Re: [Patch, target] Sort out some issues in config{,/*}/darwin*.h

[IainS <developer at sandoe-acoustics dot co dot uk>]

On 14 Apr 2010, at 18:44, Peter O'Gorman wrote:
> >    o macro-ized DSYMUTIL so that we can change it without editing  
> > the specs.
> >    o Provided a wrapper for the system-supplied dsymutil that  
> > allows us to edit out bogus warnings.
> >        this is installed in $(libexecdir) . The initial version  
> > filters the message that gives us problems
> >            like PR43254.
>
> I think this is not such a bad idea. People building linux -> darwin
> cross compilers (At least, I think they exist) currently may have to
> edit the specs so that dsymutil is not called, because it is not an  
> open
> source project at Apple.

well.. they would have to provide a dummy dsymutil or some configury  
stuff to edit that out of the script.
(but that's a pretty easy thing to do).

> Since your wrapper always exits successfully
> they shouldn't need to do that anymore... Are you sure that the  
> wrapper
> should always exit successfully?

tricky one that...
the problem is detecting a case which is should fail but is not the  
case we wish to edit
	specifically, there could be more than one error message so editing  
ours out is not sufficient proof there are no more.

I reconciled this internally by saying to myself :
A. this is the last phase - it will not mislead the User into thinking  
the compile succeeded if it didn't.
B. Any other messages from dsymutil will still be presented - and  
therefore cause testsuite fails - or alert the end user.

> >        (a) Always runs when we are generating an exe - IMO this is  
> > more intuitive to the end user - and
> >             it side-steps the problem with the existing spec.
>
> I don't actually agree with this, but don't feel that strongly about  
> it
> either.
>
> >    o Filter out -lm and re-apply it where needed.  [I've provided a  
> > bolt-hole for anyone who wants to
> >       force -lm (-force_lm) at least until we confirm that this is  
> > not an issue.]
>
> There are multpile libraries that are just symlinks to libSystem  
> (libc,
> libinfo, libpthread, libdl, for example). It seems odd to treat libm
> specially here. If the original problem was a test suite failure, why
> not fix the test suite to not add -lm on darwin? Sorry, I know this  
> will
> drive you nuts, but I don't get from the PR why you chose this
> workaround. If memory serves correctly libm has always been a  
> symlink to
> libSystem (or the System framework in 10.0). There is no need to treat
> different darwin versions differently here.

Well, I think that compiler cognoscenti will easily understand why we  
run dsymutil in the case of temp files .. .
... but it ain't so obvious to a non-compiler-thinking end user why he  
doesn't get a dSYM file from his make output.

> >    o fixed darwin10 not to provide no_compact_unwind when building  
> > for <10.6.
> Was this actually causing a problem?

I don't think we have metrics on this one way or the other - I am  
merely preserving the code-generation status quo.
If one is targeting 10.4 from a 10.6 system - it could conceivably be  
an issue.

> Sorry, I didn't spend much time looking at the PRs, I will look more
> closely this evening.

thanks, all input is appreciated.

Iain