Fix clearing test output buffer

[m-novikov]

If we try to run a test without closing previous results buffer it will fail with error: Text is read-only in mini-buffer.

[wbolster]

hmmm, strange. repeated invocations work perfectly fine for me, and i use this all the time.

can you reproduce this in a clean environment? which emacs version?

[m-novikov]

My emacs version is 26.1 (spaceemacs)
Reproducible in clean environment if comint-prompt-read-only variable is true.

[m-novikov]

Maybe proper fix should add

(let ((comint-prompt-read-only nil))
   ...)

before running test process.
What do you think?

[mpanarin]

I experience exactly same issue with repeated test runs.

Using spacemacs/develop and emacs28

[wbolster]

ah, comint-prompt-read-only is nil by default which is why i didn't notice this myself.

@m-novikov re your comment about comint-prompt-read-only vs inhibit-read-only... not sure, what do you think is best?

totally forgot about this issue, would be happy to merge if someone tells me what the best variable is here...

[pinetr2e]

Instead of 'erase-buffer', how about using 'comint-clear-buffer', which works regardless of comint-prompt-read-only.
Note that it should be called after the comint buffer is created like below:

diff --git a/python-pytest.el b/python-pytest.el
index 1f5ba4c..5b9d569 100644
--- a/python-pytest.el
+++ b/python-pytest.el
@@ -374,7 +374,6 @@ With a prefix ARG, allow editing."
           (user-error "Aborting; pytest still running")))
       (when process
         (delete-process process))
-      (erase-buffer)
       (unless (eq major-mode 'python-pytest-mode)
         (python-pytest-mode))
       (compilation-forget-errors)
@@ -387,6 +386,7 @@ With a prefix ARG, allow editing."
          nil t))
       (run-hooks 'python-pytest-setup-hook)
       (make-comint-in-buffer "pytest" buffer "sh" nil "-c" command)
+      (comint-clear-buffer)
       (run-hooks 'python-pytest-started-hook)
       (setq process (get-buffer-process buffer))
       (set-process-sentinel process #'python-pytest--process-sentinel)

[wbolster]

wouldn't that clear the buffer after starting the process? that's a race condition that can cause output to get lost i think? 🤔

[pinetr2e]

You are right. I did not consider that make-comint-in-buffer starts the process right away.
I just tried to find a simpler way but it seems that it was not so great.

By the way, the change in this PR, relying on inhibit-read-only seems to be fine as a proper way. For example, comint-clear-buffer is implemented based on inhibit-read-only in the same way and many other packages do the same.

(comint-prompt-read-only simply does not work and the usage is not designed to be like inhibit-read-only)

[wbolster]

inhibit-read-only it is. finally merged... thanks all!