Re: Configure bug under Win9x-Cygwin systems.

[Kevin Ryde]

address@hidden writes:
>
>   case $host_os in
>   msdosdjgpp* | cygwin*)
>     # On DJGPP, this test can blow up pretty badly due to problems in libc
>     # (any single argument exceeding 2000 bytes causes a buffer overrun
>     # during glob expansion).  Even if it were fixed, the result of this
>     # check would be larger than it should be.
>     lt_cv_sys_max_cmd_len=12288;    # 12K is about right
>     ;;

Thanks, I'll add something along those lines.

> Originally, it was saying I had like 365k for command line (under Win9x),
> and 12288 for command line under Win2k.  I do not know how to code into the
> configure a check for Win9x (which is the only thing I am suggesting should
> skip this check).  My hack is certainly not something for the real GMP real
> release, but I am sure there must be some way of doing this skip of the
> check IF a Win9x Cygwin (or probably Mingw) build is taking place.

That test comes from libtool, there's a address@hidden mailing list
for libtool discussions.  Maybe the problem is actually in cygwin or
something, but it'd be good for libtool to do the right thing.

The test looks like it's supposed to stop probing at 512k, if you run
"sh -x configure" you might be able to see if that doesn't work, or if
it goes into slow mode before reaching that size.

> NOTE I tried different config "switches" and was not able to work around
> the listed "bug".

You can override the test using the cache variable, for instance

        ./configure lt_cv_sys_max_cmd_len=20000

> I have tried --enable-static --disable-shared

That should be the default actually.

(Some people have reported problems with a --enable-shared build under
cygwin, apparently related to stdio stuff, but we haven't got to the
bottom of that.)