This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: Getting Apple's libstdc++ debug mode into the FSF tree
- From: snyder <snyder at fnal dot gov>
- To: Matt Austern <austern at apple dot com>
- Cc: Nathan Myers <ncm at cantrip dot org>, libstdc++ at gcc dot gnu dot org
- Date: 17 Jul 2003 12:33:30 -0500
- Subject: Re: Getting Apple's libstdc++ debug mode into the FSF tree
- References: <51DA6DFF-B81B-11D7-BE8A-000393B2ABA2@apple.com>
>>>>> "Matt" == Matt Austern <austern@apple.com> writes:
Matt> This seems to me like a big red flag. Do you know whether
Matt> your unreferenced-types flag applies to STABS as well as to
Matt> Dwarf?
The patch i submitted was only for dwarf; it works by walking the
trees built in dwarf2out.
Mark mentioned at the time i first submitted it that it would be
better to do the unreferenced type suppression higher up, at the
tree level, so that this stuff would be independent of the debugging
format. I agree, but that's a much more difficult project if you're
not intimately familiar with the compiler --- at least if you're
munging structures used only in dwarf2out, you can be pretty sure
you're not going to break some other part of the compiler.
Matt> I'd like to do some tests. Some of Apple's customers are
Matt> very concerned about the sizes of executables when debugging
Matt> is turned on. (And if you think those concerns are silly,
Matt> you should realize that in some cases we're talking about
Matt> sizes in the half-gigabyte range.)
That's a big concern for us too. With gcc 3.1, with my original
unreferenced-types removal patch (which was more aggressive than
what eventually made it into 3.4), the directory of .a files from our
weekly builds is up to about 1.3G. Binary sizes in the 100's of MB
are common, and links may need ~1GB of virtual memory.
sss