On Fri, Jan 11, 2019 at 07:49:54PM +0000, Richard Sandiford wrote: > FWIW, I think this is papering over a deeper issue,
Most certainly. At the very least there's the fact that many places in the compiler that call attr_min_value (via other functions) don't bother checking for an INT_MAX return. I also didn't try to analyse why bb-reorder.c making wrong decisions due to confused copy_bb_p and similar predicates led to hppa not replacing short branches with longer ones. So there are likely to be some final.c issues too. -- Alan Modra Australia Development Lab, IBM
