bug#16315: Slow paste

bug#16315: Slow paste

[Perry Smith]

[-- Attachment #1: Type: text/plain, Size: 2605 bytes --]

Hi,

I have a Mac (10.8.5) running JollysFastVNC (1.48) viewing a RealVNC server (5.0.1) on AIX (6.1 TL08 SP03) running emacs (24.3).

When I copy from a Mac window and paste into emacs (yank), there is a long delay.  When I started seriously testing, it was always 5 seconds but I have memories that sometimes it was much longer.  I'm calling these memories "false" at this point.

I believe it started when emacs 24 was being developed but I'm not sure.  I know the issue existed in 24.2 and did not exist in emacs 23 that I was using but I don't recall which subversion that was.

I've turned on TRACE_SELECTION in xselect.c and I've also duplicated those macros in keyboard.c and process.c.  When I start emacs -q I get this output:

10748140: get_input_pending returning 0
10748140: last check: read_kdb=-1 NILP(wait_for_cell):1
10748140: get_input_pending returning 0
10748140: no_avail:0
10748140: pselect 2

followed by a half second pause.  This is the normal polling of the input (I guess its normal).  I go to a Mac window, select some text, then hit command-C to copy it into the cut buffer, then select the VNC window (where emacs is already the active window) and hit control-Y to paste.

I get:

10748140: get_input_pending returning 1
10748140: get_input_pending returning 1
10748140: Get selection UTF8_STRING, type _EMACS_TMP_
10748140:   Start waiting 5 secs for SelectionNotify
10748140: last check: read_kdb=0 NILP(wait_for_cell):0
10748140: Received SelectionNotify
10748140: get_input_pending returning 0
10748140: no_avail:0
10748140: pselect 2

followed by a 5 second pause, then I get

10748140:   Got event = 1
10748140: Reading selection data
10748140: Read 7 bytes from property _EMACS_TMP_
10748140:   Delete property _EMACS_TMP_

and then I see the text show up in the emacs window.  I *think* what is "wrong" here is in this case, get_input_pending is actually pulling the text out and via the SelectionNotify handler placing the text into reading_selection_reply but then returns zero because, at that point in time, there is no input to be read.  Thus this if test, is not true:

     if ((read_kbd || !NILP (wait_for_cell))
	  && detect_input_pending ())

I'm thinking that after detect_input_pending is called in wait_reading_process_output, another two lines like this:

     if (! NILP (wait_for_cell) && ! NILP (XCAR (wait_for_cell)))
	break;

should be added before the if (!no_avail) line (around line 4612 in the copy I am looking at.

I made this change in my copy and so far, it seems to work.

Perry Smith


[-- Attachment #2: Message signed with OpenPGP using GPGMail --]
[-- Type: application/pgp-signature, Size: 495 bytes --]

bug#16315: Slow paste

[Lars Ingebrigtsen]

Perry Smith <pedzsan@gmail.com> writes:

> When I copy from a Mac window and paste into emacs (yank), there is a
> long delay.  When I started seriously testing, it was always 5 seconds
> but I have memories that sometimes it was much longer.  I'm calling
> these memories "false" at this point.

[...]

> and then I see the text show up in the emacs window.  I *think* what
> is "wrong" here is in this case, get_input_pending is actually pulling
> the text out and via the SelectionNotify handler placing the text into
> reading_selection_reply but then returns zero because, at that point
> in time, there is no input to be read.  Thus this if test, is not
> true:
>
>      if ((read_kbd || !NILP (wait_for_cell))
> 	  && detect_input_pending ())
>
> I'm thinking that after detect_input_pending is called in
> wait_reading_process_output, another two lines like this:
>
>      if (! NILP (wait_for_cell) && ! NILP (XCAR (wait_for_cell)))
> 	break;
>
> should be added before the if (!no_avail) line (around line 4612 in
> the copy I am looking at.
>
> I made this change in my copy and so far, it seems to work.

(This bug report unfortunately got no response at the time.)

The wait_reading_process_output function has changed a bit over the
years since this was first reported.  Are you still seeing this issue in
more recent Emacs versions?

If so, could you include a proposed patch?

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

bug#16315: Slow paste

[Perry Smith]

I no longer work as the initial bug report said.  I’m not using AIX, RealVNC,
nor JollysFastVNC.  The original report had a change which I don’t recall
not working but if I’m the only person reporting the issue, I would just close
it and leave sleeping dogs lie.

> On Dec 6, 2020, at 9:11 AM, Lars Ingebrigtsen <larsi@gnus.org> wrote:
> 
> Perry Smith <pedzsan@gmail.com> writes:
> 
>> When I copy from a Mac window and paste into emacs (yank), there is a
>> long delay.  When I started seriously testing, it was always 5 seconds
>> but I have memories that sometimes it was much longer.  I'm calling
>> these memories "false" at this point.
> 
> [...]
> 
>> and then I see the text show up in the emacs window.  I *think* what
>> is "wrong" here is in this case, get_input_pending is actually pulling
>> the text out and via the SelectionNotify handler placing the text into
>> reading_selection_reply but then returns zero because, at that point
>> in time, there is no input to be read.  Thus this if test, is not
>> true:
>> 
>>     if ((read_kbd || !NILP (wait_for_cell))
>> 	  && detect_input_pending ())
>> 
>> I'm thinking that after detect_input_pending is called in
>> wait_reading_process_output, another two lines like this:
>> 
>>     if (! NILP (wait_for_cell) && ! NILP (XCAR (wait_for_cell)))
>> 	break;
>> 
>> should be added before the if (!no_avail) line (around line 4612 in
>> the copy I am looking at.
>> 
>> I made this change in my copy and so far, it seems to work.
> 
> (This bug report unfortunately got no response at the time.)
> 
> The wait_reading_process_output function has changed a bit over the
> years since this was first reported.  Are you still seeing this issue in
> more recent Emacs versions?
> 
> If so, could you include a proposed patch?
> 
> -- 
> (domestic pets only, the antidote for overdose, milk.)
>   bloggy blog: http://lars.ingebrigtsen.no

bug#16315: Slow paste

[Lars Ingebrigtsen]

Perry Smith <pedzsan@gmail.com> writes:

> I no longer work as the initial bug report said.  I’m not using AIX, RealVNC,
> nor JollysFastVNC.  The original report had a change which I don’t recall
> not working but if I’m the only person reporting the issue, I would just close
> it and leave sleeping dogs lie.

OK; doesn't sound like we'll make further progress here, so I'm closing
this bug report.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no