Re: Handle -o w/ and w/o $EXEEXT during manifest embedding

[Ralf Wildenhues]

* Peter Rosin wrote on Sun, Aug 31, 2008 at 09:11:51AM CEST:
> Den 2008-08-31 07:29, skrev Ralf Wildenhues:
>>
>> thank you for starting a new thread on this.  The other, monster thread,
>> scares me a bit too much for wanting to dig through it.
>
> Should I repost the pending patches as new fresh individual
> messages?

That would certainly help.  Even cooler would be if you posted a link to
the post of the patch within the thread too, so I could read myself up
from there.  (When reviewing, I try to at least skim the discussion
pertaining to a thread, in order to check that the discussed issues have
been addressed.)  Up to you if you want to take the effort though.

>> * Peter Rosin wrote on Sat, Aug 30, 2008 at 10:26:45PM CEST:
>>> During my exercise with libsndfile I found a problem with
>>> the manifest embedding code. $output may or may not contain
>>> the trailing $EXEEXT, so the code in $postlink_cmds has
>>> to handle both cases (or some c14n has to be added to ltmain).
>>
>> Is this really a problem with the libtool variable exeext not being set
>> correctly, or with the libtool script being passed '-o prog' vs. '-o
>> prog.exe'?
>
> The patch solves the latter, i.e. '-o prog' vs. '-o prog.exe'. I
> have not seen any problem with exeext not being correct.

Hmm, ok.

>> If it's a problem with exeext not being set correctly, then shouldn't we
>> fix that?  Markus Duft posted a patch for that on this list before,
>> which I've never found the time to evaluate.
>>
>> Also, this needs testsuite exposure if not already done.
>
> By "this", I assume you mean a check if both '-o prog' and
> "-o prog$EXEEXT" work?

Yes.  At least if such a test is easily possible.

> Should that test do what is needed to
> expose a failure to embed the manifest? I.e. do you want the
> test to install both progs and then try to run them?

If a runtime test is the only way to find out if things went alright,
then I guess that would be best.  In this case the test should ensure
to SKIP for cross compilation.  It's also OK to SKIP the test generally
when CL is not used, I guess.

Cheers,
Ralf