Re: ZSTD TOAST compression, and an extensible compression method encoding

Hi Greg,

On Sat, Nov 14, 2026 at 1:50 PM Greg Burd <greg(at)burd(dot)me> wrote:
> Most of the changes are in 0001 and 0002, since those are the ones
> closest to being committed.

I found nothing broken in 0001 or 0002, and one check in 0002 that no
test exercises.

0001: reading pglz and lz4 values through text, substr(),
pg_column_*(), verify_heapam() and test_decoding gives byte-identical
output (58,990 lines) on master and master+0001, and each of four wrong
conversions I planted changes that output.

0002: on 0001-0003 under cassert, reverting any one check in
toast_compression_id_resolve() fails t/007, except method_byte ==
TOAST_INVALID_COMPRESSION_ID. From 0004 on, with the last ID at 5, that
check alone keeps a long pointer with byte 3 away from the Assert in
toast_compression_id_needs_cmid_byte(), and t/007 has no byte-3 case.
A byte-3 row in the long-form part of t/007 would cover it.

Old data reads the same with a master data directory started under
0001-0003, and after pg_upgrade from REL_18 and REL_19 to 0001-0003.
The scripts and the results, forged pointers included, are attached.

Manu