https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70594
--- Comment #10 from Patrick Palka <ppalka at gcc dot gnu.org> --- (In reply to Jakub Jelinek from comment #8) > (In reply to Nathan Sidwell from comment #7) > > copy_node_stat allocates a new UID for decls. so despite its name > > 'copy_decl_no_change' does cause a change. > > > > That decl never escapes out of the const-expr machinery, but of cause causes > > later decls to have different numberings. (which is what we're observing). > > > > I suppose the constexpr machinery could restore the next_decl_uid (and > > perhaps others?) after copy_fn -- or somewhere in that call chain. But that > > does seem rather icky. > > Yeah, that is not really the right thing. > > > Alternatively copy_decl_no_change could not allocate new UIDs? > > copy_fn is used solely by the constexpr stuff, so the primary question is if > we really need to duplicate any decls for those or not. Changing > copy_decl_no_change is undesirable, it is used for plenty of other cases. > But, to avoid any decl remapping one could just use another function for > that, or could abuse > id->prevent_decl_creation_for_types = true; > id->remapping_type_depth = 1; > etc. The thing is, what is copy_fn copying the fn from (constexpr function > with just parsed body, or parsed and already genericized, or parsed, > genericized and gimplified, or even after some GIMPLE optimizations) and Looks like it's copying the fn after it's been parsed and genericized. > what undesirable effects not copying the decl might have (e.g. during GIMPLE > phases perhaps DECL_VALUE_EXPR could be set for various decls, or their > alignment could be changed, whatever other changes are done) and what > exactly Patrick's patch did (did we in the past always copy_fn just once > each constexpr, but now we do it multiple times, if we GC in between)? Before the patch the function was copied for each constexpr call that's evaluated. After the patch the function is copied only if the freelist is empty (in practice this means after GC or during recursive constexpr calls to the same function).
