RFC: configuring mt_alloc

[Benjamin Kosnik]

There are three issues remaining with __mt_alloc.

1) configuring mt_alloc
2) dealing with -fno-weak fallout
3) trying to make the no-threads case correct, and the threads case more elegant.

Here's something to think about, that addresses #1, and a partial #2. I
should have more on #2 in a bit.

It moves all the truly constant data into a nested class,
__mt_alloc::tune. Then, getters/setters in __mt_alloc can switch out the
default settings for something else, provided the instantiation has not
been initialized already.

I think this might work. It's roughly based on the configuration ideas
in Yalloc (thanks Felix!)

tested x86/linux

-benjamin

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: p.20040220-1.patch
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20040221/a657c141/attachment.ksh>