Bizarre conflict with auto-revert-mode

[raxod502]

I've been gradually driven crazy over the last several months with the symptoms of this problem, but it wasn't until you introduced the aggressive-indent-sit-for-time variable (thanks!) that I was finally able to track it down reproducibly. Now, it appears to me that the issue may have been fixed just a few hours ago in commit 2f54a58 (in which case this can be closed), but I had to deal with this problem for a long time and couldn't find any information about how to reproduce it, so I think it should be reported.

Here is what you need to do from emacs -Q with a pre- 2f54a58 aggressive-indent:

(load-file "~/path/to/aggressive-indent.el")
(aggressive-indent-mode 1)
(setq aggressive-indent-sit-for-time 1)
(auto-revert-mode 1)

Now, the problem still exists without increasing aggressive-indent-sit-for-time, but it's virtually impossible to reproduce consistently with the default settings since the behavior only occurs when you press a key within aggressive-indent-sit-for-time.

Here are a few of the funky things that start happening once all of the above setup is done:

• If you type too quickly (i.e. with less than one second between successive keypresses), often half of your keypresses will be silently dropped.
• In the terminal, if you type an open paren and then press e.g. the right arrow key, the text OC is inserted instead of the cursor moving. With aggressive-indent-sit-for-time increased to one second, you can mash the arrow keys to get a long string of OAOBOAOCOD inserted into your buffer, as long as the delay between successive keypresses is less than one second. I bet this is because the ESC keypress is being dropped as per the previous bullet point.

My experience before figuring this out was that as long as aggressive-indent-mode was enabled, sometimes (perhaps once every few days, with no discernible pattern), my keypresses would start occasionally being silently dropped, or A-D would occasionally be inserted into my buffer when I pressed an arrow key, or the meta key would be pressed for me (turning e.g. previous-line into paredit-splice-sexp-killing-backward -- very fun). I now think that the occurrences must have been roughly correlated with the performance of aggressive-indent (which would of course be very nondeterministic and difficult to predict). That is, when aggressive-indent was running slowly, it would have the same effect as increasing aggressive-indent-sit-for-time.

Hope this information helps.

[raxod502]

Actually, I think this is still happening :(

[raxod502]

I wonder if it has anything to do with syl20bnr/spacemacs#5413 ?

[raxod502]

I'll leave this issue closed unless I can come up with some productive way to provide directions to reproduce.

[raxod502]

This was a manifestation of the buggy behavior of while-no-input discussed in https://debbugs.gnu.org/cgi/bugreport.cgi?bug=31692.

cc @Malabarba

I suggest modifying the code currently in aggressive-indent so that people don't encounter this problem on Emacs 26.