Modify orioledb.undo_buffers descriptiom and add asserts for 2 *

O_MAX_UNDO_RECORD_SIZE floor

Wrap wal_commit->current_oxid_commit code into CRIT_SECTION

Wrap wal_commit->current_oxid_commit code into CRIT_SECTION

Use VALGRIND_MAKE_MEM_DEFINED() instead of null_unused_bytes()

Save on non-emptying undefined page bytes which and satisfy
VALGRIND_CHECK_MEM_IS_DEFINED checks in btree_page_reorg()

Drop null_unused_bytes() function at all

Advance bridge bitmap iterator on per-page seq scan exhaustion

When dead bridge_ctids accumulate on a TIDBitmap page (left by earlier
UPDATEs; GIN only sees them removed at VACUUM), the page's keybitmap
resolves to fewer live PKs than page_ntuples. The per-page primary seq
scan then exhausts before BRIDGE_NEXT_TUPLE marks the page exhausted,
and the previous code treated the NULL as end-of-scan, silently dropping
every later bitmap page.

Fixes #930

test: full undo image for splits during an active seq scan (non-temp)

Adds a shared-table regression test that an active sequential scan keeps
seeing every pre-operation row when a split rewrites, underneath it, the
on-disk extents it has already collected downlinks for.

A seq scan collects its ON_DISK downlinks up front; a split then moves
tuples to a new page that is not in that set, so the scan can only recover
them from a full pre-split page-level undo image.  That image is kept while a
seq scan is active (meta_page_get_num_seq_scans() > 0) even when the page has
no retained pre-op undo of its own (a freshly written leaf has an invalid
undoLocation, which the retain-horizon test alone treats as "no reader needs
the image").

This mirrors TempLocalPoolTest.test_evict_pages_with_concurrent_seq_scan on a
regular table: with the numSeqScans guard the scan returns all 5000 ids; drop
the guard and it loses the rows the splits moved off the collected pages.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

test: full undo image for splits during an active seq scan (non-temp)

Adds a shared-table regression test that an active sequential scan keeps
seeing every pre-operation row when a split rewrites, underneath it, the
on-disk extents it has already collected downlinks for.

A seq scan collects its ON_DISK downlinks up front; a split then moves
tuples to a new page that is not in that set, so the scan can only recover
them from a full pre-split page-level undo image.  That image is kept while a
seq scan is active (meta_page_get_num_seq_scans() > 0) even when the page has
no retained pre-op undo of its own (a freshly written leaf has an invalid
undoLocation, which the retain-horizon test alone treats as "no reader needs
the image").

This mirrors TempLocalPoolTest.test_evict_pages_with_concurrent_seq_scan on a
regular table: with the numSeqScans guard the scan returns all 5000 ids; drop
the guard and it loses the rows the splits moved off the collected pages.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

test: full undo image for splits during an active seq scan (non-temp)

Adds a shared-table regression test that an active sequential scan keeps
seeing every pre-operation row when a split rewrites, underneath it, the
on-disk extents it has already collected downlinks for.

A seq scan collects its ON_DISK downlinks up front; a split then moves
tuples to a new page that is not in that set, so the scan can only recover
them from a full pre-split page-level undo image.  That image is kept while a
seq scan is active (meta_page_get_num_seq_scans() > 0) even when the page has
no retained pre-op undo of its own (a freshly written leaf has an invalid
undoLocation, which the retain-horizon test alone treats as "no reader needs
the image").

This mirrors TempLocalPoolTest.test_evict_pages_with_concurrent_seq_scan on a
regular table: with the numSeqScans guard the scan returns all 5000 ids; drop
the guard and it loses the rows the splits moved off the collected pages.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

test: full undo image for splits during an active seq scan (non-temp)

Adds a shared-table regression test that an active sequential scan keeps
seeing every pre-operation row when a split rewrites, underneath it, the
on-disk extents it has already collected downlinks for.

A seq scan collects its ON_DISK downlinks up front; a split then moves
tuples to a new page that is not in that set, so the scan can only recover
them from a full pre-split page-level undo image.  That image is kept while a
seq scan is active (meta_page_get_num_seq_scans() > 0) even when the page has
no retained pre-op undo of its own (a freshly written leaf has an invalid
undoLocation, which the retain-horizon test alone treats as "no reader needs
the image").

This mirrors TempLocalPoolTest.test_evict_pages_with_concurrent_seq_scan on a
regular table: with the numSeqScans guard the scan returns all 5000 ids; drop
the guard and it loses the rows the splits moved off the collected pages.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Add test_checkpoint_snapshot_resurrects_aborted_oxid for checkpoint m…

…id-abort tx bug deterministic repro