> On Sat, Jun 20, 2026 at 3:35 AM Jose E. Marchesi <[email protected]> wrote: >> >> >> Hi Andrea, Kanishka. >> >> > On Fri, Jun 19, 2026 at 2:18 PM Kanishka Solanki >> > <[email protected]> wrote: >> >> >> >> Hi, >> >> >> >> This patch disables strict aliasing in the Algol 68 frontend by setting >> >> flag_strict_aliasing to 0 in a68_post_options(). >> >> >> >> The frontend builds successfully and make check-algol68 passes with this >> >> change. >> > >> > I am not sure that is the correct approach. If a68 does not want TBAA, >> > the front-end should return 0 from LANG_HOOKS_GET_ALIAS_SET. >> >> We need to reflect a bit on whether Algol 68 can benefit from >> optimizations based on strict aliasing, and in fact on whether the >> optimizations are correct with respect to the language's rules. > > Understandable. The reason why I mention turning off via the flag is > not the correct approach in this case is for cross language LTO. > And -fno-strict-aliasing does not turn off TBAA set assignments but > rather short circuits the aliasing code not to compare the sets. So if > the language does not specify aliasing "sets", it should just return 0 > from the language hook. You might want a front-end option which turns > on/off the sets (that is return 0 vs -1 from that language hook).
I actually remember you mentioning this long ago in IRC, but I forgot about it :( Ok, so until we look at this closely I take it it makes sense to turn off the set via the hook. > > Thanks, > Andrea > >> >> > >> > Thanks, >> > Andrea >> > >> >> >> >> Thanks, >> >> Kanishka Solanki >> >> >> >>
