On Wed, Aug 13, 2014 at 6:31 AM, Jeff Law <l...@redhat.com> wrote: > On 08/13/14 06:12, Felix Yang wrote: >> >> Hi all, >> >> The qsort library function may have different behavior on >> different hosts (say Linux vs MinGW). >> We may have different sorting results with qsort when there are >> elements with the same key value. >> GCC uses qsort a lot. And the output of certain optimizations, >> such as IRA, relies on the sorting result of this library function. >> The problem is that we may have different assembly code of GCC on >> different hosts even with the same source file. >> Normally this is not what a GCC user expect to see. >> In order to fix this issue, I am adding Berkeley qsort to >> libiberty in order to override the one from the library. >> >> Bootstrapped on x86_64-suse-linux, OK for trunk? Please help >> commit this patch if it's OK. > > Or we find the cases where the sort comparison routine can be tweaked to > make it stable which makes this class of problems go away.
I tend to agree. When we want a fully specified sort, it seems to me we should use a fully specified comparison routine. Ian
