entriesFilter: Complete rewrite to properly handle all error paths - knn_data, queryVector, topk_rowids, distances, candidates, taken not properly freed on errors These fixes address memory corruption and database corruption issues that occurred during vec0 INSERT/DELETE operations and KNN queries. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> (cherry picked from commit 12259fa)
Oldes
pushed a commit
to Siskin-framework/sqlite-vec
that referenced
this pull request
Feb 23, 2026
Add comprehensive test suite that exercises code paths fixed in the memory leak fixes from PR asg017#258: - vec0_free: Tests create/destroy cycles with partition, auxiliary, and metadata columns to verify proper cleanup of column names - vec0_init: Tests error conditions (too many partition keys, too many metadata columns, invalid syntax) to verify cleanup on error paths - vec0Update_Insert: Tests batch inserts with various column types - vec0Update_Delete_ClearMetadata: Tests DELETE operations with metadata - vec0Filter_knn: Tests KNN queries with metadata filters - vec0_metadata_filter_text: Tests KNN queries with text filters - vec0_write_metadata_value: Tests inserts with metadata values The tests include stress tests for rapid create/insert/drop cycles and mixed operations that would surface memory leaks when run under memory checkers like valgrind or ASAN. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> (cherry picked from commit 4eeb2f2)
vlasky
added a commit
to vlasky/sqlite-vec
that referenced
this pull request
Aug 22, 2026
Ports the fixes from upstream PR asg017#258 (see also asg017#265), adapted to this fork, plus a few additional leaks found while reviewing the same code paths: - vec_slice: route malloc-failure branches through the cleanup label instead of returning early and leaking the parsed vector - vec_eachFilter: free the error message from vector_from_value - vec0_free: free partition, auxiliary, and metadata column names and metadata shadow-table names, previously leaked on every table close - vec0_init: free the auxiliary shadow-table SQL string; free the vtab struct itself on the error path; assign the column counters at the error label so names copied into the vtab during a failed constructor parse are freed too - vec0_metadata_filter_text: free the rowids buffer when the blob read fails - vec0Filter_knn: free knn_data on all error paths, clean up metadata-IN arrays and their text entries on error, and free the top-k buffers when a step after a successful chunk scan fails - vec0_write_metadata_value: free the two SQL strings; always close the metadata blob handle, including on error paths - vec0Update_Insert, vec0Update_Delete_ClearMetadata, vec0Update_UpdateAuxColumn: free SQL strings after prepare; finalize the statement when a step fails - vec_static_blob_entriesFilter: restructure with a cleanup label; the query vector and candidate/taken bitmaps previously leaked even on success, and everything leaked on error paths Verified with macOS leaks(1) on a workload exercising inserts, updates, deletes, KNN queries with IN constraints, metadata/partition/auxiliary columns, and table create/drop: 17 leaks (1168 bytes) before, 0 after.
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.
I found a few memory leaks in the code and attempted to fix them. Could you please take a look and confirm whether the fixes are correct? I’m not entirely sure, but the changes are as follows: