Re: [lmi] Can't get GUI test to run for pc-linux-gnu

[Greg Chicares]

On 3/23/21 10:00 PM, Vadim Zeitlin wrote:
> On Thu, 18 Mar 2021 23:55:12 +0000 Greg Chicares <gchicares@sbcglobal.net> 
> wrote:
> 
> GC> I feel certain that lmi's GUI tests passed for pc-linux-gnu at
> GC> some time in the not-too-distant past,
> 
>  Yes, I'm almost certain they did. We wanted to enable them in the CI
> build, but there were some sporadic failures there that we've never quite
> tracked down, so we didn't do it...
> 
> GC> but now they fail for me:
> 
>  And me too. And this is not specific to the "about_dialog_version" test
> case, other tests fail as well.
> 
> GC> I think that's just saying that an exception wasn't caught,
> GC> and the output continues:
> 
> [FWIW, IMHO this is one of those circumstances when it would have been
>  useful to do "LMI_NO_EXCEPTION_DUMP=1 ./wx_test" to suppress all the
>  extra expression backtraces output]

Recent experience led me to a different viewpoint. I ran one of those
tests in isolation (perhaps it was "about_dialog_version"), after
first manually blocking the stack trace, and it gave me only:

NOTE: starting the test suite
about_dialog_version: started
Abnormal-termination handler called. Please report this problem.

Then I reverted the suppression of the stack trace, and got more
information. It wasn't information that I was actually able to
use to diagnose the problem in the time available; but it was
more information.

> GC> Vadim--Have you seen this?
> 
>  Yes, and I hope to fix this soon, but as I still haven't done it almost a
> week later, I've decided to finally reply to this message to at least let
> you know that I was working on this.

Thank you.

> I'd like to ask you once again if you could please merge the PR fixing the
> tests with autotools and adding them to the CI builds from
> https://github.com/let-me-illustrate/lmi/pull/174
Maybe I can look at it tomorrow. At a quick glance, it seems
that the changes span several categories. Some have pervasive
effect, but are simple ('alert_cli.cpp'). I don't know whether
I even have the tools to test this change in 'boost_regex.hpp':
  -#   if 9 <= __GNUC__
  +#   if 10 <= __GNUC__
but I haven't read the commit messages yet (and it looks like
there are twenty-eight of them). Some files ('configure.ac',
e.g.) I don't use anyway. At least one change affects only
a unit test ('numeric_io_test.cpp'), and that should be very
easy to test. OTOH, three scripts that rebuild libraries have
changed; to test them, I'll have to do a complete reinstallation,
and I'll have to do it on the corporate server as well as my own
machine.

And it looks like you've updated the wx version. To us, that's
a large manual testing effort. Is that commit separable from
the rest? Looking in the commit messages...is the only purpose
to avoid a clang warning?