Pg18 changes pqsignal() return

[pramsey]

And this in turn breaks our build. Consultation with pgsql-hackers indicates that we've been Doing It Wrong all along.

​https://www.postgresql.org/message-id/flat/4E796D56-79B5-4447-9DA7-5428BC1A6B53%40cleverelephant.ca

[pramsey]

The correct answer is to not try to catch signals at all, leave that to PostgreSQL. Instead, just look at the current value of QueryCancelPending || ProcDiePending from miscadmin.h to find out what the interrupt state is. This turns out to be not hard to do with the existing callback facilities in geos and liblwgeom.

[Paul Ramsey <pramsey@…>]

In 1f7c3150/git:

Rework interrupt handling and remove signal handling
from PostGIS. PostGIS does not need to capture signals,
it just needs to look at the state flags exposed by

PgSQL (QueryCancelPending

ProcDiePending) from

time to time and interrupt itself if they are set.
References #5841

[pramsey]

Wondering if this should be back-ported into 3.5, since the feedback from pgsql-hackers was that basically our implementation has been broken forever.

[Paul Ramsey <pramsey@…>]

In 0549494/git:

Add interrupt support to interpolate grid, and
confirm interrupt support on GDAL contour.
References #5841

[Paul Ramsey <pramsey@…>]

In 0b003a8/git:

Remove use of pqsignal in favour of checking pgsql interrupt
variables in order to respect user interrupt request.
References #5841

[pramsey]

See #5137 for information on former use of executorHook

[robe]

Looks like the commit broke interruptability on winnie, tried triggering a number of times and fails in same spot. The 3.5 branch of hers was green before this. 5 minutes is hard to ignore, but since the other interrupts are passing perhaps its only the GEOS one at issue.

On winnie I see this, but interrupt and interrupt_relate are okay

 ./regress/core/interrupt_buffer .. failed (diff expected obtained: /projects/postgis/tmp/3.5.1dev_pg16_geos3.13_gdal3.9.2w64/test_126_diff)
-----------------------------------------------------------------------------
--- ./regress/core/interrupt_buffer_expected	2024-10-29 18:48:10.401068900 -0400
+++ /projects/postgis/tmp/3.5.1dev_pg16_geos3.13_gdal3.9.2w64/test_126_out	2025-01-25 12:40:10.602262300 -0500
@@ -1,3 +1,3 @@
 ERROR:  canceling statement due to statement timeout
-buffer interrupted on time
+buffer interrupted late: 00:04:53.254363 (00:00:06.07376 tolerated)
 1|5
-----------------------------------------------------------------------------

Dronie did fail once, but I think this is a random interrupt failure

On dronie:

 regress/core/interrupt .. failed (diff expected obtained: /tmp/pgis_reg/test_124_diff)

-----------------------------------------------------------------------------

--- /drone/src/regress/core/interrupt_expected	2025-01-24 23:12:20.455468619 +0000

+++ /tmp/pgis_reg/test_124_out	2025-01-25 00:19:54.053246531 +0000

@@ -1,3 +1,3 @@

 ERROR:  canceling statement due to statement timeout

-segmentize interrupted on time

+ERROR:  canceling statement due to statement timeout

 1|LINESTRING(0 0,2 0,4 0)

-----------------------------------------------------------------------------

 regress/core/interrupt_relate .. ok in 1428 ms

 regress/core/interrupt_buffer .. ok in 618 ms

[Regina Obe <lr@…>]

In f5048c59/git:

Fix windows interrupt
References #5841 for PostGIS 3.6.0

[Regina Obe <lr@…>]

In 35e55942/git:

Fix windows interrupt
Closes #5841 for PostGIS 3.5.3