On Fri, 26 Apr 2019, Richard Biener wrote: > Can you move the GIMPLE/RTL FE specific data in c_declspecs to > a substructure accessed via indirection? I guess enlarging that > isn't really what we should do. You'd move gimple_or_rtl_pass > there and make that pointer one to a struct aux_fe_data > (lifetime managed by the respective RTL/GIMPLE FE, thus > to be freed there)? Joseph, do you agree or is adding more > stuff to c_declspecs OK (I would guess it could be a few more > elements in the future).
c_declspecs is shortlived (on the parser obstack, freed after each external declaration is parsed). I don't think its size is particularly important (this is however a guess, not based on any actual measurements of cc1 memory usage). -- Joseph S. Myers jos...@codesourcery.com
