bug#48500: 28.0.50; url-retrieve-synchronously exits abnormally due to pending keyboard input from terminal

[Shane Mulligan <mullikine@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 5686 bytes --]

I have done a write-up on this issue to clarify everything.

https://mullikine.github.io/posts/fixing-googling-tooling-in-emacs/

Shane Mulligan

How to contact me:
🇦🇺 00 61 421 641 250
🇳🇿 00 64 21 1462 759 <+64-21-1462-759>
mullikine@gmail.com


On Wed, May 19, 2021 at 6:49 PM Shane Mulligan <mullikine@gmail.com> wrote:

> Hi Eli,
>
> I tagged you back in this conversation because I may have fixed the issue
> :).
>
> Shane Mulligan
>
> How to contact me:
> 🇦🇺 00 61 421 641 250
> 🇳🇿 00 64 21 1462 759 <+64-21-1462-759>
> mullikine@gmail.com
>
>
> On Wed, May 19, 2021 at 6:48 PM Shane Mulligan <mullikine@gmail.com>
> wrote:
>
>> I may have resolved this issue with the following patch to
>> `url-retrieve-synchronously`.
>> What this achieves is to trigger a `quit` in a controlled environment
>> rather than allowing it to occur when `accept-process-output` is run.
>> It's not always wanted to trigger a quit when `(input-pending-p)` is `t`.
>> But I noticed from placing `while-no-input` around `accept-process-output`
>> to avoid the `quit` that `url-retrieve-synchronously` would then hang but
>> with the controlled `quit` happening beforehand, `accept-process-output` no
>> longer needs `while-no-input` around it. The end result is buttery smooth
>> helm with no accidental `quit` from typing too fast. I think this may have
>> resulted in GUI helm faster too.
>>
>> ```
>> 29c29,30
>> <       (let ((proc (get-buffer-process asynch-buffer)))
>> ---
>> >       (let ((proc (get-buffer-process asynch-buffer))
>> >             (counter 0))
>> 71a73,78
>> >             (with-local-quit
>> >               (if (input-pending-p)
>> >                   (progn
>> >                     (setq counter (1+ counter))
>> >                     (if (> counter 20)
>> >                         (keyboard-quit)))))
>> ```
>> Shane Mulligan
>>
>> How to contact me:
>> 🇦🇺 00 61 421 641 250
>> 🇳🇿 00 64 21 1462 759 <+64-21-1462-759>
>> mullikine@gmail.com
>>
>>
>> On Wed, May 19, 2021 at 11:32 AM Shane Mulligan <mullikine@gmail.com>
>> wrote:
>>
>>> My apologies. It was literally 5am when I wrote that. I think I have
>>> misunderstood `C-g` being generated with `quit-flag`. The bad behaviour is
>>> very clearly still happening. I will try to clarify this problem further by
>>> experimenting with it. I need to figure out why `while-no-input` suppresses
>>> the `quit` but merely setting `quit-flag` does not. Something indirect may
>>> be happening.
>>>
>>> Thank you all,
>>> Shane Mulligan
>>>
>>> How to contact me:
>>> 🇦🇺 00 61 421 641 250
>>> 🇳🇿 00 64 21 1462 759 <+64-21-1462-759>
>>> mullikine@gmail.com
>>>
>>>
>>> On Wed, May 19, 2021 at 4:54 AM Eli Zaretskii <eliz@gnu.org> wrote:
>>>
>>>> > From: Shane Mulligan <mullikine@gmail.com>
>>>> > Date: Wed, 19 May 2021 04:32:16 +1200
>>>> >
>>>> > Thanks for looking into this so quickly. First some background on the
>>>> problem. I managed to work through
>>>> > this issue (https://github.com/emacs-helm/helm/issues/2417) with the
>>>> `emacs-helm` maintainer and we
>>>> > found what appears to be that the call to `accept-process-output`
>>>> inside of `url-retrieve-synchronously` will
>>>> > generate a `C-g` when there is pending input of any char. As far as I
>>>> can tell this is an issue only with
>>>> > terminal emacs. As I understand it, `inhibit-quit`, as used in
>>>> `accept-process-output` allows a `C-g` to be
>>>> > propagated outwards and handled and from what I can see by the
>>>> comments surrounding,
>>>> >
>>>> > ```
>>>> >               ;; accept-process-output returned nil, maybe because
>>>> the process
>>>> >               ;; exited (and may have been replaced with another).
>>>> If we got
>>>> >               ;; a quit, just stop.
>>>> > ```
>>>> >
>>>> > the `C-g` in this case is expected.
>>>> >
>>>> > But I wonder if `C-g` was meant to be generated if the user was
>>>> simply mashing keys on the keyboard.  In
>>>> > this case, the `C-g` emanating from `accept-process-output` was
>>>> bubbling up into `helm` and `helm` was
>>>> > treating it like an error.
>>>> >
>>>> > Here, you can see a quit being generated from the visual cue in the
>>>> minibuffer.
>>>> > https://asciinema.org/a/nAIB8Z1lGgZJqJg9Mt8YiNEM9
>>>> >
>>>> > Here, I have added `while-no-input` and I no longer get the `quit`.
>>>> > https://asciinema.org/a/x9ELZhwDP1IUtmOz0M1cly42H
>>>> >
>>>> > However, as I test the addition of `while-no-input` with
>>>> `helm-google-suggest` (as above), though Quit is no
>>>> > longer being generated from mashing keys, the key input which would
>>>> have generated the `quit` is not
>>>> > immediately shown in `helm`. Instead, it only appears on the next key
>>>> press. So my implementation may not
>>>> > be perfect or complete.
>>>> >
>>>> > Finding the solution would make the minibuffer far less interrupted
>>>> while typing when
>>>> > `url-retrieve-synchronously` is used in the background.
>>>> >
>>>> > Thank you.
>>>> >
>>>> > Shane Mulligan
>>>>
>>>> Please in the future send your responses with the bug address,
>>>> 48500@debbugs.gnu.org, on the CC list, so that others will see your
>>>> detailed descriptions.  I won't have time to take a good look into
>>>> that in the next few days, so it's good to make others aware of your
>>>> findings, because they might look into it meanwhile.
>>>>
>>>> Personally, I find it very strange that typing on the keyboard
>>>> produces C-g, it shouldn't happen, neither on a TTY nor on a GUI
>>>> display.
>>>>
>>>

[-- Attachment #2: Type: text/html, Size: 13232 bytes --]