Re: new autocrasher errors

[Phil Edwards <phil at jaj dot com>]

On Fri, Oct 11, 2002 at 04:55:21PM -0500, Benjamin Kosnik wrote:
> 
> Mainline has additional symbols, due to the codecvt work, and missing
> symbols, due to the first round of allocator cleanups, and cannot
> reasonably expected to be binary compatible with the 3_2-branch library. 
> 
> I consider this ok. Mainline is not a release branch so I'm not too
> worried about it at the moment. These patches are easy to revert, in
> any case, if a new stable branch is cut and ABI compatibility is
> desired.

Agreed.


> However, when the autocrasher catches up, the 3_2-branch library
> sources, built with g++'s of gcc and gcc_3_2-branch should be
> interchangeable.
> 
> It doesn't look like they are right now:
> http://www.devphil.com/build/newchain_oldlib.txt

There have been additional FAILs, too, since Wednesday:

    http://www.devphil.com/build/reports/tests-trunk/regress.png
    http://www.devphil.com/build/reports/tests-trunk/2002-10-09

> Any more details possible?

Right now sources are only update once a day, at 3:30 in the morning,
my time.  Which... *checks gcc-patches* ...was right in the middle of
when you checked your changes in.  So in another couple hours we should
see something new.

I want to change the scripts to do builds more often, and complain when
they break, but coursework has grabbed my free evenings this week.


> Also, abi_check.cc will have to be augmented to deal with related
> symbol versions, ie GLIBCPP_3.2.1 is not something to worry about.
> Anybody have any good ideas about how this should be done? Symbol
> versions are stored as strings right now so I guess that helps.

I'll have to sleep on this one.  New ground for me.


Phil

-- 
I would therefore like to posit that computing's central challenge, viz. "How
not to make a mess of it," has /not/ been met.
                                                 - Edsger Dijkstra, 1930-2002