Dear Andre, Whilst it is probably OK in most circumstances, I would change: s/rank_remap = ss->dimen < ndim/rank_remap = ss->dimen < ndim != 0
Apart from that, it is indeed OK for trunk, in spite of your expectations :-) Thanks for the patch Paul On 6 July 2015 at 14:58, Andre Vehreschild <ve...@gmx.de> wrote: > Hi all, > > this is a proposal to patch PR 66578 > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66578 . It extends work of Mikael > Morin. The patch fixes two issues: > > 1. a source'd allocate in a block: allocate(c, source=a(:)). The issues occurs > because due to the new handling of source-expressions in trans_allocate() an > array descriptor is created where previously just a plain array was used. > I.e., > GFC_DESCRIPTOR_TYPE_P (source) is true now and GFC_ARRAY_TYPE_P (source) > false, > which made gfortran use the wrong bounds for the descriptor (zero-based > instead > of one-based). This was fixed by Mikael's proposal. > > 2. a two-level array addressing lead to a segfault. I.e., when in a > source-expression an array was used to index another object, then the offset > was computed incorrectly. > > Bootstraps and regtests fine on x86_64-linux-gnu/f21. > > Comments welcome! > > Regards, > Andre > > PS: Experience shows that asking whether this ok for trunk is useless ;-) > There > is always something that could be improved. Open for suggestions. > -- > Andre Vehreschild * Email: vehre ad gmx dot de -- Outside of a dog, a book is a man's best friend. Inside of a dog it's too dark to read. Groucho Marx
