zregexparse.ztst fails with latest CVS

zregexparse.ztst fails with latest CVS

[Bart Schaefer]

It goes into an infinite loop getting an I/O error on read, according to
strace.  This is with the standard default dynamic linkage.

I'll try a completely clean rebuild later just in case; out of time now.

-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

Zsh: http://www.zsh.org | PHPerl Project: http://phperl.sourceforge.net   

53completion.ztst (Re: zregexparse.ztst fails with latest CVS)

[Bart Schaefer]

On Feb 6,  8:23pm, Bart Schaefer wrote:
} Subject: zregexparse.ztst fails with latest CVS
}
} It goes into an infinite loop getting an I/O error on read, according to
} strace.  This is with the standard default dynamic linkage.

This is on linux, redhat 5.2; sorry for not mentioning it sooner.

It's not the zregexparse test, it's the completion test.

With ZTST_verbose=1, I get to:

...
../../zsh-3.1.6/Test/52zregexparse.ztst: all tests successful.
Running test: directories and files

and then it hangs forever.  It appears that the zsh started on the zpty
slave has died, leaving zpty in a read loop getting I/O error.

A statically linked zsh passes all tests successfully.

I noticed that "make clean" doesn't remove the module .so files, so I
explicitly blew them away and rebuilt.  Makes no difference.

Tricking comptest into attaching an strace to the zsh on the pty, I find
this (after reams of output):

10480 open("/usr/src/local/zsh/zsh-3.1.6-dynamic/Test/../../zsh-3.1.6/Test/../Completion/Core/_setup", O_RDONLY|O_NOCTTY) = 7
10480 lseek(7, 0, SEEK_END)             = 1941
10480 lseek(7, 0, SEEK_SET)             = 0
10480 read(7, "#autoload\n\nlocal val nm=\"$com"..., 1941) = 1941
10480 close(7)                          = 0
10480 sigprocmask(SIG_BLOCK, [CHLD], [CHLD]) = 0
10480 sigprocmask(SIG_BLOCK, [CHLD], [CHLD]) = 0
10480 sigprocmask(SIG_UNBLOCK, [CHLD], [CHLD]) = 0
10480 sigprocmask(SIG_BLOCK, [CHLD], []) = 0
10480 sigprocmask(SIG_BLOCK, [CHLD], [CHLD]) = 0
10480 sigprocmask(SIG_BLOCK, [CHLD], [CHLD]) = 0
10480 sigprocmask(SIG_UNBLOCK, [CHLD], [CHLD]) = 0
10480 sigprocmask(SIG_BLOCK, [CHLD], []) = 0
10480 --- SIGSEGV (Segmentation fault) ---
10480 +++ killed by SIGSEGV +++

I was able to locate the core file, here's a partial stack trace:

(gdb) where 
#0  0x1d42 in ?? ()
#1  0x808a3a8 in dyn_boot_module (m=0x80d6fa0)
    at ../../zsh-3.1.6/Src/module.c:625
#2  0x808a525 in boot_module (m=0x80d6fa0) at ../../zsh-3.1.6/Src/module.c:677
#3  0x808b263 in load_module (name=0x80de630 "zsh/complist")
    at ../../zsh-3.1.6/Src/module.c:846


-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

Zsh: http://www.zsh.org | PHPerl Project: http://phperl.sourceforge.net   

Re: 53completion.ztst (Re: zregexparse.ztst fails with latest CVS)

[Vin Shelton]

I don't see this on my Mandrake 7.1 system.  uname -a reports:
  Linux zion.rcn.com 2.2.18 #2 SMP Mon Dec 11 21:52:20 EST 2000 i686 unknown

Built from latest CVS sources, with gcc--version ==> 2.95.3

Every test reports 'all tests successful'.

rpm -a glibc ==> glibc-2.1.3-17mdk

in case that helps.

  - vin

"Bart Schaefer" <schaefer@candle.brasslantern.com> writes:
> On Feb 6,  8:23pm, Bart Schaefer wrote:
> } Subject: zregexparse.ztst fails with latest CVS
> }
> } It goes into an infinite loop getting an I/O error on read, according to
> } strace.  This is with the standard default dynamic linkage.
> 
> This is on linux, redhat 5.2; sorry for not mentioning it sooner.
> 
> It's not the zregexparse test, it's the completion test.
> 
> With ZTST_verbose=1, I get to:
> 
> ...
> ../../zsh-3.1.6/Test/52zregexparse.ztst: all tests successful.
> Running test: directories and files
> 
> and then it hangs forever.  It appears that the zsh started on the zpty
> slave has died, leaving zpty in a read loop getting I/O error.
> 
> A statically linked zsh passes all tests successfully.
> 
> I noticed that "make clean" doesn't remove the module .so files, so I
> explicitly blew them away and rebuilt.  Makes no difference.
> 
> Tricking comptest into attaching an strace to the zsh on the pty, I find
> this (after reams of output):
> 
> 10480 open("/usr/src/local/zsh/zsh-3.1.6-dynamic/Test/../../zsh-3.1.6/Test/../Completion/Core/_setup", O_RDONLY|O_NOCTTY) = 7
> 10480 lseek(7, 0, SEEK_END)             = 1941
> 10480 lseek(7, 0, SEEK_SET)             = 0
> 10480 read(7, "#autoload\n\nlocal val nm=\"$com"..., 1941) = 1941
> 10480 close(7)                          = 0
> 10480 sigprocmask(SIG_BLOCK, [CHLD], [CHLD]) = 0
> 10480 sigprocmask(SIG_BLOCK, [CHLD], [CHLD]) = 0
> 10480 sigprocmask(SIG_UNBLOCK, [CHLD], [CHLD]) = 0
> 10480 sigprocmask(SIG_BLOCK, [CHLD], []) = 0
> 10480 sigprocmask(SIG_BLOCK, [CHLD], [CHLD]) = 0
> 10480 sigprocmask(SIG_BLOCK, [CHLD], [CHLD]) = 0
> 10480 sigprocmask(SIG_UNBLOCK, [CHLD], [CHLD]) = 0
> 10480 sigprocmask(SIG_BLOCK, [CHLD], []) = 0
> 10480 --- SIGSEGV (Segmentation fault) ---
> 10480 +++ killed by SIGSEGV +++
> 
> I was able to locate the core file, here's a partial stack trace:
> 
> (gdb) where 
> #0  0x1d42 in ?? ()
> #1  0x808a3a8 in dyn_boot_module (m=0x80d6fa0)
>     at ../../zsh-3.1.6/Src/module.c:625
> #2  0x808a525 in boot_module (m=0x80d6fa0) at ../../zsh-3.1.6/Src/module.c:677
> #3  0x808b263 in load_module (name=0x80de630 "zsh/complist")
>     at ../../zsh-3.1.6/Src/module.c:846
> 
> 
> -- 
> Bart Schaefer                                 Brass Lantern Enterprises
> http://www.well.com/user/barts              http://www.brasslantern.com
> 
> Zsh: http://www.zsh.org | PHPerl Project: http://phperl.sourceforge.net   

RE: 53completion.ztst (Re: zregexparse.ztst fails with latest CVS)

[Andrej Borsenkow]

>
> On Feb 6,  8:23pm, Bart Schaefer wrote:
> } Subject: zregexparse.ztst fails with latest CVS
> }
> } It goes into an infinite loop getting an I/O error on read, according to
> } strace.  This is with the standard default dynamic linkage.
>
> This is on linux, redhat 5.2; sorry for not mentioning it sooner.
>

I cannot reproduce it either on MIPS based Unix. I have CVS + my patch I've
sent at Monday (but it should not affect this anyway). OTOH I cannot run
anything related to zpty for a long time under Cygwin (hoping to look at it
closer). Any chance it may be related?

>
> I noticed that "make clean" doesn't remove the module .so files,

Sorry? You really mean Zsh, do not you? :-)

bor@itsrm2% gmake clean
...
gmake[2]: Entering directory `/u3/u1/build/zsh/Src'
gmake[3]: Entering directory `/u3/u1/build/zsh/Src/Builtins'
rm -f *.o *.export *.so
                         ^^^^
rm -f rlimits.h
rm -f *.o.c *.syms *.pro *.epro *.mdh *.mdhi *.mdhs *.mdh.tmp
gmake[3]: Leaving directory `/u3/u1/build/zsh/Src/Builtins'
gmake[3]: Entering directory `/u3/u1/build/zsh/Src/Modules'
rm -f *.o *.export *.so
                         ^^^^
... etc

-andrej

Re: 53completion.ztst (Re: zregexparse.ztst fails with latest CVS)

[Bart Schaefer]

On Feb 7, 10:50am, Andrej Borsenkow wrote:
} Subject: RE: 53completion.ztst (Re: zregexparse.ztst fails with latest CVS
}
} I cannot reproduce it either on MIPS based Unix. I have CVS + my patch I've
} sent at Monday (but it should not affect this anyway). OTOH I cannot run
} anything related to zpty for a long time under Cygwin (hoping to look at it
} closer). Any chance it may be related?

No, it happens with plain zsh -f too.  Transcript follows; don't let the
"zsh-3.1.9" in path names confuse you, that's my local CVS sandbox name.

(gdb) run -f
Starting program: /scratch/local/src/zsh/zsh-3.1.9-dynamic/Src/zsh -f
zagzig% module_path=($PWD/Test/Modules)
zagzig% setopt extendedglob
zagzig% fpath=(/usr/src/local/zsh/zsh-3.1.9/(Completion|Functions)/*~*/CVS(/)) 
zagzig% autoload -U compinit
zagzig% compinit -u
zagzig% zmodload
zsh/zutil
zsh/compctl
zsh/complete
zsh/main
zsh/zle
zsh/parameter
zagzig% zstyle ":completion:*:default" list-colors "no=<NO>" "fi=<FI>" "di=<DI>" "ln=<LN>" "pi=<PI>" "so=<SO>" "bd=<BD>" "cd=<CD>" "ex=<EX>" "mi=<MI>" "tc=<TC>" "sp=<SP>" "lc=<LC>" "ec=<EC>\n" "rc=<RC>"
zagzig% zstyle ":completion:*" group-name ""
zagzig% zstyle ":completion:*:messages" format "<MESSAGE>%d</MESSAGE>
dquote> "
zagzig% zstyle ":completion:*:descriptions" format "<DESCRIPTION>%d</DESCRIPTION>
dquote> "
zagzig% zstyle ":completion:*:options" verbose yes
zagzig% zstyle ":completion:*:values" verbose yes
zagzig% setopt noalwayslastprompt listrowsfirst completeinword
zagzig% zmodload -i zsh/complist
zsh: failed to load module: zsh/complist

At this point, if I try a second time to load zsh/complist, zsh crashes:

Program received signal SIGSEGV, Segmentation fault.
0x1d42 in ?? ()
(gdb) up
#1  0x808a3a8 in dyn_boot_module (m=0x80d6f88)
    at ../../zsh-3.1.9/Src/module.c:625
625             return fn(m);
(gdb) where 5
#0  0x1d42 in ?? ()
#1  0x808a3a8 in dyn_boot_module (m=0x80d6f88)
    at ../../zsh-3.1.9/Src/module.c:625
#2  0x808a525 in boot_module (m=0x80d6f88) at ../../zsh-3.1.9/Src/module.c:677
#3  0x808b263 in load_module (name=0x80de3e0 "zsh/complist")
    at ../../zsh-3.1.9/Src/module.c:846
#4  0x808b6ad in require_module (nam=0x400e4588 "zmodload", 
    module=0x80de3e0 "zsh/complist", res=1, test=0)
    at ../../zsh-3.1.9/Src/module.c:889
(More stack frames follow...)
(gdb) print fn
$3 = (int (*)()) 0x40165198 <domenuselect+5984>
(gdb) print m
$4 = (struct module *) 0x80d6f88
(gdb) print *m
$5 = {nam = 0x80de3e0 "zsh/complist", flags = 20, u = {handle = 0x80dfc00, 
    linked = 0x80dfc00, alias = 0x80dfc00 ""}, deps = 0x80dd090, wrapper = 0}


} > I noticed that "make clean" doesn't remove the module .so files,
} 
} Sorry? You really mean Zsh, do not you? :-)

Ah, I see now what confused me:  Because "make check" died, it didn't
remove the Test/Modules/ temporary installation directory; I did an
"ls **/*.so" and saw there were some, without noticing where.  The
"mostlyclean-here" rule in Test/Makefile.in doesn't clean those out.

-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

Zsh: http://www.zsh.org | PHPerl Project: http://phperl.sourceforge.net   

Solved Re: 53completion.ztst (Re: zregexparse.ztst fails with latest CVS)

[Bart Schaefer]

It was a module_path thing -- modules from the previous 3.1.9-dev-8 build
were being linked in when zle started up, before comptest had a chance to
reset the module_path.  I changed the version number to something that had
never been installed, so that no modules other than zsh/main were loadable
until after module_path was changed, and now all the tests pass.

I working on a patch to the test scripts to avoid this happening again.

-- 
Bart Schaefer                                 Brass Lantern Enterprises
http://www.well.com/user/barts              http://www.brasslantern.com

Zsh: http://www.zsh.org | PHPerl Project: http://phperl.sourceforge.net