GLIBC 2.25 and up test failures listed on wiki

GLIBC 2.25 and up test failures listed on wiki

[Lee Mracek]

While building GLIBC I've been running into the issues with
test-cond-printers, etc as listed in the architecture-independent section
of https://sourceware.org/glibc/wiki/Release/2.25. Since I am sure that
Python is in my path, I'm assuming that this is due to the potential cause
added to the wiki regarding "gcc 7.8 or newer" but I'd like to confirm that.

Currently my error presents as an unexpected output, and running the binary
itself doesn't provide any information. I can provide a full log if
necessary to compare.

Furthermore, I'm wondering what changed in gcc 7.8 to make this test pass.
I'd love to deeply understand what is going on before I deploy a glibc that
is failing unit tests.

Regards,
Lee Mracek

Re: GLIBC 2.25 and up test failures listed on wiki

[Carlos O'Donell]

On 6/11/19 4:47 PM, Lee Mracek wrote:
> While building GLIBC I've been running into the issues with
> test-cond-printers, etc as listed in the architecture-independent section
> of https://sourceware.org/glibc/wiki/Release/2.25. Since I am sure that
> Python is in my path, I'm assuming that this is due to the potential cause
> added to the wiki regarding "gcc 7.8 or newer" but I'd like to confirm that.

Look closely, it's "gdb 7.8 or newer", you need:

- New enough python.

   * Python 3.4 or later

     Python is required to build the GNU C Library.  As of release time,
     Python 3.7.1 is the newest verified to work for building and
     testing the GNU C Library.

- New enough PExpect.

   * PExpect 4.0

     The pretty printer tests drive GDB through test programs and
     compare its output to the printers'.  PExpect is used to capture
     the output of GDB, and should be compatible with the Python version
     in your system.  As of release time PExpect 4.3 is the newest
     verified to work to test the pretty printers.

- New enough GNU Debugger (gdb).

   * GDB 7.8 or later with support for Python 2.7/3.4 or later

     GDB itself needs to be configured with Python support in order to
     use the pretty printers.  Notice that your system having Python
     available doesn't imply that GDB supports it, nor that your
     system's Python and GDB's have the same version.  As of release
     time GNU 'debugger' 8.2 is the newest verified to work to test the
     pretty printers.

     Unless Python, PExpect and GDB with Python support are present, the
     printer tests will report themselves as 'UNSUPPORTED'.  Notice that
     some of the printer tests require the GNU C Library to be compiled
     with debugging symbols.

All three of these things come togethe to test the pretty printers.

Notes taken from glibc/INSTALL.

> Currently my error presents as an unexpected output, and running the binary
> itself doesn't provide any information. I can provide a full log if
> necessary to compare.
> 
> Furthermore, I'm wondering what changed in gcc 7.8 to make this test pass.
> I'd love to deeply understand what is going on before I deploy a glibc that
> is failing unit tests.

It is documented as the minimum for the test to work in glibc/INSTALL.

We don't explicitly explain why, but I expect it's a combination of what was
the oldest tested gdb which works. Feel free to report back with an older
gdb if it works :-)

-- 
Cheers,
Carlos.

Re: GLIBC 2.25 and up test failures listed on wiki

[Florian Weimer]

* Lee Mracek:

> While building GLIBC I've been running into the issues with
> test-cond-printers, etc as listed in the architecture-independent section
> of https://sourceware.org/glibc/wiki/Release/2.25. Since I am sure that
> Python is in my path, I'm assuming that this is due to the potential cause
> added to the wiki regarding "gcc 7.8 or newer" but I'd like to confirm
> that.

I don't think GCC 7.8 exists.  Do you mean GDB?

> Currently my error presents as an unexpected output, and running the binary
> itself doesn't provide any information. I can provide a full log if
> necessary to compare.

Perhaps backporting this commit will help?

commit 56a731d4ad319e473851893980e734b7ee42c1a0
Author: Jan Kratochvil <jan.kratochvil@redhat.com>
Date:   Fri Feb 8 13:49:19 2019 -0500

    Avoid readline conflicts in pexpect
    
    In some cases, sensitive to readline version and the user's
    environment, gdb might emit escape codes while run under python's
    pexpect (i.e. testing pretty printers).  This patch, suggested
    by Jan, helps isolate the test from the user's environment.

If not, you need to look at the respective .out files and check what is
reported there.

Thanks,
Florian