Re: PATCH C++: speed up templates

[Mark Mitchell <mark at codesourcery dot com>]

--On Friday, February 21, 2003 11:36:34 AM -0800 Mike Stump 
<mstump at apple dot com> wrote:

> On Thursday, February 20, 2003, at 06:11 PM, Mark Mitchell wrote:
> > --On Thursday, February 20, 2003 05:59:11 PM -0800 Ziemowit Laski
> > <zlaski at apple dot com> wrote:
> >
> > > So, can you point me to a date or release tag where the bug is
> > > reproducible?
> >
> > Try the test case in the PR 8338.
> >
> > You're going to have a very hard time getting me to revert this patch.
> >  If you want to make me, I can come up with a valid test case that
> > will take *a long time* to run without the hash table.  The hash table
> > fixes the big-O complexity of the function.
>
> I've already submitted a patch to solve that problem:
>
> http://gcc.gnu.org/ml/gcc-patches/2003-02/msg01245.html
>
> It needs cleaning up per your instructions.  I'll see about turning it
> around today.
>
> With that patch in, then another patch, see below, can be used to turn
> off the hash tables, and remove the regression what was put into the
> compiler, while preserving the goodness of the walk.  Does this part look
> ok for 3.3, pending the other work?

Why is OK not to avoid circularity if fn is 0?

Does the original test case in the PR still work with your patch?

If those questions can be answered satisfactorily, then your patch is fine. 
I'm just nervous about taking out the hash table, since there are 
definitely places where we have circularities in the tree structure, and 
you can now get burned by that.

--
Mark Mitchell                mark at codesourcery dot com
CodeSourcery, LLC            http://www.codesourcery.com