On Wed, Sep 18, 2024 at 10:51:49AM -0600, Jeff Law wrote:
> On 9/18/24 10:04 AM, Jakub Jelinek wrote:
> > Hi!
> >
> > Similarly to the previous patch, dwarf2asm.cc had
> > HAVE_DESIGNATED_INITIALIZERS support, and as fallback a huge switch.
> > The switch from what I can see is expanded as a jump table with 256
> > label pointers and code at those labels then loads addresses of
> > string literals.
> > The following patch instead uses a table with 256 const char * pointers,
> > NULL for ICE, non-NULL for returning something, similarly to the
> > HAVE_DESIGNATED_INITIALIZERS case.
> >
> > Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
> >
> > 2024-09-18 Jakub Jelinek <ja...@redhat.com>
> >
> > * dwarf2asm.cc (eh_data_format_name): Use constexpr initialization
> > of format_names table for C++14 instead of a large switch.
> Do we want to defer this while the discussion around moving to c++14 plays
> out? Or would you prefer to move forward and if we move to c++14 come back
> and simplify?
I'm fine either way.
Note, the #if HAVE_DESIGNATED_INITIALIZERS parts could be removed too now
that we compile the stuff with C++ (for years). clang++ actually does
support int a[3] = { [2] = 1, [0] = 2 }; but warns about it by default.
Jakub
