newlib/winsup/cygwin/how-to-debug-cygwin.txt

Copyright 2001 Red Hat Inc., Egor Duda

So, your favorite program has crashed? And did you say something about
'stackdump'? Or it just prints its output from left to right and upside-down?
Well, you can file an angry bug report and wait until some of the core
developers try to reproduce your problem, try to find what's the matter
with your program and cygwin and fix the bug, if any. But you can do something
better than that. You can debug the problem yourself, and even if you can't
fix it, your analysis may be very helpful. Here's the (incomplete) howto on
cygwin debugging.

1. The first thing you'll need to do is to build cygwin1.dll and your crashed
application from sources. To debug them you'll need debug information, which
is normally stripped from executables.

2. Create known-working cygwin debugging environment.
- create a separate directory, say, c:\cygdeb, and put known-working
  cygwin1.dll and gdb.exe in it.
- create a wrapper c:\cygdeb\debug_wrapper.cmd:

========= debug_wrapper.cmd =========
rem setting CYGWIN_TESTING environment variable makes cygwin application
rem not to interfere with other already running cygwin applications.
set CYGWIN_TESTING=1
c:\cygdeb\gdb.exe -nw %1 %2
===================================

3. Try to use cygwin's JIT debugging facility:
- add 'error_start=c:\cygdeb\debug_wrapper.cmd' to CYGWIN environment
  variable. When some application encounters critical error, cygwin will stop
  it and execute debug_wrapper.cmd, which will run gdb and make it to attach to
  the crashed application.

4. Strace.
  You can run your program under 'strace' utility, described if user's manual.
  If you know where the problem approximately is, you can add a bunch of
  additional debug_printf()s in the source code and see what they print in
  strace log. There's one common problem with this method, that some bugs
  may mysteriously disappear once the program is run under strace. Then the
  bug is likely a race condition. strace has two useful options to deal with
  such situation: -b enables buffering of output and reduces additional
  timeouts introduced by strace, and -m option allows you to mask certain
  classes of *_printf() functions, reducing timeouts even more.

5. Problems at early startup.
  Sometimes, something crashes at the very early stages of application
  initialization, when JIT debugging facility is not yet active. Ok, there's
  another environment variable that may help. Create program_wrapper.cmd:

========= program_wrapper.cmd =========
rem setting CYGWIN_SLEEP environment variable makes cygwin application
rem to sleep for x milliseconds at startup
set CYGWIN_SLEEP=20000
c:\some\path\bad_program.exe some parameters
===================================
 
  Now, run program_wrapper.cmd. It should print running program pid.
  After starting program_wrapper.cmd you've got 20 seconds to open another
  window, cd to c:\cygdeb in it, run gdb there and in gdb prompt type

  (gdb) attach <pid>

  where <pid> is the pid that program_wrapper.cmd have printed.
  After that you can normally step through the code in cygwin1.dll and
  bad_program.exe

6. Heap corruption.
  If your program crashes at malloc() or free() or when it references some
  malloc()'ed memory, it looks like heap corruption. You can configure and
  build special version of cygwin1.dll which includes heap sanity checking.
  To do it, just add --enable-malloc-debugging option to configure. Be warned,
  however, that this version of dll is _very_ slow (10-100 times slower than
  normal), so use it only when absolutely necessary.

7. Program dies when running under strace.
  If your program crashes when you run it using strace but runs ok (or has a
  different problem) otherwise, then there may be a problem in one of the
  strace *_printf statements.  Usually this is caused by a change in arguments
  resulting in a %s being used with something other than a pointer to a
  string.

  To debug this scenario, do something like this:

   bash$ gdb -nw yourapp.exe
   (gdb) dll cygwin1
   (gdb) l dll_crt0_1
   (gdb) bp <<first line in the function>>
   (gdb) run
   (gdb) set strace.active=1
   (gdb) continue

  The program will then run in "strace mode", calling each strace *_printf,
  just like it does when run under the strace program.  Eventually, the
  program will crash, probably in small_printf.  At that point, a 'bt'
  command should show you the offending call to strace_printf with the
  improper format string.