[emacs plugin] marking words under parenthesis about tern

[marijnh]

How are you marking them, and what did you expect to happen? Do mark commands still work? As in, it is just that the marked region isn't highlighted, or that you can't mark at all?

[jxs]

i am marking them with shift and arrows , and mark commands such as mark-word or mark-sexp work, although with arrows i can't mark at all messages buffer outputs :
+1: Mark set
Quit

[marijnh]

I've never done shift-arrow selection in emacs, not does it work in a vanilla setup. Which setting/minor-mode are you using to get that?

[magnars]

I had a similar problem yesterday, but that was using normal C-SPC to set mark. Once the marked region moved over some specific part of the buffer, the mark vanished.

I had no idea what was happening, but next time I'll make sure to disable tern and see if that solves the issue.

[marijnh]

The emacs API is unfortunately full of subtle side effects. It's very possible that Tern's post-command-hook is causing this. If you can give me a way to reproduce, I'll try to figure out why.

[magnars]

If I should wager a guess, it would be with-current-buffer in tern-run-query that messes with the mark. I've seen similar issues with with-temp-buffer at least. I'll let you know if I can reproduce it.

[jxs]

@marijnh it does come in my vanilla setup, version 24.3
http://www.gnu.org/software/emacs/manual/html_node/emacs/Shift-Selection.html
you can reproduce it by loading any javascript file, loading js2-mode and for example do set-mark and mark a word inside parenthesis, or even parenthesis alone like a function call: foo() or foo(bar)

[magnars]

Yes, quite right. Just set-mark and move point over a function call. Once the type hints pop up, the mark is deactivated.

[marijnh]

If I turn on shift-select-mode (or try to reproduce this by setting the mark with C-space), I still don't see this. Whether I start outside of a call, and go inside it, or start inside and go outside, or just move around inside of the call, the mark stays where it used to be. This is Emacs version 24.1.1.

[magnars]

I'm on Emacs 24.3.1.

Here's a reproduction:

• create a file under a directory with a .tern-project that has the browser libs.
• type in document.appendChild("abc")
• if you set the mark and move about, everything works fine. This is because the lookup happened when you typed it in.
• so, go to point 1, save the file, kill the buffer, and reopen it.
• set the mark, move forward one char at a time until you reach (
• moving the cursor one more, inside the (, will disable the mark
• this coincides with the lookup of the function signature - the first time.

Doing it again, won't trigger the bug.

However, I have been able to get it triggering every time. Maybe this has
to do with it not finding the function, and so it keeps looking it up.

[marijnh]

Thanks for the detailed instructions. However, for me, this really doesn't happen. Most likely it's a difference between 24.1 and 24.3. I'll try to upgrade.

[marijnh]

Unfortunately, no, on 24.3.1 it still isn't happening for me. Very odd.

[magnars]

That is weird. It is definitely tern. Here's an ascii-cast with emacs -Q reproducing it:

http://ascii.io/a/2956

[marijnh]

Using your code (conveniently copy-pasted from the asciicast) I can reproduce the problem. I've reduced it to the http request. Simply calling url-http appears to have the side effect of setting the mark (*Messages* shows a 'Mark set' message at this point). I haven't been able to figure out why -- there are no calls to set-mark or similar in the source of the url-http module.

I suspect this can be solved by shadowing the right dynamic variables. I haven't figured out which one to shadow yet, though.

[magnars]

Yes, you're probably right about the shadowing. Have you tried save-excursion tho?

[marijnh]

Hadn't tried, but it appears save-excursion also doesn't help.

[marijnh]

My previous analysis was off, it's not url-http but json-encode that causing this, and it is causing it by calling insert inside a temp buffer. I still haven't found a way to prevent this side effect, and I'm more or less tired of looking. Here's a simple way to reproduce it, if you feel like digging deeper:

(defun shift-selection-disruptor ()
  (interactive)
  (with-temp-buffer
    (insert "x")))

Shift-select something, do M-x, and run shift-selection-disruptor. The selection will be cleared by the insert command.

[magnars]

Here you go:

(defun shift-selection-disruptor ()
  (interactive)
  (let (deactivate-mark)
    (with-temp-buffer
      (insert "x"))))

[marijnh]

Great! See attached patch.