On Wed, Nov 8, 2017 at 12:27 AM, Steve Kargl <s...@troutmask.apl.washington.edu> wrote: > On Tue, Nov 07, 2017 at 10:38:43PM +0200, Janne Blomqvist wrote: >> On Tue, Nov 7, 2017 at 10:06 PM, Steve Kargl >> <s...@troutmask.apl.washington.edu> wrote: >> > On Tue, Nov 07, 2017 at 08:26:15PM +0200, Janne Blomqvist wrote: >> >> + >> >> + logical_type_node = gfc_get_logical_type (4); >> > >> > Can you add a comment to note that the 4 is purposely chosen? >> > A year or so from now, someone might change this to >> > gfc_default_logical_kind >> > without understand/recalling why 4 is used. >> >> The patch contains a comment explaining that in trans-types.h. > > Well, did say I scanned the patch ;-) > >> However, a bit of further experimenting with my patch revealed that >> one can cause these latent bugs I feared to appear by playing with >> -fdefault-integer-8 / -finteger-4-integer-8 with the example from >> bugzilla (so we end up passing a pointer to a logical(kind=4) variable >> but we tell libgfortran it's kind=8). I did fix the most obvious ones >> in the attached patch (on top of the previous one), but I'm not >> convinced it's exhaustive. >> >> So I wonder, do we actually care enough about -fdefault-integer-8 / >> -finteger-4-integer-8 to do this kind of micro-optimization (i.e. use >> 4 byte logical temporary variables if we can even when we use >> -fdefault-integer-8 / -finteger-4-integer-8), or should we play it >> safe and just use default_logical_type_node everywhere? > > I certainly don't care about fallout when one uses -fdefault-* > options. I've stated more than once I think these should > be removed, because these options break storage association > (among other things). I a little surprise that -finteger-4-integer-8 > have an effect, but I admit I haven't looked closely at the > internal representation of a Fortran logical and its relationship > to INTEGER(4). > > -- > steve
So in the end I decided to play it safe and do the simpler approach. Patch that I committed as r254526 attached. Unless there are objections, I'll apply it to the GCC-7 tree in a few days as well since it's a regression. -- Janne Blomqvist
0001-Introduce-logical_type_node-and-use-it.patch.gz
Description: GNU Zip compressed data
