PG16 PostgreSQL 16 crashing on raster install

[robe]

I thought it was just my setup, but debbie is having the same issue with her PG16 run as I am

22:54:44  topology has been added to end of database search_path 
22:54:44 (1 row)
22:54:44 
22:54:44 NOTICE:  schema "public" already exists, skipping
22:54:44 NOTICE:  schema "public" already exists, skipping
22:54:44 psql:/var/lib/jenkins/workspace/postgis/branches/3.4/regress/00-regress-install/share/contrib/postgis/rtpostgis.sql:111: server closed the connection unexpectedly
22:54:44 	This probably means the server terminated abnormally
22:54:44 	before or while processing the request.
22:54:44 psql:/var/lib/jenkins/workspace/postgis/branches/3.4/regress/00-regress-install/share/contrib/postgis/rtpostgis.sql:111: error: connection to server was lost
22:54:44 -----------------------------------------------------------------------------
22:54:44 make: *** [regress/runtest.mk:24: check-regress] Error 1
22:54:44 Build step 'Execute shell' marked build as failure
22:54:44 Triggering a new build of PostgreSQL stop
22:54:44 Finished: FAILURE

It just crashes. ON mine it crashes on line 100 of rtpostgis.sql

As best I can tell, this started happening with the commit of

​https://git.postgresql.org/gitweb/?p=postgresql.git;a=commit;h=a73952b795632b2cf5acada8476e7cf75857e9be

I haven't checked to see what that line corresponds to yet.

Commit a73952b795632b2cf5acada8476e7cf75857e9be by michael

Add check on initial and boot values when loading GUCs

This commit adds a function to perform a cross-check between the initial
value of the C declaration associated to a GUC and its actual boot
value in assert-enabled builds.  The purpose of this is to prevent
anybody reading these C declarations from being fooled by mismatched
values before they are loaded at program startup.

The following rules apply depending on the GUC type:
* bool - can be false, or same as boot_val.
* int - can be 0, or same as the boot_val.
* real - can be 0.0, or same as the boot_val.
* string - can be NULL, or strcmp'd equal to the boot_val.
* enum - equal to the boot_val.

This is done for the system as well custom GUCs loaded by external
modules, which may require extension developers to adapt the C
declaration of the variables used by these GUCs (testing this change
with some of my own modules has allowed me to catch some stupid typos,
FWIW).  This may finish by being a bad experiment depending on the
feedbcak received, but let's see how it goes.

Author: Peter Smith
Reviewed-by: Nathan Bossart, Tom Lane, Michael Paquier, Justin Pryzby
Discussion: https://postgr.es/m/CAHut+PtHE0XSfjjRQ6D4v7+dqzCw=d+1a64ujra4EX8aoc_Z+w@mail.gmail.com

On my setup lines 95 - 107 are these and it is crashing at 100

-- part of Raster type
-- expects input to be Hex WKB
CREATE OR REPLACE FUNCTION raster_in(cstring)
    RETURNS raster
    AS 'C:/ming64gcc81/projects/postgis/postgis-git/regress/00-regress-install/lib/postgis_raster-3.4','RASTER_in'
    LANGUAGE 'c' IMMUTABLE STRICT PARALLEL SAFE;
# line 100
-- part of Raster type
-- expects output to be Hex WKB
CREATE OR REPLACE FUNCTION raster_out(raster)
    RETURNS cstring
    AS 'C:/ming64gcc81/projects/postgis/postgis-git/regress/00-regress-install/lib/postgis_raster-3.4','RASTER_out'
    LANGUAGE 'c' IMMUTABLE STRICT PARALLEL SAFE;

[robe]

I should add debbie doesn't build with every push of postgresql master. So the error would lie anywhere after this commit

​https://git.postgresql.org/gitweb/?p=postgresql.git;a=commit;h=d9d873bac67047cfacc9f5ef96ee488f2cb0f1c3: Clean up some inconsistencies with GUC declarations (details)

and on or before the above commit - ​https://git.postgresql.org/gitweb/?p=postgresql.git;a=commit;h=a73952b795632b2cf5acada8476e7cf75857e9be

but on closer inspection, I'm not seeing any commits between these, so I guess ​https://git.postgresql.org/gitweb/?p=postgresql.git;a=commit;h=a73952b795632b2cf5acada8476e7cf75857e9be is probably the smoking gun.

[robe]

yap confirmed running this is sufficient to crash my backend:

CREATE OR REPLACE FUNCTION raster_in(cstring)
    RETURNS raster
    AS 'C:/ming64gcc81/projects/postgis/postgis-git/regress/00-regress-install/lib/postgis_raster-3.4','RASTER_in'
    LANGUAGE 'c' IMMUTABLE STRICT PARALLEL SAFE;

Sadly gdb is of no help here.

There is no backtrace of meaning

Reading symbols from C:\ming64gcc81\projects\postgresql\rel\pg16w64gcc81\bin\postgres.exe...done.
(gdb) cont
Continuing.
[Thread 17048.0xad4 exited with code 0]
[Thread 17048.0x2874 exited with code 3]
[Thread 17048.0x590c exited with code 3]
[Thread 17048.0x2e98 exited with code 3]
[Thread 17048.0x4d70 exited with code 3]
[Thread 17048.0x64d8 exited with code 3]
[Inferior 1 (process 17048) exited with code 03]

[robe]

I'm guessing it's failing in the _PG_init(void) of raster, which is where default boot values are set.

[Regina Obe <lr@…>]

In 5df6277/git:

Fix crash in PG16 on raster.
Closes #5277 for PostGIS 3.4.0

[Regina Obe <lr@…>]

In 4aabcd4/git:

Fix multi-def error introduced in last commit. References #5277 for PostGIS 3.4.0

[Regina Obe <lr@…>]

In a18bfd0/git:

Mark them as all static. References #5277 for PostGIS 3.4.0

[Regina Obe <lr@…>]

In 76f44de/git:

Support build on Pg16devel for PostGIS 3.3.3

Closes #5257
Closes #5261
Closes #5277
Closes #5308
Closes #5374
Closes ​https://github.com/postgis/postgis/pull/716