bug#52941: 29.0.50; fido-mode: characters typed into minibuffers inserted multiple times

bug#52941: 29.0.50; fido-mode: characters typed into minibuffers inserted multiple times

[Sean Whitton]

Hello,

If the system is under heavy load, and/or Emacs is compiled with debug
options which substantially slow it down, and fido-mode is enabled, then
typing quickly into the minibuffer can result in typed characters
inserted more than once.  Here is how I can reproduce it:

    emacs -Q
    M-x fido-mode RET
    C-h v

and quickly try to type "font-weight-table".  This is what I get in the
minibuffer:

    font-weweweweweweweight-tatabe

Although I made at least one typo (I didn't type the 'l'), I definitely
did not type 'we' and 'ta' that many times.

I haven't been able to reproduce with icomplete-mode, but as the problem
does not occur consistently, I am not confident it does not occur with
icomplete-mode.

-- 
Sean Whitton

bug#52941: 29.0.50; pgtk: characters typed into minibuffers inserted multiple times

[Sean Whitton]

Hello,

On Sat 01 Jan 2022 at 04:14PM -07, Sean Whitton wrote:

> If the system is under heavy load, and/or Emacs is compiled with debug
> options which substantially slow it down, and fido-mode is enabled, then
> typing quickly into the minibuffer can result in typed characters
> inserted more than once.  Here is how I can reproduce it:
>
>     emacs -Q
>     M-x fido-mode RET
>     C-h v
>
> and quickly try to type "font-weight-table".  This is what I get in the
> minibuffer:
>
>     font-weweweweweweweight-tatabe
>
> Although I made at least one typo (I didn't type the 'l'), I definitely
> did not type 'we' and 'ta' that many times.
>
> I haven't been able to reproduce with icomplete-mode, but as the problem
> does not occur consistently, I am not confident it does not occur with
> icomplete-mode.

It occurred to me to try --with-x-toolkit=lucid instead of --with-pgtk.
Sure enough, I cannot reproduce the bug.  So I think the problem is
something about how pgtk receives and processes keyboard input.

In summary: I can reproduce with pgtk running as a native Wayland
application, and I cannot reproduce with the Lucid build running under
Xwayland.  In both cases, swaywm.  I have retitled the bug.

Thanks.

-- 
Sean Whitton

bug#52941: 29.0.50; pgtk: characters typed into minibuffers inserted multiple times

[João Távora]

[-- Attachment #1: Type: text/plain, Size: 1637 bytes --]

Thanks, Sean, that's very good investigation and I'll admit induces
a "phew" on my part ;-)

And yes, I'm pretty sure this will happen with icomplete-mode as
well.

João

On Tue, Jan 11, 2022 at 12:11 AM Sean Whitton <spwhitton@spwhitton.name>
wrote:

> Hello,
>
> On Sat 01 Jan 2022 at 04:14PM -07, Sean Whitton wrote:
>
> > If the system is under heavy load, and/or Emacs is compiled with debug
> > options which substantially slow it down, and fido-mode is enabled, then
> > typing quickly into the minibuffer can result in typed characters
> > inserted more than once.  Here is how I can reproduce it:
> >
> >     emacs -Q
> >     M-x fido-mode RET
> >     C-h v
> >
> > and quickly try to type "font-weight-table".  This is what I get in the
> > minibuffer:
> >
> >     font-weweweweweweweight-tatabe
> >
> > Although I made at least one typo (I didn't type the 'l'), I definitely
> > did not type 'we' and 'ta' that many times.
> >
> > I haven't been able to reproduce with icomplete-mode, but as the problem
> > does not occur consistently, I am not confident it does not occur with
> > icomplete-mode.
>
> It occurred to me to try --with-x-toolkit=lucid instead of --with-pgtk.
> Sure enough, I cannot reproduce the bug.  So I think the problem is
> something about how pgtk receives and processes keyboard input.
>
> In summary: I can reproduce with pgtk running as a native Wayland
> application, and I cannot reproduce with the Lucid build running under
> Xwayland.  In both cases, swaywm.  I have retitled the bug.
>
> Thanks.
>
> --
> Sean Whitton
>


-- 
João Távora

[-- Attachment #2: Type: text/html, Size: 2282 bytes --]

bug#52941: 29.0.50; pgtk: characters typed into minibuffers inserted multiple times

[Po Lu via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Sean Whitton <spwhitton@spwhitton.name> writes:

> It occurred to me to try --with-x-toolkit=lucid instead of --with-pgtk.
> Sure enough, I cannot reproduce the bug.  So I think the problem is
> something about how pgtk receives and processes keyboard input.

> In summary: I can reproduce with pgtk running as a native Wayland
> application, and I cannot reproduce with the Lucid build running under
> Xwayland.  In both cases, swaywm.  I have retitled the bug.

You forgot to test the following combinations:

  --with-x-toolkit=gtk3, x-gtk-use-native-input set to nil
  --with-x-toolkit=gtk3, x-gtk-use-native-input set to t
  --with-pgtk, GDK_BACKEND=x11

Could you please try all three setups and see which of those the
problem exists in?

Thanks in advance.

bug#52941: 29.0.50; pgtk: characters typed into minibuffers inserted multiple times

[Sean Whitton]

Hello Po,

On Tue 11 Jan 2022 at 06:21PM +08, Po Lu wrote:

> Sean Whitton <spwhitton@spwhitton.name> writes:
>
>> It occurred to me to try --with-x-toolkit=lucid instead of --with-pgtk.
>> Sure enough, I cannot reproduce the bug.  So I think the problem is
>> something about how pgtk receives and processes keyboard input.
>
>> In summary: I can reproduce with pgtk running as a native Wayland
>> application, and I cannot reproduce with the Lucid build running under
>> Xwayland.  In both cases, swaywm.  I have retitled the bug.
>
> You forgot to test the following combinations:
>
>   --with-x-toolkit=gtk3, x-gtk-use-native-input set to nil
>   --with-x-toolkit=gtk3, x-gtk-use-native-input set to t

Cannot reproduce.  I did try both, but I can't find any mention of
x-gtk-use-native-input in the Emacs source code -- is it possible you
meant a similarly-named variable?

>   --with-pgtk, GDK_BACKEND=x11

Can reproduce, but it happens less often, i.e., you have to load up the
system further and type faster to reproduce :)

-- 
Sean Whitton

bug#52941: 29.0.50; pgtk: characters typed into minibuffers inserted multiple times

[Po Lu via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Sean Whitton <spwhitton@spwhitton.name> writes:

> Hello Po,
>
> On Tue 11 Jan 2022 at 06:21PM +08, Po Lu wrote:
>
>> Sean Whitton <spwhitton@spwhitton.name> writes:
>>
>>> It occurred to me to try --with-x-toolkit=lucid instead of --with-pgtk.
>>> Sure enough, I cannot reproduce the bug.  So I think the problem is
>>> something about how pgtk receives and processes keyboard input.
>>
>>> In summary: I can reproduce with pgtk running as a native Wayland
>>> application, and I cannot reproduce with the Lucid build running under
>>> Xwayland.  In both cases, swaywm.  I have retitled the bug.
>>
>> You forgot to test the following combinations:
>>
>>   --with-x-toolkit=gtk3, x-gtk-use-native-input set to nil
>>   --with-x-toolkit=gtk3, x-gtk-use-native-input set to t
>
> Cannot reproduce.  I did try both, but I can't find any mention of
> x-gtk-use-native-input in the Emacs source code -- is it possible you
> meant a similarly-named variable?

It's a relatively new feature, you should update from master and test
again.

> Can reproduce, but it happens less often, i.e., you have to load up the
> system further and type faster to reproduce :)

Thanks, I will look into this, but please also update and test native
input.

bug#52941: 29.0.50; pgtk: characters typed into minibuffers inserted multiple times

[Sean Whitton]

Hello,

On Fri 14 Jan 2022 at 08:39AM +08, Po Lu wrote:

> Sean Whitton <spwhitton@spwhitton.name> writes:
>
>> Hello Po,
>>
>> On Tue 11 Jan 2022 at 06:21PM +08, Po Lu wrote:
>>
>>> Sean Whitton <spwhitton@spwhitton.name> writes:
>>>
>>>> It occurred to me to try --with-x-toolkit=lucid instead of --with-pgtk.
>>>> Sure enough, I cannot reproduce the bug.  So I think the problem is
>>>> something about how pgtk receives and processes keyboard input.
>>>
>>>> In summary: I can reproduce with pgtk running as a native Wayland
>>>> application, and I cannot reproduce with the Lucid build running under
>>>> Xwayland.  In both cases, swaywm.  I have retitled the bug.
>>>
>>> You forgot to test the following combinations:
>>>
>>>   --with-x-toolkit=gtk3, x-gtk-use-native-input set to nil
>>>   --with-x-toolkit=gtk3, x-gtk-use-native-input set to t
>>
>> Cannot reproduce.  I did try both, but I can't find any mention of
>> x-gtk-use-native-input in the Emacs source code -- is it possible you
>> meant a similarly-named variable?
>
> It's a relatively new feature, you should update from master and test
> again.

Ah, sorry.  Now done.  Cannot reproduce in either case.

In the latter case I had to kill fcitx5 before I could type into Emacs.
So I thought I'd also test pgtk with and without GDK_BACKEND=x11 without
fcitx5 running.  I could reproduce in both cases.

-- 
Sean Whitton

bug#52941: 29.0.50; pgtk: characters typed into minibuffers inserted multiple times

[Po Lu via Bug reports for GNU Emacs, the Swiss army knife of text editors]

Sean Whitton <spwhitton@spwhitton.name> writes:

> Ah, sorry.  Now done.  Cannot reproduce in either case.
>
> In the latter case I had to kill fcitx5 before I could type into Emacs.
> So I thought I'd also test pgtk with and without GDK_BACKEND=x11 without
> fcitx5 running.  I could reproduce in both cases.

Thanks for testing.

bug#52941: 29.0.50; pgtk: characters typed into minibuffers inserted multiple times

[Yuuki Harano]

[-- Attachment #1: Type: text/plain, Size: 412 bytes --]


On Tue, 11 Jan 2022 09:11:13 +0900,
Sean Whitton wrote:
> > If the system is under heavy load, and/or Emacs is compiled with debug
> > options which substantially slow it down, and fido-mode is enabled, then
> > typing quickly into the minibuffer can result in typed characters
> > inserted more than once.

Thanks for the report.
Please try the attached patch, which should solve the issue.

-- 
Yuuki Harano


[-- Attachment #2: events-processed-twice.diff --]
[-- Type: application/octet-stream, Size: 919 bytes --]

diff --git a/src/pgtkterm.c b/src/pgtkterm.c
index 0155ae991d..5be9af7160 100644
--- a/src/pgtkterm.c
+++ b/src/pgtkterm.c
@@ -153,10 +153,24 @@ evq_enqueue (union buffered_input_event *ev)
 evq_flush (struct input_event *hold_quit)
 {
   struct event_queue_t *evq = &event_q;
-  int i, n = evq->nr;
-  for (i = 0; i < n; i++)
-    kbd_buffer_store_buffered_event (&evq->q[i], hold_quit);
-  evq->nr = 0;
+  int n = 0;
+
+  while (evq->nr > 0)
+    {
+      /* kbd_buffer_store_buffered_event may do longjmp, so
+	 we need to shift event queue first and pass the event
+	 to kbd_buffer_store_buffered_event so that events in
+	 queue are not processed twice.  Bug#52941 */
+      union buffered_input_event ev = evq->q[0];
+      int i;
+      for (i = 1; i < evq->nr; i++)
+	evq->q[i - 1] = evq->q[i];
+      evq->nr--;
+
+      kbd_buffer_store_buffered_event (&ev, hold_quit);
+      n++;
+    }
+
   return n;
 }
 

bug#52941: 29.0.50; pgtk: characters typed into minibuffers inserted multiple times

[Sean Whitton]

Hello,

On Sun 16 Jan 2022 at 12:08AM +09, Yuuki Harano wrote:

> On Tue, 11 Jan 2022 09:11:13 +0900,
> Sean Whitton wrote:
>> > If the system is under heavy load, and/or Emacs is compiled with debug
>> > options which substantially slow it down, and fido-mode is enabled, then
>> > typing quickly into the minibuffer can result in typed characters
>> > inserted more than once.
>
> Thanks for the report.
> Please try the attached patch, which should solve the issue.

Looks like it does -- thanks!

-- 
Sean Whitton

bug#52941: 29.0.50; pgtk: characters typed into minibuffers inserted multiple times

[Yuuki Harano]

On Sun, 16 Jan 2022 06:10:17 +0900,
Sean Whitton wrote:
> > Please try the attached patch, which should solve the issue.
> 
> Looks like it does -- thanks!

Applied.  Thanks for the confirmation.

-- 
Yuuki Harano