It seems we should just remove the use of register?
I have a vague idea that it was once important to say "register" if
you are going to use the variable in an asm snippet that requires it
to be in a register. That might be wrong, or it might be obsolete
even if once true. We could try taking these out and seeing if the
buildfarm complains.
We have several inline asm statements not using register despite using
variables in a register (e.g. pg_atomic_compare_exchange_u32_impl()), so I
wouldn't expect a problem with compilers we support.
Should we make configure test for -Wregister? There's at least one additional
use of register that we'd have to change (pg_regexec).
From a compilation perspective, "register" tells the compiler that you
cannot have a pointer on a variable, i.e. it generates an error if someone
adds something like:
void * p = ®ister_variable;
Removing the "register" declaration means that such protection would be
removed, and creating such a pointer could reduce drastically compiler
optimization opportunities.
--
Fabien.
