Sandbox branch for gfortran 4.5 patches

[Tobias Burnus]

Hi all,

regarding the sandbox branch: I'm not sure how useful it is; after all,
we want to apply the patches one-by-one when 4.5 opens and not as a huge
chunk - and splitting them will not be easy. Additionally, one needs to
merge the trunk into the branch regularly, which also not always works.
Thus, I think it does not make much sense to have a branch for all the
little patches. However, one could consider a sandbox branch for the I/O
patches, which allows better (regression)testing and makes sense since
at least 3 developers are involved.

Regarding the 4.4 release: Currently there are 128 "serious regressions"
(P1 to P3) of which 9 are P1, 112 are P2 and 7 have to be triaged (=
P3). Richard estimated yesterday 33% C++ regressions (was ~50% in
earlier releases); tree-optimization is second and there are quite a lot
missed-optimization P2s. Release criterion is no P1 regressions (fixed
or downgraded to P2) and less than 100 P1+P2+P3 regressions. -- For
Fortran: There are 4 regressions (in 4.4) of which 1 is suspended (array
descriptor) and 1 is the I/O performance PR.


Paul Richard Thomas wrote:
> PS - The memory leak patch was actually FX's; I just added some extra
> frills, including a working gfc_current_locus in trans*, and a bug
> fix.
>
> I can remove the extra options with no problem (FX, did you have
> anything ready that used them?).  If it bootstraps and regtests OK,
> should I commit?  After fixing the last round of bit-rot on this
> patch, I do not want to have to do it again!
>   
Have you posted a new patch? I only saw
http://gcc.gnu.org/ml/fortran/2008-11/msg00380.html which had some minor
issues, cf. http://gcc.gnu.org/ml/fortran/2008-11/msg00410.html  What do
you mean with "commit"? 4.4 trunk, 4.5 trunk or the sandbox branch? I
think it is too large for the 4.4 trunk.

Regarding the new options: -fruntest-check=recursion and
-fruntest-check=do are implemented by the reviewed patches
http://gcc.gnu.org/ml/fortran/2008-12/msg00096.html and
http://gcc.gnu.org/ml/fortran/2008-12/msg00104.html

Tobias