On Wed, Feb 24, 2016 at 4:43 PM, Pierre-Marie de Rodat <dero...@adacore.com> wrote: > On 02/24/2016 03:53 PM, Martin Liška wrote: >> >> On 02/24/2016 03:27 PM, Michael Matz wrote: >>> >>> But nothing can set ENABLE_CHECKING anymore (the macro is meanwhile >>> called >>> CHECKING_P), so all that code is dead anyway. So either the new macro >>> should be used or that code should be removed. >> >> >> Good point, well the change is quite recent (12-2015). I'm adding >> the author of the code to make a decision about it. > > > Thanks for the heads up! That’s kind of funny: the check associated with > this dw_loc_frame_offset field revealed a bug to us (at AdaCore) very > recently, so I think we should keep it in one form or another. > > This field takes one int slot in the dw_loc_descr_node structure, so I guess > not having it in release mode is important (that’s what Jason said in > <https://gcc.gnu.org/ml/gcc-patches/2015-11/msg02267.html>). Here’s what I > think: > > * This field is used only in the resolve_args_picking graph traversal > for a consistency check in already visited nodes. > > * resolve_args_picking has a hash set to remember the already visited > nodes. > > * The consistency check itself has almost no runtime cost: we’re > doing the graph traversal in release mode anyway. > > So what about removing the field (in struct dw_loc_descr_node) and replacing > the visited hash set with a frame_offset hash map (in resolve_args_picking)? > This hash map would remember the information we currently store in the > field.
Sounds reasonable. > This is a little change, but I can take care of this if you want. I’m a > little bit desynchronized with the development pace these days: would this > be for stage 4 or GCC 7? Technically it's GCC 7 material. Can you either change the ENABLE_CHECKING use to CHECKING_P for GCC 6 or remove the guarded code/fields? Thanks, Richard. > -- > Pierre-Marie de Rodat
