Re: [PATCH v2 10/12] soft-fp: Add the lack of implementation for 128 bit

[Joseph Myers <joseph at codesourcery dot com>]

On Wed, 18 Jul 2018, Zong Li wrote:

> > I think consideration of (a revised version of) this patch is most
> > appropriately deferred until 2.29 (and probably likewise the new port
> > itself, given the associated risks and the lack of posting before the
> > freeze - posting before the freeze being particularly important for any
> > architecture-independent changes involved in a new port).
> 
> I know this concern and mechanism in glibc. Is there a chance to
> include this modification at this moment? Because this new port
> is the only one port to use it.

Well, being in the freeze you could test the patch much more extensively 
than would normally be expected for such a patch to justify it as not 
affecting existing configurations.  I'm thinking of a before-and-after 
comparison of installed stripped shared libraries built with 
build-many-glibcs.py - build unmodified glibc for all configurations 
build-many-glibcs.py supports, using the --strip option, save the install 
directories, rebuild for all configurations with the patch applied (and no 
other changes), compare the <configuration>/lib* directories with the 
saved copies from the previous build to make sure that none of the 
installed stripped shared libraries have changed at all (and also make 
sure that the test results are identical for the two builds).

That would be overkill for this patch normally - but late in the freeze as 
we are, we need to be much, much more careful about not breaking other 
configurations, and so much more thorough testing is appropriate (and of 
course the patch submission message then needs to describe that testing).

-- 
Joseph S. Myers
joseph@codesourcery.com