cygwin build_alias issue

[Ralf Corsepius]

On Wed, 2004-09-01 at 08:27, Chris Johns wrote:
> Ralf Corsepius wrote:
> > 
> > The symptoms match (aborting on the first "option=value" argument passed
> > to a configure script). 
> > 
> > But ... still nobody seems to have been able to "sh -x" or strace this
> > issue, ... so all this is wild guesses.
> > 
> 
> Agreed. To trace my bug down I had to do as you suggest and use 'set -x' 
> and 'set +x' in various places until I narrowed down the problem. Not 
> nice given the size of configure scripts.
I know, ... but given the dynamical nature of this problem ...

Theoretically, it should be sufficient to add "set +x"/"sh -x"'s the
failing configure script. I.e. once you see a configure script failing,
edit this configure script, "rm -f" the build-tree and re-run the
command.

> > 
> > I still think it's more likely a "resources problem", because the bug
> > doesn't seem to be deterministically. 
> > 
> 
> Given the same command line then yes. If the command lines of the nested 
> configures vary then maybe.
Given the toplevel command line, they normally don't vary.

> I could generate the bug every time on Win98/WinME which was an 
> interesting pattern.

Are you trying to say, you could not reproduce this issue with later
WIN-versions?

I recall having read a remark on in a Cygwin forum/mailing list or
similar, that Win95/Win98 has a hardcoded limit of "40" on the number of
file descriptions somewhere, while later WINs' limits are supposed to be
"unlimited" (Several 10000s).

> > So, I am inclined to think, that it's likely expr returning a bogus
> > return value to be the trigger for these failures, but I doubt this is
> > the origin.

FWIW: Did somebody try the "usual windows resources test", i.e. reboot,
close all apps on the desktop, open a window and run the test, open
another window and reiterate the test, continue until the system
collapses or the test fails in a different way?

> Sure. For the bug I listed I managed to get a small test case so this 
> was not related to resource issues (that I could see).
Great, is this test case portable/applicable to Cygwin? 

Ralf