Re: RFA: More demangler patches, with fixes for gcc/13244 and gcc/13304

[Ian Lance Taylor <ian at wasabisystems dot com>]

DJ Delorie <dj@redhat.com> writes:

> > +/* We avoid pulling in the ctype tables, to prevent pulling in
> > +   additional unresolved symbols when this code is used in a library.
> > +   FIXME: Is this really a valid reason?  This comes from the original
> > +   V3 demangler code.
> 
> I think the original problem was that the old demangler used to be
> copied into libstdc++.a and thus dependencies on other parts of
> libiberty were a no-no.  This is no longer the case, so I expect that
> any libiberty module should be free to use other libiberty modules as
> appropriate.

I think you are right.

Tom Tromey indicated that there might be some use for the new
demangler somewhere in the Java libraries, where they need a demangler
which doesn't call libstdc++ routines.  Tom, is that still true?  How
would you feel about a demangler which used the safe-ctype routines
from libiberty?  For that matter, how would you feel about a demangler
which used the floatformat routines, even though those would never be
useful for Java and only rarely for C++?

Ian