I know CVS is hosed (this time for sure)

[Charles Wilson]

> I told Chuck in private email that I was *just* able to duplicate his
> make problem exactly once.  This happened, of course, just as I had
> uploaded 1.3.3 to sources.redhat.com.
> 
> The program segv'ed while building the winsup directory.  Since cygwin
> was stripped, I couldn't easily see where it had died.
> 
> I haven't been able to get it to fail ever since.
> 
> So, I think I'm going to release it as is.  If there is a bug, it
> seems to be extremely sporadic.
> 
> I will wait to see if Chuck can duplicate any problems first, though.


So, after building cygwin (-g -O2 -fnative-struct) and installing it, 
AND building cygwin (-g -O2 -fnative-struct) with the NEW kernel 
running, I then stripped the active DLL, (cp cygwin1.dll to tmp.dll, 
strip, exit all cygwin, command.com->copy tmp.dll cygwin1.dll)

And rebuilt cygwin (-g -O2 -fnative-struct) AGAIN.

The first time (with unstripped kernel in use) --> no error.  This time 
(with stripped kernel in use) --> error.  Sigh.

I've attached the backtrace and everything else I could think of (no 
rhyme or reason -- I just went thru my email and tried to do all of the 
gdb things that had been asked of me before.)

The good news is, after quitting gdb, the build continued successfully. 
  I *think* the error is happening when a sub-process of make is just 
terminating because it is finished in a certain directory.  It "Leaves 
directory X" and coredumps. But, this is perfect stopping place.  When 
when gdb returns (successfully) the parent make picks right up and 
continues, unaware that anything bad happened -- because the child make 
was terminating anyway!  (Without JIT, the child make would return an 
error code after stackdumping and the parent make quits; With JIT, gdb 
returns success and the parent make continues!)

Anyway, after that build finished, I tried to build again (after 
cleaning out my build directory)...it's still running; I'll report back 
later.  (Perhaps I too can join Chris in the "it failed ONCE" category).

To tell you the truth, I think 1.3.3 ought to go out regardless.  It 
aint perfect, but this bug seems to be tickled quite rarely--after the 
recent (semi)fixes--and there's lots of improvements/fixes in the 
current CVS begging for a release.  (And don't forget, no errors at all 
if the running kernel has symbols.  I'm not sure why that would be, but...)

--Chuck