Crash with =~ in script, but not interactively

Crash with =~ in script, but not interactively

[Mikael Magnusson]

I was going to make a wrapper script that replaces the :43: in gcc
warnings with +43 so i can open in vim easierly, but ran into some
problems.

#!/bin/zsh
#this makes zsh segfault
if [[ $#@ = 1 ]]; then
	if [[ $1 =~ /?[^/]+:[0-9]+:$ ]]; then
		echo hello
	fi
fi

If i save that in a file and run 'zsh file' it crashes, but if i run
'source file' in an open shell it works as expected. Here's a
backtrace with -g3 -O0

Program received signal SIGSEGV, Segmentation fault.
0x080d0d16 in taddstr (s=0xc93fa008 <Address 0xc93fa008 out of
bounds>) at text.c:68
68	    int sl = strlen(s);
(gdb) bt
#0  0x080d0d16 in taddstr (s=0xc93fa008 <Address 0xc93fa008 out of
bounds>) at text.c:68
#1  0x080d1ce2 in gettext2 (state=0xafb7dc3c) at text.c:538
#2  0x080d1008 in getjobtext (prog=0xa7f11480, c=0xa7f114bc) at text.c:172
#3  0x0807656b in execpline2 (state=0xafb7e22c, pcode=259, how=18,
input=0, output=0, last1=0)
    at exec.c:1481
#4  0x0807597b in execpline (state=0xafb7e22c, slcode=26626, how=18,
last1=0) at exec.c:1275
#5  0x080751df in execlist (state=0xafb7e22c, dont_change_job=0,
exiting=0) at exec.c:1081
#6  0x08074dee in execode (p=0xa7f11480, dont_change_job=0, exiting=0)
at exec.c:939
#7  0x0808ddb6 in loop (toplevel=1, justonce=0) at init.c:181
#8  0x08090f15 in zsh_main (argc=2, argv=0xafb7e384) at init.c:1352
#9  0x0805ead6 in main (argc=Cannot access memory at address 0xffffffff
) at ./main.c:93
(gdb) bt full
#0  0x080d0d16 in taddstr (s=0xc93fa008 <Address 0xc93fa008 out of
bounds>) at text.c:68
	sl = -918577144
	c = 0 '\0'
#1  0x080d1ce2 in gettext2 (state=0xafb7dc3c) at text.c:538
	end = (Wordcode) 0xaa0b5378
	s = (Tstack) 0x0
	n = (Tstack) 0x819bd30
	stack = 0
	code = 959262863
	c1 = {0x815d5ab "=", 0x815d5ad "!=", 0x815d5b0 "<", 0x815d5b2 ">",
0x815d5b4 "-nt",
  0x815d5b8 "-ot", 0x815d5bc "-ef", 0x815d5c0 "-eq", 0x815d5c4 "-ne",
0x815d5c8 "-lt",
  0x815d5cc "-gt", 0x815d5d0 "-le", 0x815d5d4 "-ge", 0x815d5d8 "=~"}
#2  0x080d1008 in getjobtext (prog=0xa7f11480, c=0xa7f114bc) at text.c:172
	s = {prog = 0xa7f11480, pc = 0xa7f11534,
  strs = 0xa7f11524 "/\224\217\206/\220+:\2170-9\220+:\205"}
	jbuf = "if [[ \205\204@ = 1 ]]; then; if [[ \2051 =~
/\224\217\206/\220+:\2170-9\220+:\205 ]]; then; echo hello; \005rE "
#3  0x0807656b in execpline2 (state=0xafb7e22c, pcode=259, how=18,
input=0, output=0, last1=0)
    at exec.c:1481
	pid = 134828847
	pipes = {2, -1346904904}
#4  0x0807597b in execpline (state=0xafb7e22c, slcode=26626, how=18,
last1=0) at exec.c:1275
	ipipe = {0, 0}
	opipe = {0, 0}
	pj = 0
	newjob = 1
	old_simple_pline = 0
	slflags = 0
	code = 259
---Type <return> to continue, or q <return> to quit---
	lastwj = 0
	lpforked = 0
#5  0x080751df in execlist (state=0xafb7e22c, dont_change_job=0,
exiting=0) at exec.c:1081
	donedebug = 0
	next = (Wordcode) 0xa7f11520
	code = 26626
	ret = -1346903576
	cj = 0
	csp = 0
	ltype = 18
	old_pline_level = 0
	old_list_pipe = 0
	oldlineno = 8
	oldnoerrexit = 0
	donetrap = 0
#6  0x08074dee in execode (p=0xa7f11480, dont_change_job=0, exiting=0)
at exec.c:939
	s = {prog = 0xa7f11480, pc = 0xa7f114bc,
  strs = 0xa7f11524 "/\224\217\206/\220+:\2170-9\220+:\205"}
#7  0x0808ddb6 in loop (toplevel=1, justonce=0) at init.c:181
	toksav = 1
	prog = (Eprog) 0xa7f11480
#8  0x08090f15 in zsh_main (argc=2, argv=0xafb7e384) at init.c:1352
	t = (char **) 0xafb7e38c
	t0 = 157
#9  0x0805ead6 in main (argc=Cannot access memory at address 0xffffffff
) at ./main.c:93
No locals.

Using revision 'Based on 24334: Documentation for run-help
improvements.' which is currently the latest.

-- 
Mikael Magnusson

Re: Crash with =~ in script, but not interactively

[Mikael Magnusson]

On 04/01/2008, Mikael Magnusson <mikachu@gmail.com> wrote:
> I was going to make a wrapper script that replaces the :43: in gcc
> warnings with +43 so i can open in vim easierly, but ran into some
> problems.
>
> #!/bin/zsh
> #this makes zsh segfault
> if [[ $#@ = 1 ]]; then
>         if [[ $1 =~ /?[^/]+:[0-9]+:$ ]]; then
>                 echo hello
>         fi
> fi
>
> If i save that in a file and run 'zsh file' it crashes, but if i run
> 'source file' in an open shell it works as expected. Here's a
> backtrace with -g3 -O0

Just checked, this happens directly on the commit "23375: Phil
Pennock: =~, zsh/regex etc. etc.", so the bug wasn't introduced after
that. Likely the bug is somewhere else, but I can't reproduce it with
just a = instead of the =~.

-- 
Mikael Magnusson

Re: Crash with =~ in script, but not interactively

[Peter Stephenson]

On Fri, 4 Jan 2008 10:27:40 +0100
"Mikael Magnusson" <mikachu@gmail.com> wrote:
> I was going to make a wrapper script that replaces the :43: in gcc
> warnings with +43 so i can open in vim easierly, but ran into some
> problems.
> 
> #!/bin/zsh
> #this makes zsh segfault
> if [[ $#@ = 1 ]]; then
> 	if [[ $1 =~ /?[^/]+:[0-9]+:$ ]]; then
> 		echo hello
> 	fi
> fi
> 
> If i save that in a file and run 'zsh file' it crashes, but if i run
> 'source file' in an open shell it works as expected.

Thanks, the problem was fairly obvious (luckily, since debugging the
wordcode is another thing that's a nightmare).

Index: Src/text.c
===================================================================
RCS file: /cvsroot/zsh/zsh/Src/text.c,v
retrieving revision 1.20
diff -u -r1.20 text.c
--- Src/text.c	1 May 2007 22:05:06 -0000	1.20
+++ Src/text.c	4 Jan 2008 14:28:51 -0000
@@ -732,7 +732,8 @@
 			    taddstr(" ");
 			    taddstr(ecgetstr(state, EC_NODUP, NULL));
 			    if (ctype == COND_STREQ ||
-				ctype == COND_STRNEQ)
+				ctype == COND_STRNEQ ||
+				ctype == COND_REGEX)
 				state->pc++;
 			} else {
 			    /* Unary test: `-f foo' etc. */ 
Index: Test/C02cond.ztst
===================================================================
RCS file: /cvsroot/zsh/zsh/Test/C02cond.ztst,v
retrieving revision 1.19
diff -u -r1.19 C02cond.ztst
--- Test/C02cond.ztst	13 Dec 2007 21:21:29 -0000	1.19
+++ Test/C02cond.ztst	4 Jan 2008 14:28:51 -0000
@@ -211,6 +211,18 @@
 0:-nt shouldn't abort on non-existent files
 >status = 1
 
+# core dumps on failure
+  if zmodload -i zsh/regex 2>/dev/null; then
+     echo >regex_test.sh 'if [[ $# = 1 ]]; then
+	if [[ $1 =~ /?[^/]+:[0-9]+:$ ]]; then
+	  :
+	fi
+      fi
+      exit 0'
+      $ZTST_testdir/../Src/zsh -f ./regex_test.sh
+  fi
+0:regex tests shouldn't crash
+
 %clean
   # This works around a bug in rm -f in some versions of Cygwin
   chmod 644 unmodish



-- 
Peter Stephenson <pws@csr.com>                  Software Engineer
CSR PLC, Churchill House, Cambridge Business Park, Cowley Road
Cambridge, CB4 0WZ, UK                          Tel: +44 (0)1223 692070

Re: Crash with =~ in script, but not interactively

[Mikael Magnusson]

On 04/01/2008, Peter Stephenson <pws@csr.com> wrote:
> On Fri, 4 Jan 2008 10:27:40 +0100
> "Mikael Magnusson" <mikachu@gmail.com> wrote:
> > I was going to make a wrapper script that replaces the :43: in gcc
> > warnings with +43 so i can open in vim easierly, but ran into some
> > problems.
> >
> > #!/bin/zsh
> > #this makes zsh segfault
> > if [[ $#@ = 1 ]]; then
> >       if [[ $1 =~ /?[^/]+:[0-9]+:$ ]]; then
> >               echo hello
> >       fi
> > fi
> >
> > If i save that in a file and run 'zsh file' it crashes, but if i run
> > 'source file' in an open shell it works as expected.
>
> Thanks, the problem was fairly obvious (luckily, since debugging the
> wordcode is another thing that's a nightmare).

Yep that worked. Does the =~ operator support backreferences? I didn't
find anything searching through the manual for the word. This is what
I ended up with instead which works too:

#!/bin/zsh
setopt extendedglob
if [[ $# = 1 ]]; then
	if [[ $1 = *[^/]##:(#b)([0-9]##)(#B): ]]; then
		2=+${1[$mbegin[1],$mend[1]]}
		1=${1[1,$mbegin[1]-2]}
	fi
fi
if [[ -z $DISPLAY ]] ; then vim $@; else; gvim $@; fi

in case anyone ever wanted something like that.

-- 
Mikael Magnusson

Re: Crash with =~ in script, but not interactively

[Peter Stephenson]

"Mikael Magnusson" wrote:
> Yep that worked. Does the =~ operator support backreferences? I didn't
> find anything searching through the manual for the word. This is what
> I ended up with instead which works too:

It uses $MATCH and $match in the same way (no flags required), unless
the option BASH_REMATCH is set in which case it uses the array
$BASH_REMATCH.  See the description of =~ in zshmisc.

-- 
Peter Stephenson <pws@csr.com>                  Software Engineer
CSR PLC, Churchill House, Cambridge Business Park, Cowley Road
Cambridge, CB4 0WZ, UK                          Tel: +44 (0)1223 692070