Re: [lmi] Manual test of automated GUI test

[Greg Chicares]

On 2014-10-19 00:16Z, Vadim Zeitlin wrote:
> On Sat, 18 Oct 2014 22:44:58 +0000 Greg Chicares <address@hidden> wrote:
> 
> GC> [Others could easily do this by accident, so would it make sense
> GC> to forestall it...e.g., by issuing an 'Esc' keypress before doing
> GC> anything else in the GUI test, if that's possible?]
> 
>  The patch below (to be applied on top of the previous patch) does this and
> also incidentally fixes the crash, as explained in the commit message below.

I applied this to my local tree, rebuilt, and ran the test in my
"mischievous" way--without the secret backdoor argument. The
following messagebox popped up--several times in succession, IIRC:

Programming error
A document created during unattended test hasn't been closed, please report 
this.

The log says:

00:22:31: Starting automatic tests:
00:22:31:     about_dialog_version: ok (8ms)
00:22:31: Error: can't open file 'wx_test.conf' (error 2: the system cannot 
find the file specified.)
[due to that error, I expect failures downstream]
[...]
00:24:15: FAILURE: 21 out of 24 tests failed.

But something really curious happened: a floating context menu
popped up. It contained "Census" and "Illustration". It's exactly
the menu I would expect if I started the production system and
clicked the "open a new file" button on the toolbar. It appeared
at the mouse-cursor position (as expected), but (unexpectedly)
even if I carefully moved the mouse cursor into the msw-xp
"taskbar" before starting the test, then didn't touch the mouse
or the keyboard while the test was running.

Well, once that menu appears, the test program seems to pause
for input. If I click one of the two choices it offers, then
apparently a single test is run, and the menu appears for
another, and another...and then the "Programming error"
messagebox doesn't appear. The number of tests that succeed
with my random clicking is probably a function of which of
the two choices I randomly pick; on one run, only 13 failed.

With the patch, though, it no longer crashes.

And if I run the test non-mischievously, with the secret backdoor
argument, then no such curious thing happens, and a greater
number of tests succeed--the same number as with my previous
non-mischievous trial without this latest patch:

00:32:10: FAILURE: 10 out of 24 tests failed.