Question on pseudo console and legacy console

Question on pseudo console and legacy console

[Koichi Murase]

I have a question on the pseudo console mode and the legacy console
mode.

Question:

  Does it cause problems to set `disable_pcon' in the default value of
  the environment variable `CYGWIN'?  When is the pseudo console mode
  recommended, and when can I use the legacy console mode?  I noticed
  that the terminal emulators with the pseudo console mode is quite
  slower than with the legacy console mode, so I am thinking of
  setting `disable_pcon' by default if it does not cause problem.


Background:

* There are large performance differences of terminal emulators
  between the legacy and pseudo console modes.  I show some test and
  results in my laptop as follows:

  For example, one can measure a performance of the terminal emulators
  by the following commands.  It measures the processing time of the
  terminal for 1M lines.

  $ yes | head -1000000 > yes.txt
  $ time cat yes.txt

  The results with different terminals and different console modes
  are summarized below:

  urxvt     ... legacy: 0.410s,   pcon: 33.517s   (~ 80x slower)
  (A) win   ... legacy: 1.593s,   pcon: 36.064s   (~ 20x slower)
  (B) x11   ... legacy: 1.573s,   pcon: 35.611s   (~ 20x slower)
  (C) tty   ... legacy: 1.614s,   pcon: 35.129s   (~ 20x slower)
  mintty    ... legacy: 3.811s,   pcon: 39.406s   (~ 10x slower)
  xterm -j  ... legacy: 39.809s,  pcon: 1m2.237s  (~ 1.5x slower)

  [ Note: `legacy' and `pcon' are the legacy and pseudo console
  modes, respectively.  (A)--(C) are three different modes of the
  terminal that I wrote by myself and also I primarily use daily.
  When xterm is tested, the option -j is passed to explicitly turn
  off the smooth scrolling mode. ]

  This is the results for a specific computer, but the difference
  between legacy and pseudo console modes is clear.  I think the
  reason why pseudo console is so slow is that it actually
  internally processes terminal sequences and constructs its
  terminal contents in backgrounds, which is completely redundant
  process when the user uses terminal emulators.  And the background
  terminal is as slow as xterm, which is at least ten times slower
  than the other terminal emulators.

* I understand that, with the pseudo console mode, the Cygwin console
  have more powerful terminal features compared to the legacy console
  mode.  However, I do not usually use the console but another
  terminal emulator.  Also the default Cygwin Terminal (which can be
  launched from the shortcut icon on desktop or in the start menu) is
  actually Mintty, so I believe most other people also use terminal
  emulators.

  Is there any reason to enable the pseudo console mode for all the
  programs including those which does not use the console window?  For
  example, does the Cygwin PTY with the legacy console mode have some
  limitation which is not present with the pseudo console mode?  Or,
  are there some problems caused by the legacy console mode?  I
  thought maybe some Windows Command-Line application (based on
  Windows Console API) could have troubles with the legacy console
  mode, but with a quick check for `cmd' and `PowerShell' it appears
  to work in the terminal emulators with the legacy console mode
  though I haven't tested it thoroughly.

* For these reasons, if there is no functional differences between the
  pseudo console mode and the legacy console mode as far as I use
  terminal emulators, I tend to think about setting `disable_pcon' as
  the default value of the environment variable `CYGWIN'.  But looking
  at the discussion at

  https://cygwin.com/ml/cygwin-patches/2020-q1/threads.html#00060

  it appears the setting `disable_pcon' is only introduced as a
  workaround for the programs incompatible with the pseudo console
  mode (such as `cgdb'), and the pseudo console mode is still
  considered to be appropriate for normal programs.

* If the console mode would not cause any differences when there is no
  visible console window, would it make sense to change Cygwin's
  behavior so that it switches to the pseudo console mode only when
  the console window is present (if it is not too difficult
  technically)?


Best regards,

Koichi

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Question on pseudo console and legacy console

[Takashi Yano]

On Sat, 8 Feb 2020 22:50:56 +0800
Koichi Murase wrote:
> I have a question on the pseudo console mode and the legacy console
> mode.
> 
> Question:
> 
>   Does it cause problems to set `disable_pcon' in the default value of
>   the environment variable `CYGWIN'?  When is the pseudo console mode
>   recommended, and when can I use the legacy console mode?  I noticed
>   that the terminal emulators with the pseudo console mode is quite
>   slower than with the legacy console mode, so I am thinking of
>   setting `disable_pcon' by default if it does not cause problem.

Do you mean "without pseudo console support" by "legacy console mode"?
If so, pty should behave as before 3.1.0 if you set disable_pcon,
except that char code conversion, from code page which you set, to char
code which is declared by locale, is still enabled.

> Background:
> 
> * There are large performance differences of terminal emulators
>   between the legacy and pseudo console modes.  I show some test and
>   results in my laptop as follows:
> 
>   For example, one can measure a performance of the terminal emulators
>   by the following commands.  It measures the processing time of the
>   terminal for 1M lines.
> 
>   $ yes | head -1000000 > yes.txt
>   $ time cat yes.txt
> 
>   The results with different terminals and different console modes
>   are summarized below:
> 
>   urxvt     ... legacy: 0.410s,   pcon: 33.517s   (~ 80x slower)
>   (A) win   ... legacy: 1.593s,   pcon: 36.064s   (~ 20x slower)
>   (B) x11   ... legacy: 1.573s,   pcon: 35.611s   (~ 20x slower)
>   (C) tty   ... legacy: 1.614s,   pcon: 35.129s   (~ 20x slower)
>   mintty    ... legacy: 3.811s,   pcon: 39.406s   (~ 10x slower)
>   xterm -j  ... legacy: 39.809s,  pcon: 1m2.237s  (~ 1.5x slower)
> 
>   [ Note: `legacy' and `pcon' are the legacy and pseudo console
>   modes, respectively.  (A)--(C) are three different modes of the
>   terminal that I wrote by myself and also I primarily use daily.
>   When xterm is tested, the option -j is passed to explicitly turn
>   off the smooth scrolling mode. ]
> 
>   This is the results for a specific computer, but the difference
>   between legacy and pseudo console modes is clear.  I think the
>   reason why pseudo console is so slow is that it actually
>   internally processes terminal sequences and constructs its
>   terminal contents in backgrounds, which is completely redundant
>   process when the user uses terminal emulators.  And the background
>   terminal is as slow as xterm, which is at least ten times slower
>   than the other terminal emulators.

This seems to be too much than I expected... 

> * I understand that, with the pseudo console mode, the Cygwin console
>   have more powerful terminal features compared to the legacy console
>   mode.  However, I do not usually use the console but another
>   terminal emulator.  Also the default Cygwin Terminal (which can be
>   launched from the shortcut icon on desktop or in the start menu) is
>   actually Mintty, so I believe most other people also use terminal
>   emulators.
> 
>   Is there any reason to enable the pseudo console mode for all the
>   programs including those which does not use the console window?  For
>   example, does the Cygwin PTY with the legacy console mode have some
>   limitation which is not present with the pseudo console mode?  Or,
>   are there some problems caused by the legacy console mode?  I
>   thought maybe some Windows Command-Line application (based on
>   Windows Console API) could have troubles with the legacy console
>   mode, but with a quick check for `cmd' and `PowerShell' it appears
>   to work in the terminal emulators with the legacy console mode
>   though I haven't tested it thoroughly.

With out pseudo console, for example:
1) cmd.exe echos command twice.
2) dir /p does not work in cmd.exe.
3) Many of windows native console apps such as windows native gnuplot
   does not work at all.

> * For these reasons, if there is no functional differences between the
>   pseudo console mode and the legacy console mode as far as I use
>   terminal emulators, I tend to think about setting `disable_pcon' as
>   the default value of the environment variable `CYGWIN'.  But looking
>   at the discussion at
> 
>   https://cygwin.com/ml/cygwin-patches/2020-q1/threads.html#00060
> 
>   it appears the setting `disable_pcon' is only introduced as a
>   workaround for the programs incompatible with the pseudo console
>   mode (such as `cgdb'), and the pseudo console mode is still
>   considered to be appropriate for normal programs.
> 
> * If the console mode would not cause any differences when there is no
>   visible console window, would it make sense to change Cygwin's
>   behavior so that it switches to the pseudo console mode only when
>   the console window is present (if it is not too difficult
>   technically)?

I am not sure what you mean "the console window is present", but
that is possible if you accept the behaviour that the output of
cygwin native apps disappears everytime when windows native app
is executed.

That is for example:

Start mintty, then shell prompt displayed as:
----
[yano@Express5800-S70 ~]$
----

Execute chcp 65001, the first shell prompt disappears.
----
Active code page: 65001
[yano@Express5800-S70 ~]$
----

Execute ps.
----
Active code page: 65001
[yano@Express5800-S70 ~]$ ps
      PID    PPID    PGID     WINPID   TTY         UID    STIME COMMAND
     1334    1330    1334       1052  pty0      197609 08:36:01 /usr/bin/ps
     1329       1    1329       1524  ?         197609 08:33:25 /usr/bin/mintty
     1330    1329    1330      13544  pty0      197609 08:33:25 /usr/bin/tcsh
[yano@Express5800-S70 ~]$
----

Execute chcp 65001 again, then shell prompt and result of ps
disappears.
----
Active code page: 65001
Active code page: 65001
[yano@Express5800-S70 ~]$
----

-- 
Takashi Yano <takashi.yano@nifty.ne.jp>

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Question on pseudo console and legacy console

[Koichi Murase]

Thank you for the explanation.  I think I'm going to by default add
`disable_pcon' to the environment variable `CYGWIN' before calling
`posix_openpt'.

2020年2月9日(日) 7:43 Takashi Yano
> Do you mean "without pseudo console support" by "legacy console mode"?

Yes.

> If so, pty should behave as before 3.1.0 if you set disable_pcon,

I see. Actually I was particularly interested in whether the
introduction of the pseudo console support is related to solving any
existing problems before 3.1.0 or not.

> except that char code conversion, from code page which you set, to
> char code which is declared by locale, is still enabled.

I haven't recognized this change. This is really nice!

> With out pseudo console, for example:
> 1) cmd.exe echos command twice.
> 2) dir /p does not work in cmd.exe.
> 3) Many of windows native console apps such as windows native gnuplot
>    does not work at all.

OK, there are in fact functional differences between the pseudo
console mode and the legacy console mode.  So, is the purpose of the
introduction of the pseudo console mode is not just to enhance the
Cygwin console but to make Windows Command-Line applications work
properly in all Cygwin terminals including terminal emulators as well
as console?

Now I understand the difficulties of the pseudo console modes.  This
is really challenging as Windows Console API and ANSI/VT terminal
sequences are independent ecosystems and based on different models.
I'm not sure if it is even possible to make it realize in a completely
transparent way for both types (Windows and Cygwin) of applications.

> > * If the console mode would not cause any differences when there
> >   is no visible console window, would it make sense to change
> >   Cygwin's behavior so that it switches to the pseudo console mode
> >   only when the console window is present (if it is not too
> >   difficult technically)?
>
> I am not sure what you mean "the console window is present", but

I am sorry I did not put it clearly.  By "the console window is
present", I meant the case that a console will be allocated for the
terminal emulator but the corresponding GUI console window will not be
shown in the foreground.  My suggestion was to choose/switch a mode
for each PTY session but not for each program.

Specifically I thought about completely disabling the pseudo console
support with terminal emulators, and turning on it only for Cygwin
consoles.  But this is based on my assumption that the pseudo console
mode is aimed to provide enhanced Cygwin consoles and does not affect
the functional differences in terminal emulators.

----

Nevertheless, IMHO, the current implementation of the pseudo console
mode is better to be turned off for terminal emulators by default and
only turned on for Cygwin consoles.  Of course, it can still
optionally be turned on in terminal emulators when the user or the
terminal emulator explicitly request that.

The TTY/PTY mechanism should be transparent to the types of terminals,
but the current implementation of PTY by the pseudo console mode
requires a specific set of terminal capabilities and does not allow
any extensions which is not in the pseudo console.  This is not a good
design as a basic system API.  If this pseudo console mode is
defaulted for Cygwin PTY, that means any terminal emulator cannot be
implemented properly on Cygwin except the case that the terminal
emulator is a perfect copy of the pseudo console.

In fact there are already many reports caused by the behavior
difference of the pseudo console from an expected one.  I don't think
this kind of troubles stops by just modifying the superficial behavior
of the current implementation.  We want to avoid supporting new
features (Windows Command-Line applications) by breaking many existing
applications which opens PTY.  I think it is better to wait until we
could find a way to make the PTY behavior transparent to the terminal
types.

----

Thank you,

Koichi

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Question on pseudo console and legacy console

[Thomas Wolff]

Am 09.02.2020 um 04:20 schrieb Koichi Murase:
> Specifically I thought about completely disabling the pseudo console
> support with terminal emulators, and turning on it only for Cygwin
> consoles.  But this is based on my assumption that the pseudo console
> mode is aimed to provide enhanced Cygwin consoles and does not affect
> the functional differences in terminal emulators.
Actually it's just the other way round. The pseudo console serves the 
adaptation of client-side (i.e. native Windows programs) console usage 
to server-side (i.e. pty-accessed terminals) interaction, transforming 
character sets, raw input, signal handling. Most of this is not needed 
inside the Windows console / "Cygwin console".

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple