[Patch, Fortran] PR fortran/38887: Fix MVBITS with runtime zero-length array

[Daniel Kraft]

Hi all,

this "trivial" patch fixes PR 38887, where a call to MVBITS aborted at 
runtime if the array was empty (but this not detected at runtime).

This regression was introduced by my wrong-code fix for MVBITS; prior to 
it, elemental dependencies in the arguments to MVBITS were simply 
ignored, leading to that wrong-code bug.  Now, 
gfc_conv_elemental_dependencies is called which in turn called 
internal_unpack for copying the result back after the whole evaluation. 
  internal_unpack however aborted when its argument was empty.

I'm not completely sure why it was done that way, as the "ordinary" 
unpack library function simply returns in this case; my patch changes 
internal_unpack to do so, also.  I believe this is the best solution and 
should not hurt; my guess is that the abort() was introduced first 
simply because it was (probably) not possible at that time that 
internal_unpack could be called with empty arrays.  Now it is, and 
instead of generating runtime-check code for this and call 
internal_unpack only if the array is not empty, this seems to be the 
best solution.  Comments on this welcome though, if someone can point me 
out why the abort() was there!

I did exclude the regenerated files from the patch, hope this is ok. 
Regression-testing on gnu/linux-x86-32 (but there shouldn't be any) at 
the moment.  Ok for trunk if no failures?

Cheers,
Daniel

-- 
Done:  Arc-Bar-Cav-Rog-Sam-Tou-Val-Wiz
To go: Hea-Kni-Mon-Pri-Ran
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch.changelog
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20090121/0615d45b/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch.diff
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20090121/0615d45b/attachment-0001.ksh>