This reverts commit feac531415.
This patch has deactivated 253 test cases without a sufficient reasoning.
Instead it speaks about a shotgun in its commite message.
The patch is reverted because
- It hasn't been discussed on emacs-devel. It should have, because it is
a serious change in our infrastructure (new official tag :nobatch). Any
documentation of this change, for example in test/README, is missing.
- The proper way to deactivate such tests would have been
(skip-when noninteractive)
Even better to skip for the respective reasons.
- There is no fault report. There is no information about how these tests
have failed. Since it hasn't been a problem so far for us, nobody will
work on a fix forever.
There were a bunch of tests that were breaking make check and should
never be run in batch mode, because they do things like assuming there
is a controlling tty or assuming we can access network services when
we can't (e/g. in a CI/CD environment). I have shotgunned this
problem by tagging all the failing tests with :nobatch and then
changing the default and expensive selectors so make check won't barf
all over its shoes.
As many of these :nobatch should be individually removed as possible, after
upgrading the test harness to mock the environmental stuff they need.
Investigate these failures with "make check-nobatch".
* test/lisp/server-tests.el (server-tests/can-create-frames-p):
Define to be nil on Cygwin unless Emacs has at least one of the
features gfilenotify, dbus, or threads. This causes three tests
to be skipped.
* test/src/keyboard-tests.el (keyboard-unread-command-events):
Skip on Cygwin unless Emacs has at least one of the features
gfilenotify, dbus, or threads.
* test/lisp/server-tests.el
(server-tests/server-force-stop/keeps-frames): Delete every new frame
created during the test. On some systems, 'delete-terminal' will
delete the frames for us, so this ensures that if there are no new
frames, nothing happens.
(cherry picked from commit 3785fe52e4)
This may or may not fix the test run from CI.
* test/lisp/server-tests.el (server-tests/can-create-frames-p):
Don't attempt to create frames if TERM=dumb, which what we have
if run from M-x compile (for instance).
(server-tests/server-force-stop/keeps-frames): Delete created frame so
that it doesn't cause trouble for other tests.
* lisp/emacs-lisp/ert.el (ert--insert-infos): Allow 'message' to be a
function that is called when inserting the info.
(ert-info): Update docstring to describe using a function for
MESSAGE-FORM.
* lisp/server.el (server-start): Log when the server is starting.
* test/lisp/server-tests.el (server-tests/can-create-frames-p): New
constant. Use it to skip tests that need to create frames.
(server-tests/start-emacsclient): Rename to...
(server-tests/start-client): ... this, and set the process's buffer.
(server-tests/with-server): Put the server file in a temporary
directory so we don't conflict with real Emacs servers.
(server-tests/with-client): New macro...
(server-tests/server-start/stop-prompt-with-client)
(server-tests/emacsclient/server-edit)
(server-tests/emacsclient/create-frame)
(server-tests/emacsclient/create-frame): ... use it.
(server-tests/server-start/stop-prompt-with-client): Simplify.
This eliminates a useless error prompt when killing Emacs from a
client frame when there are no other frames (bug#58877).
* lisp/server.el (server-running-external): New error.
(server--file-name): New function...
(server-eval-at): ... use it.
(server-start): Factor out server stopping code into...
(server-stop): ... here.
(server-force-stop): Use 'server-stop', and tell it not to delete
frames.
* test/lisp/server-tests.el
(server-tests/server-force-stop/keeps-frames): New test.
* lisp/server.el (server-mode-map): New keymap...
(server-mode): ... use it.
(server-start): Update the 'server-mode' variable (and sync to
'global-minor-modes') when starting/stopping the server.
* test/lisp/server-tests.el: New file (bug#58909).
