[Debian GNUstep maintainers] Bug#453242: gnustep-base: FTBFS: checking if ffcall trampolines work... no

[Yavor Doganov]

On Thu, Nov 29, 2007 at 12:50:56PM +0100, Lucas Nussbaum wrote:
> On 28/11/07 at 23:51 -0500, Hubert Chathi wrote:
> 
> However, the test probably tries to *run* the code, not just compile
> it. 

That's right, it is run via the AC_RUN_IFELSE macro, where the
condition for cross-compiling is set to "have_working_trampoline=yes"
-- but your configuration is not cross-compiling so the test program
is being run.

I don't understand the code, but AFAICS it cannot return success in
that case.  Is there are a reliable way to detect the host cpu in such
an environment?

> > i.e. if it only fails in that configuration, should this bug be
> > "serious" or "important"?

IMHO, Hubert, this is certainly not a release-critical bug.

> However, a better question is probably "do we want it fix or not?".

It is worth fixing if GNUstep upstream (or ffcall upstream) consider
it a bug.  Trampolines are an evil art.  If this environment can be
detected either by gnustep-base's configure or the trampoline test, it
is the way to go.  Failing that, the remaining solutions are just "to
make it build", which is easy -- but not an actual bugfix.

> Note that 1.13.0-7 built fine, too.

Certainly, IIRC upstream indroduced this trampiline configure test in
0.14 precisely because they were receiving many bogus bug reports from
Solaris/HP-UX users.  GNUstep has always heavily relied on ffcall/ffi
so when that's not working the behaviour of the whole framework is
unpredictable, and it's very hard to figure out what's going on. 

So from their (upstream's) point of view, I believe that's a good
change.