Horton

Re: Potential ABI breakage in upcoming minor releases

Noah Misch <noah@leadboat.com>

View thread
From: Noah Misch <noah@leadboat.com>
To: Christoph Berg <myon@debian.org>
Cc: Pavan Deolasee <pavan.deolasee@gmail.com>, pgsql-hackers <pgsql-hackers@postgresql.org>, Tom Lane <tgl@sss.pgh.pa.us>
Date: 2024-11-14T19:17:01Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Undo unintentional ABI break in struct ResultRelInfo.

  2. For inplace update durability, make heap_update() callers wait.

  3. Fix btmarkpos/btrestrpos array key wraparound bug.

  4. Fix calculation of which GENERATED columns need to be updated.

On Thu, Nov 14, 2024 at 10:26:58AM -0800, Noah Misch wrote:
> On Thu, Nov 14, 2024 at 05:29:18PM +0100, Christoph Berg wrote:
> > Re: Noah Misch
> > > Based on a grep of PGXN code, here are some or all of the modules that react
> > > to sizeof(ResultRelInfo):
> > > 
> > > $ grepx -r 'lloc.*ResultRelInfo' | tee /tmp/1 | sed 's/-[^:]*/:/'|sort -u
> > > apacheage::    resultRelInfo = palloc(sizeof(ResultRelInfo));
> > 
> > Confirmed, crashing: AGE for 14..17 (12..13 seem fine)
> 
> Can you share more about the crash, perhaps the following?
> 
> - stack trace
> - any server messages just before the crash
> - which architecture (32-bit x86, 64-bit ARM, etc.)
> - asserts enabled, or not?
> 
> It's not immediately to clear to me why this would crash in a non-asserts
> build.  palloc issues a 512-byte chunk for sizeof(ResultRelInfo)==368 on v16,
> so I expect no actual writing past the end of the chunk.  I don't see
> apacheage allocating a ResultRelInfo other than via one palloc per
> ResultRelInfo (no arrays of them, no stack-allocated ResultRelInfo).  I'll
> also work on installing apacheage to get those answers locally.

On x86_64, I ran these with and without asserts:
  install PostgreSQL 16.4
  install https://github.com/apache/age/tree/master
  make -C age installcheck
  install PostgreSQL 16.5
  make -C age installcheck

The non-asserts build passed.  The asserts build failed with "+WARNING:
problem in alloc set ExecutorState: detected write past chunk" throughout the
diffs, but there were no crashes.  (Note that AGE "make installcheck" creates
a temporary installation, unlike PostgreSQL "make installcheck".)  What might
differ in how you tested?