8.0B2 psql \e

On Mon, Sep 13, 2004 at 12:47:54PM -0700, elein wrote:

Has anyone else experienced a segfault in psql on 8.0B2
when using \e? I can reproduce this consistently.

I do it all the time and I haven't seen it ... is your query too long or
something? Care to show psql's backtrace?

--
Alvaro Herrera (<alvherre[a]dcc.uchile.cl>)
Syntax error: function hell() needs an argument.
Please choose what hell you want to involve.

On Mon, Sep 13, 2004 at 12:47:54PM -0700, elein wrote:

Has anyone else experienced a segfault in psql on 8.0B2
when using \e? I can reproduce this consistently.

If not I'll trace it down if I can, but if it is a known
problem, I won't bother.

I'm using the latest CVS sources on Solaris 9 and can't reproduce
this problem. The segfault was identified in 8.0.0beta1 a month
ago but I thought it was fixed.

http://archives.postgresql.org/pgsql-bugs/2004-08/msg00103.php

--
Michael Fuhr
http://www.fuhr.org/~mfuhr/

elein <elein@varlena.com> writes:

Has anyone else experienced a segfault in psql on 8.0B2
when using \e? I can reproduce this consistently.

Hmm, we fixed a bug in \e not long ago, but it was pre-beta2 ...

regards, tom lane

This looks like the same bug as Michael's
http://archives.postgresql.org/pgsql-bugs/2004-08/msg00103.php

trim_directory needs to check for a NULL pointer in path.

if ( path == NULL || path[0] == '\0' )

I'll verify I have the version I think I have...

(gdb) bt
#0 0x0805dd15 in trim_trailing_separator (path=0x0) at path.c:442
#1 0x0805d6b2 in canonicalize_path (path=0x0) at path.c:123
#2 0x0804af80 in exec_command (cmd=0x808f958 "e", scan_state=0x808ba30, query_buf=0x808bb80)
at command.c:378
#3 0x0804a964 in HandleSlashCmds (scan_state=0x808ba30, query_buf=0x808bb80) at command.c:86
#4 0x0804fd31 in MainLoop (source=0x40217a60) at mainloop.c:257
#5 0x080517b2 in main (argc=2, argv=0xbffff1f4) at startup.c:310
#6 0x401177ee in __libc_start_main () from /lib/libc.so.6

419 static void
420 trim_directory(char *path)
421 {
422 char *p;
423
424 if (path[0] == '\0')
425 return;
426
427 for (p = path + strlen(path) - 1; IS_DIR_SEP(*p) && p > path; p--)
428 ;
429 for (; !IS_DIR_SEP(*p) && p > path; p--)
430 ;
431 *p = '\0';
---Type <return> to continue, or q <return> to quit---q
Quit
(gdb) print path[0]
Cannot access memory at address 0x0

This is the source for beta2. Perhap the previous
fix was to add the path[0] check but the path == NULL
check was forgotten?

OR, is it a problem if path==NULL?

--elein

elein <elein@varlena.com> writes:

I'll verify I have the version I think I have...

(gdb) bt
#0 0x0805dd15 in trim_trailing_separator (path=0x0) at path.c:442
#1 0x0805d6b2 in canonicalize_path (path=0x0) at path.c:123
#2 0x0804af80 in exec_command (cmd=0x808f958 "e", scan_state=0x808ba30, query_buf=0x808bb80)
at command.c:378

You seem to be missing this patch:

http://developer.postgresql.org/cvsweb.cgi/pgsql-server/src/bin/psql/command.c.diff?r1=1.122;r2=1.123

regards, tom lane