This is the mail archive of the cygwin-developers@cygwin.com mailing list for the Cygwin project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: cygwin 1.3.3 announcment -- extra words solicited


On Tue, Sep 04, 2001 at 10:21:35PM -0400, Charles Wilson wrote:
>egor duda wrote:
>
>
>>CW> Thread ID 1 not known.
>>
>>what does 'info target' show?
>>
>>hmm, and does gdb print any error or warning messages while starting?
>
>
>Okay, after making a little batch file and pointing error_start at that, 
>I can get gdb -nw to start up with state information.
>
>First, I get many many many "Program received signal SIGSEGV, 
>Segmentation fault." messages.  Eventually, I just hit q<RET> to get 
>past those messages.

Yes, this is standard.  It was the best way I could find to have gdb
point at useful info when it attaches to the target.  Hitting 'q' is
the correct response.

>Then, info target shows:
>Symbols from "/usr/bin/make.exe".
>Win32 child process:
>	Using the running image of child thread 1008.0x220.
>	While running this, GDB does not access memory from...
>Local exec file:
>	`/usr/bin/make.exe', file type pei-i386
>	Entry point: 0x401000
>	0x00401000 - 0x0042e200 is .text
>	0x0042f000 - 0x0042fa00 is .data
>	0x00430000 - 0x00434c5c is .bss
>	0x00435000 - 0x00435a00 is .idata
>
>Then I typed 'thread 1' -- and gdb coredumped.  Oops.  Okay, so trying 
>again, (rebuilding cygwin1.dll....wait wait wait, crash!)
>
>This time, (without 'thread 1' -- just 'bt') a simple backtrace shows
>
>...child thread 544.0x2dc
>
>
>#0  0x00410732 in exec_command (argv=0x5, envp=0xa01ca70)
>    at /usr/src/make/src/job.c:2317
>#1  0x61081e8a in read () at 
>/usr/src/cygwin/cygwin/winsup/cygwin/uinfo.cc:284
>#2  0x0040a5cb in func_shell (o=0xa01cd98 "", argv=0x22d52c,

I don't understand this.  This backtrace is saying that read() is in
uinfo.cc at line 284.  That's clearly incorrect.  It sounds like the
symbol table in cygwin1.dll is screwed up.

Something is calling exec_command with an argv of 5, though.
That's what is causing this problem.

Judging by the stack trace, it sounds like the make received a signal and
then maybe something scribbled on the stack.

Either that or gdb is confused.

I can't think of any way to debug this further right now.

>(Chris, I've sent in a digest-unsub/regular-subscribe message pair, if 
>you can approve those quickly we can do this realtime(ish).)

I never saw the subscribe message.  Did you respond to the confirmation?
Anyway, I just went ahead and subscribed you manually.

cgf


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]