Re: Proposal: allow database-specific role memberships

Hi,

On Fri, Jan 21, 2022 at 07:01:21PM -0800, Kenaniah Cerny wrote:
> Thanks for the feedback.
>
> I have attached an alternate version of the v5 patch that incorporates the
> suggested changes to the documentation and DRYs up some of the acl.c code
> for comparison. As for the databaseOid / InvalidOid parameter, I'm open to
> any suggestions for how to make that even cleaner, but am currently at a
> loss as to how that would look.
>
> CI is showing a failure to run pg_dump on just the Linux - Debian Bullseye
> job (https://cirrus-ci.com/task/5265343722553344). Does anyone have any
> ideas as to where I should look in order to debug that?

Did you try to reproduce it on some GNU/Linux system? FTR I had and I get a
segfault in pg_dumpall:

(gdb) bt
#0 __pthread_kill_implementation (threadid=<optimized out>, signo=signo(at)entry=6, no_tid=no_tid(at)entry=0) at pthread_kill.c:44
#1 0x00007f329e7e40cf in __pthread_kill_internal (signo=6, threadid=<optimized out>) at pthread_kill.c:78
#2 0x00007f329e7987a2 in __GI_raise (sig=sig(at)entry=6) at ../sysdeps/posix/raise.c:26
#3 0x00007f329e783449 in __GI_abort () at abort.c:79
#4 0x00007f329e7d85d8 in __libc_message (action=action(at)entry=do_abort, fmt=fmt(at)entry=0x7f329e90b6aa "%s\n") at ../sysdeps/posix/libc_fatal.c:155
#5 0x00007f329e7edcfa in malloc_printerr (str=str(at)entry=0x7f329e9092c3 "free(): invalid pointer") at malloc.c:5536
#6 0x00007f329e7ef504 in _int_free (av=<optimized out>, p=<optimized out>, have_lock=0) at malloc.c:4327
#7 0x00007f329e7f1f81 in __GI___libc_free (mem=<optimized out>) at malloc.c:3279
#8 0x00007f329e7dbec5 in __GI__IO_free_backup_area (fp=fp(at)entry=0x561775f126c0) at genops.c:190
#9 0x00007f329e7db6af in _IO_new_file_overflow (f=0x561775f126c0, ch=-1) at fileops.c:758
#10 0x00007f329e7da7be in _IO_new_file_xsputn (n=2, data=<optimized out>, f=<optimized out>) at /usr/src/debug/sys-libs/glibc-2.34-r4/glibc-2.34/libio/libioP.h:947
#11 _IO_new_file_xsputn (f=0x561775f126c0, data=<optimized out>, n=2) at fileops.c:1197
#12 0x00007f329e7cfd32 in __GI__IO_fwrite (buf=0x7ffc90bb0ac0, size=1, count=2, fp=0x561775f126c0) at /usr/src/debug/sys-libs/glibc-2.34-r4/glibc-2.34/libio/libioP.h:947
#13 0x000056177483c758 in flushbuffer (target=0x7ffc90bb0a90) at snprintf.c:310
#14 0x000056177483c4e8 in pg_vfprintf (stream=0x561775f126c0, fmt=0x561774840dec "\n\n", args=0x7ffc90bb0f00) at snprintf.c:259
#15 0x000056177483c5ce in pg_fprintf (stream=0x561775f126c0, fmt=0x561774840dec "\n\n") at snprintf.c:270
#16 0x0000561774831893 in dumpRoleMembership (conn=0x561775f09600, databaseId=0x561775f152d2 "1") at pg_dumpall.c:991
#17 0x0000561774832426 in dumpDatabases (conn=0x561775f09600) at pg_dumpall.c:1332
#18 0x000056177483049e in main (argc=3, argv=0x7ffc90bb1658) at pg_dumpall.c:596

I didn't look in detail, but:

@@ -1323,6 +1327,10 @@ dumpDatabases(PGconn *conn)
exit_nicely(1);
}

+ /* Dump database-specific roles if server is running 15.0 or later */
+ if (server_version >= 150000)
+ dumpRoleMembership(conn, dbid);
+

Isn't that trying print to OPF after the possible fclose(OPF) a bit before and
before it's reopened?