bug#65325: "make check" hangs on Cygwin

[Ken Brown <kbrown@cornell.edu>]

On 8/19/2023 2:15 AM, Eli Zaretskii wrote:
>> Date: Fri, 18 Aug 2023 18:02:46 -0400
>> From: Ken Brown <kbrown@cornell.edu>
>> Cc: 65325@debbugs.gnu.org
>>
>>> I can reproduce the hang.  But I don't get the hang in the default build
>>> on my system (no configure options).  I'll have to play with the
>>> configure options and try to figure out what's going on.
>>
>> OK, the hang in server-tests.el [and a similar hang in
>> keyboard-tests.el] occurs when getchar is called at keyboard.c:3978.
>> This call only occurs if dbus, file notification, and threads are all
>> disabled, i.e., if the configure options without-dbus,
>> without-file-notification, and without-threads are all specified or
>> implied.  I don't understand the keyboard code well enough to know (a)
>> why we reach that call to getchar in the context of the hanging tests
>> and (b) why getchar is only called in the absence of dbus, file
>> notification, and threads.
> 
> Because we have no other way to read input in such a build.  The code
> we have there obviously cannot handle anything but character input
> from stdin, so it is a small wonder that it hangs in an ERT test.
> 
> I think this is a red herring: we should skip these tests if none of
> those 3 features is available.  (We could also rewrite the tests to
> support such a build, but I have very little sympathy for people who
> build Emacs --without-all and still want to run all the test suite
> successfully, and thus I don't think this would be a worthwhile
> investment of our energy and resources.)
> 
>> I can work around the problem by always avoiding that getchar call on
>> Cygwin (i.e., by adding "&& !defined CYGWIN" at line 3972).  With that
>> change, all tests pass on Cygwin with Bruno's configure options.  But I
>> don't want to make that change without understanding what's going on.
>> Can anyone help?
> 
> I don't think it's a Cygwin-only failure.  But if it is, it is also OK
> to skip these tests only on Cygwin, in addition to testing the absence
> of those 3 features.

Sounds good to me.  So the first step is find out if this failure is 
Cygwin-only.  Bruno, have you tested a build --without-all on other 
platforms?

Ken