Re: server.el test failures

[Jim Porter <jporterbugs@gmail.com>]

On 2/28/2023 5:05 AM, Robert Pluim wrote:
>      Eli> Why does make-frame create a GUI X frame in a batch session?  (I've
>      Eli> seen that in your previous backtrace, but thought it was because you
>      Eli> ran that test in an interactive session.)

Ah ha, that's why I couldn't reproduce this. I usually run tests in an 
SSH session (most of my development work is in an isolated VM so I can 
throw it away if I mess something up really badly). Running within a GUI 
session, I see the error now.

> The test does 'emacsclient -c', changing that to 'emacsclient -c -nw'
> also makes the test succeed
> 
>      Eli> I don't think we should create GUI frames in batch, since the SIGIO
>      Eli> handler is not set, and we generally aren't prepared for receiving
>      Eli> it.  We should create a TTY frame instead, or maybe no frame at all.
> 
> I think the former would be better. Jim, since you wrote this test,
> what do you think?

We should definitely create a frame in these tests, since that's what 
the tests are trying to, well, test. (All the server tests are there to 
make sure that the usual things you might do with "emacsclient" work 
properly.)

This might be a real bug in Emacs: as Eli says, an Emacs server started 
in batch mode shouldn't create GUI frames, but on the other hand, this 
might be only be relevant to the regression tests (who starts an Emacs 
server in a batch session?). I wouldn't mind just adding "-nw" to all 
the "emacsclient -c" invocations in server-tests.el.