Implement timestamp'ed output on "make check"
authorSergio Durigan Junior <sergiodj@redhat.com>
Thu, 22 Nov 2018 21:45:33 +0000 (16:45 -0500)
committerSergio Durigan Junior <sergiodj@redhat.com>
Sun, 25 Nov 2018 23:21:26 +0000 (18:21 -0500)
commitf63c03b470036353c8c6c657e15f5ebd62ab67dd
tree7dd395d44c0b91cb95cd7b01fe550a134eabfafd
parentb5b12e1dbe0f2739624f91621828df73ef55e900
Implement timestamp'ed output on "make check"

It is unfortunately not uncommon to have tests hanging on some of the
BuildBot workers.  For example, the ppc64be/ppc64le+gdbserver builders
are especially in a bad state when it comes to testing GDB/gdbserver,
and we can have builds that take an absurd amount of time to
finish (almost 1 week for one single build, for example).

It may be hard to diagnose these failures, because sometimes we don't
have access to the faulty systems, and other times we're just too busy
to wait and check which test is actually hanging.  During one of our
conversations about the topic, someone proposed that it would be a
good idea to have a timestamp put together with stdout output, so that
we can come back later and examine which tests are taking too long to
complete.

Here's my proposal to do this.  The very first thing I tried to do was
to use "ts(1)" to achieve this feature, and it obviously worked, but
the problem is that I'm afraid "ts(1)" may not be widely available on
every system we support.  Therefore, I decided to implement a *very*
simple version of "ts(1)", in Python 3, which basically does the same
thing: iterate over the stdin lines, and prepend a timestamp onto
them.

As for testsuite/Makefile.in, the user can now specify two new
variables to enable timestamp'ed output: TS (which enables the
output), and TS_FORMAT (optional, used to specify another timestamp
format according to "strftime").

Here's an example of how the output looks like:

  ...
  [Nov 22 17:07:19] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/call-strs.exp ...
  [Nov 22 17:07:19] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/step-over-no-symbols.exp ...
  [Nov 22 17:07:20] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/all-architectures-6.exp ...
  [Nov 22 17:07:20] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/hashline3.exp ...
  [Nov 22 17:07:20] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/max-value-size.exp ...
  [Nov 22 17:07:20] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/quit-live.exp ...
  [Nov 22 17:07:46] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/paginate-bg-execution.exp ...
  [Nov 22 17:07:56] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/gcore-buffer-overflow.exp ...
  [Nov 22 17:07:56] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/gcore-relro.exp ...
  [Nov 22 17:07:56] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/watchpoint-delete.exp ...
  [Nov 22 17:07:56] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/breakpoint-in-ro-region.exp ...
  [Nov 22 17:07:56] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/vla-sideeffect.exp ...
  [Nov 22 17:07:57] [1234] Running binutils-gdb/gdb/testsuite/gdb.base/unload.exp ...
  ...

(What, gdb.base/quit-live.exp is taking 26 seconds to complete?!)

Output to stderr is not timestamp'ed, but I don't think that will be a
problem for us.  If it is, we can revisit the solution and extend it.

gdb/testsuite/ChangeLog:
2018-11-25  Sergio Durigan Junior  <sergiodj@redhat.com>

* Makefile.in (TIMESTAMP): New variable.
(check-single): Add $(TIMESTAMP) to the end of $(DO_RUNTEST)
command.
(check-single-racy): Likewise.
(check/%.exp): Likewise.
(check-racy/%.exp): Likewise.
(workers/%.worker): Likewise.
(build-perf): Likewise.
(check-perf): Likewise.
* README: Describe new "TS" and "TS_FORMAT" variables.
* print-ts.py: New file.
gdb/testsuite/ChangeLog
gdb/testsuite/Makefile.in
gdb/testsuite/README
gdb/testsuite/print-ts.py [new file with mode: 0755]
This page took 0.025816 seconds and 4 git commands to generate.