Re: glibc 2.25 seems to have broken AddressSaniitzer Inbox x glibc x sanitizer x

[Joseph Myers <joseph at codesourcery dot com>]

On Thu, 22 Feb 2018, Konstantin Serebryany wrote:

> LLVM has the sufficient tests, which is how we've learned about the problem.
> Do glibc developers run LLVM tests?
> If not, is it possible to change this?

I think the tests really need to be run by people familiar with LLVM 
development, who can keep the build infrastructure up to date and filter 
out issues relating to LLVM instability and only raise things that are 
related to glibc.  LLVM is much more aggressive than GCC about requiring 
recent versions of build tools, so there's significant effort involved in 
keeping LLVM builds going on a stable operating system version (possibly 
with host compilers, host libc, Sphinx etc. a few years old).

As is, we don't even have all the existing glibc buildbots properly 
maintained.  If you look at 
<http://glibc-buildbot.reserved-bit.com/waterfall> you'll see the i686 and 
x86_64 bots have been failing configure for a long time because they are 
running on Ubuntu 14.04 whose GCC and binutils versions are now too old 
for building glibc.  So it's very much for people who want new tests X to 
be run to set up and keep maintaining the infrastructure to run tests X 
and to report the issues found by tests X, rather than suggesting tests X 
and hoping that existing developers will maintain that infrastructure for 
them.

If people wish to contribute to glibc by running tests - which is 
certainly a very useful contribution to make, given the limited resources 
in that area at present - it's useful to have different people running a 
wide variety of tests continuously with current glibc and other tools and 
reporting issues found - everything up to and including continuous 
rebuilds of complete GNU/Linux distributions on sufficiently large build 
clusters.

-- 
Joseph S. Myers
joseph@codesourcery.com