Re: -make native exceptions test

[Nicola Pero]

I agree that this all needs fixing.  I haven't had time to look at any  
of the configure stuff yet.  I will before
release, ie in the next week or so.

At a quick look at this, the comments in gnustep-make's configure.ac  
seem to clarify that the check for the
version of GCC was added as a quick hack - the actual check required  
is the one for the uncaught exception
handler --

  # What we want to do: set USE_OBJC_EXCEPTIONS to yes if we can  
compile
  # something with @try/@catch/@finally/@synchronized in it, and it is
  # recent enough (GCC >= 4.4.0) as to be shipped with a working  
uncaught
  # exception handler.

If we move the proper check for the uncaught exception handler from - 
base to -make, then the explicit check
for the version of GCC should go away.  I agree with you that that  
sounds like the right thing to do. :-)

And, it's certainly better to have tests that test whether things work  
or not, as opposed to relying on built-in
knowledge/enumerations of which compiler/runtime supports what (as  
these lists tend to easily go out of date
when new releases of compilers, runtimes or such like appear).  I  
totally agree with you on this.

You can wait for me to do that reorganization, or if you want to do  
it, you could just do it yourself - I'm happy
for you to make these changes to gnustep-make (just please avoid  
inflammatory comments in the source code).

Thanks

On 24 Feb 2011, at 11:54, David Chisnall wrote:

> I'm trying to figure out the native exceptions test in -make, since  
> it's wrong and needs fixing before the next release, but I'm  
> hesitant to touch it because I can't work out what it's actually  
> trying to do.  The correct behaviour should be:
>
> - Test whether the exception syntax works
> - Test whether we have an uncaught exception handler
> - Enable native exceptions if both are true, or native exception  
> support was explicitly disabled by the user (in which case we can  
> skip the tests).
>
> What it seems to do:
>
> - Check we're using GCC, disable exceptions if we aren't (no idea  
> why this test is here)
> - Check whether the exception syntax works
> - Check whether the compiler (not runtime) version is 4.4 or  
> greater, and assume that this means that the uncaught exception  
> handler won't work.
>
> This breaks if:
>
> - You are using the NeXT runtime (supports native exceptions, but  
> the runtime isn't checked, only the compiler)
> - You are using the GNUstep runtime (supports native exceptions)
> - You are using a recent GCC runtime with an old compiler
> - You are using clang (detected as GCC 4.2), but with a recent(ish)  
> runtime
>
> Basically, this complex test only actually works if you are using  
> gcc 4.4 or later, and so could be replaced by a much simpler (though  
> equally wrong) test for gcc 4.4 or later.
>
> The thing that is confusing me, is that we have a test in -base for  
> an uncaught exception handler, which tests the thing that we  
> actually want to test (i.e. can we handle uncaught exceptions),  
> rather than something vaguely related (are we using GCC 4.4 or  
> later).  Why are we not using this in -make?
>
> David
>
> -- Sent from my STANTEC-ZEBRA
> _______________________________________________
> Gnustep-dev mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/gnustep-dev