Topology crash on PostgreSQL 14 mingw windows - sqlmm

[robe]

I sent this upstream as they are having trouble replicating mingw issues and there are ton of tests failing on some mingw64 bots (not all).

​https://www.postgresql.org/message-id/002001d7175f%248942b020%249bc81060%24%40pcorp.us

Crash happens here:

SELECT topology.ST_AddIsoEdge('sqlmm_topology', 1, 2, 'LINESTRING(0 0, 10 0, 5 5, 5 -5)');

stack trace below:

Thread 1 received signal SIGSEGV, Segmentation fault.
[Switching to Thread 22272.0x49ec]
0x00007ffe75f5f23f in ntdll!RtlLeaveCriticalSection ()
   from C:\WINDOWS\SYSTEM32\ntdll.dll
(gdb) bt
#0  0x00007ffe75f5f23f in ntdll!RtlLeaveCriticalSection ()
   from C:\WINDOWS\SYSTEM32\ntdll.dll
#1  0x00007ffe75fd4c0e in ntdll!memset () from C:\WINDOWS\SYSTEM32\ntdll.dll
#2  0x00007ffe75fbc872 in ntdll!__C_specific_handler ()
   from C:\WINDOWS\SYSTEM32\ntdll.dll
#3  0x00007ffe75fd1fef in ntdll!.chkstk () from
C:\WINDOWS\SYSTEM32\ntdll.dll
#4  0x00007ffe75f60939 in ntdll!RtlUnwindEx ()
   from C:\WINDOWS\SYSTEM32\ntdll.dll
#5  0x00007ffe74002eed in msvcrt!_setjmpex ()
   from C:\WINDOWS\System32\msvcrt.dll
#6  0x00000000008d929a in pg_re_throw () at elog.c:1776
#7  0x00000000008d911d in errfinish (filename=<optimized out>,
    lineno=<optimized out>, funcname=0x6b025560 <__func__.114531>
"pg_error")
    at elog.c:589
#8  0x000000006afc91d3 in pg_error (fmt=<optimized out>, ap=<optimized out>)
    at lwgeom_pg.c:250
#9  0x000000006afdfb03 in lwerror (fmt=0x0) at lwutil.c:197
#10 0x000000006b000c16 in lwt_AddIsoEdge (topo=topo(at)entry=0x83629b0,
    startNode=startNode(at)entry=1, endNode=endNode(at)entry=2,
    geom=geom(at)entry=0x7a303b8) at lwgeom_topo.c:794
#11 0x000000006afc70d2 in ST_AddIsoEdge (fcinfo=0x7a2caa8)
    at postgis_topology.c:3567
#12 0x00000000005f7604 in ExecInterpExpr (state=0x7a2c9c0,
    econtext=0x7a2c6c0, isnull=<optimized out>) at execExprInterp.c:1209
#13 0x000000000062f23f in ExecEvalExprSwitchContext (isNull=0x509ef9f,
    econtext=0x7a2c6c0, state=0x7a2c9c0)
    at ../../../src/include/executor/executor.h:315
#14 ExecProject (projInfo=0x7a2c9b8)
    at ../../../src/include/executor/executor.h:349
#15 ExecResult (pstate=<optimized out>) at nodeResult.c:136
#16 0x00000000005fc0ca in ExecProcNode (node=0x7a2c5a8)
    at ../../../src/include/executor/executor.h:247
#17 ExecutePlan (execute_once=<optimized out>, dest=0x7a2b2e8,
    direction=<optimized out>, numberTuples=0, sendTuples=<optimized out>,
    operation=CMD_SELECT, use_parallel_mode=<optimized out>,
    planstate=0x7a2c5a8, estate=0x7a2c370) at execMain.c:1531
#18 standard_ExecutorRun (queryDesc=0x8338990, direction=<optimized out>,
    count=0, execute_once=<optimized out>) at execMain.c:350
#19 0x00000000007a08cb in PortalRunSelect (portal=0x79bcc70,
    portal(at)entry=0x7fffffff, forward=forward(at)entry=true,
count=count(at)entry=0,
    dest=dest(at)entry=0x7a2b2e8) at pquery.c:912
#20 0x00000000007a20cc in PortalRun (portal=0x7fffffff,
    portal(at)entry=0x79bcc70, count=0, count(at)entry=2147483647,
    isTopLevel=isTopLevel(at)entry=true, run_once=run_once(at)entry=true,
    dest=dest(at)entry=0x7a2b2e8, altdest=altdest(at)entry=0x7a2b2e8,
    qc=qc(at)entry=0x509f2d0) at pquery.c:756
#21 0x000000000079d6ab in exec_simple_query (
    query_string=0x79fd1e0 "SELECT topology.ST_AddIsoEdge('sqlmm_topology',
1, 2, 'LINESTRING(0 0, 10 0, 5 5, 5 -5)');") at postgres.c:1173
#22 0x000000000079ee40 in PostgresMain (argc=argc(at)entry=1,
    argv=argv(at)entry=0x509f980, dbname=<optimized out>,
    username=<optimized out>) at postgres.c:4327
#23 0x0000000000708608 in BackendRun (port=0x509f7c0, port=0x509f7c0)
    at postmaster.c:4464
#24 SubPostmasterMain (argc=argc(at)entry=3, argv=argv(at)entry=0x5176af0)
    at postmaster.c:4977
#25 0x000000000093cb13 in main (argc=3, argv=0x5176af0) at main.c:186

[robe]

Okay I just repulled and recompiled without cassert and the error does not happen then. I'm going to retest with cassert again to rule out they made a change upstream.

[robe]

turning cassert off also seemed to make all the postgres regress checks pass too. Retesting now with cassert enabled again to rule out that they just fixed some things.

[robe]

I'm looking at the crashers in postgresql proper. The first one I came across copyselect, crashes in same place as I noted here - ​https://www.postgresql.org/message-id/000201d71781%24a0c5da20%24e2518e60%24%40pcorp.us

elog.c:1776

#3  0x00000000008da31a in pg_re_throw () at elog.c:1776
#4  0x00000000008da19d in errfinish (filename=<optimized out>,
    filename(at)entry=0xa291a0 <BinarySignature+24> "copyto.c",
    lineno=lineno(at)entry=490,
    funcname=funcname(at)entry=0xa29980 <__func__.112127> "BeginCopyTo")
    at elog.c:589
    at elog.c:589

So don't think topology is at fault here and just the way certain errors are being handled.

[robe]

I tested Tom Lane's patch that fixed the issue so closing this out as it will be committed I think to PG14.

​https://www.postgresql.org/message-id/000d01d71869%242dabde70%2489039b50%24%40pcorp.us