Re: New implementation of pseudo console support (experimental)

[Thomas Wolff <towo@towo.net>]

Am 18.07.2020 um 07:05 schrieb Takashi Yano via Cygwin-developers:
> Hi Thomas,
>
> Thanks for testing.
>
> On Fri, 17 Jul 2020 16:59:56 +0200
> Thomas Wolff wrote:
>> Am 17.07.2020 um 14:47 schrieb Thomas Wolff:
>>> Am 17.07.2020 um 13:19 schrieb Corinna Vinschen:
>>>> Hi Takashi,
>>>>
>>>> On Jul  1 20:47, Takashi Yano via Cygwin-developers wrote:
>>>>> On Fri, 29 May 2020 00:40:24 +0900
>>>>> Takashi Yano via Cygwin-developers <cygwin-developers@cygwin.com>
>>>>> wrote:
>>>>>> On Tue, 26 May 2020 10:09:55 +0900
>>>>>> Takashi Yano via Cygwin-developers <cygwin-developers@cygwin.com>
>>>>>> wrote:
>>>>>>> On Mon, 25 May 2020 19:53:32 +0900
>>>>>>> Takashi Yano via Cygwin-developers <cygwin-developers@cygwin.com>
>>>>>>> wrote:
>>>>>>>> On Tue, 19 May 2020 22:40:18 +0900
>>>>>>>> Takashi Yano via Cygwin-developers <cygwin-developers@cygwin.com>
>>>>>>>> wrote:
>>>>>>>>> On Sat, 16 May 2020 16:47:35 +0900
>>>>>>>>> Takashi Yano via Cygwin-developers
>>>>>>>>> <cygwin-developers@cygwin.com> wrote:
>>>>>>>>>> On Sat, 16 May 2020 09:29:56 +0900
>>>>>>>>>> Takashi Yano via Cygwin-developers
>>>>>>>>>> <cygwin-developers@cygwin.com> wrote:
>>>>>>>>>>> Fix a small bug caused when stdio is redirected to another pty.
>>>>>>>>>> Fix another bug caused when stdio is redirected to another pty.
>>>>>>>>> Revise the patch to fit the current git head.
>>>>>>>> Revise the patch again to fit the current git head.
>>>>>>> Make app, which reads stdin, work under gdb.
>>>>>> * Prevent ResizePseudoConsole() calls unless the pty is resized.
>>>>>> * Revise the patch to fit the current git head.
>>>>> Revise the patch to fit the current git head.
>>>> are you satisfied with the code?  If you want to merge it,
>>>> I'd bump Cygwin to 3.2.
>>> (blush) I apologize, I had promised to run my test cases.
>>> Beginning with it, the first succeeded, the second failed:
>>> run notepad (from mintty), click back into terminal, enter ^Z
>>> It says "Stopped" but the process is gone.
>>>
>>> Continuing may test suite soon...
>> OK, so here are finally my updated test results:
>>
>>
>> resize terminal while running Windows cmd
>> run cmd, resize, run dir/P:
>> ✓works
>>
>>
>> terminal reports in response to request escape sequences
>> ("\033[6n", "\033[0c", "\033[>c", '\033[18t', '\033]10;?\033\')
>> ✓works
>>
>>
>> output to alternate screen
>> echo -e "\e[?1047h"; cmd
>> ✓works
>> cmd
>> from other terminal: echo -e "\e[?1047h" > /dev/pty...
>> ↯no output; weird behaviour on ^C, mintty terminating (also in 3.1.6)
> I found this is a bug introduced in commit
> 0365031ce1347600d854a23f30f1355745a1765c.
>
> I will submit a patch for this issue.
>
>> signal handling/mediation; catch SIGTSTP
>> run notepad, click back into terminal, enter ^Z
>> ↯fails, notepad is terminated
>> -> this is a regression, ^Z,^C,^\ used to be ignored
> This behaviour is the same as cygwin 3.0.7. Pseudo console is not
> activated for native GUI processes.
OK, I understand this is consistent with my own proposal to apply pseudo 
console only for Windows command-line programs. On the other hand, fatal 
signal handling is another nuisance of running Windows programs from 
cygwin, so there was a benefit.
Is it possible to activate the signal interworking without setting up a 
pseudo console? If not, is it maybe worth to set it up also for Windows 
GUI programs just for this purpose? (Modifying my proposal to trigger on 
any Windows program.)

>> character set conversion from Windows cmd line program
>> run xcopy (Windows system language e.g. German), watch error message
>> ✓works in Unicode terminal
>> ✓works in non-Unicode terminal (e.g. LC_ALL=C.CP850 mintty)
>>
>>
>> handling non-ASCII characters in non-Unicode terminal + native interworking
>> LC_ALL=en_US mintty, check locale charmap -> ISO-8859-1
>> echo ö | od -t x1 -> 0xF6
>> ✓works
>> cmd /c echo ö | od -t x1 -> 0x94
>> ↯fails (also in cygwin 3.1.6)
> In this case, stdout is not a pty. So this is not a pty problem.
>
>> using wincon.c, compiled with x86_64-w64-mingw32-gcc
>> run program, check colour of second output, give non-ASCII input, check echo
>> ✓works in Unicode mintty
>> ✓works in non-Unicode mintty
>>
>>
>> using attached wincon.c, compiled with gcc
>> run program, check colour of second output, give non-ASCII input, check echo
>> ↯no output (regression, works in 3.1.6)
> This is due to the disadvantage 3) I mentioned in the first post.
> Disadvantage:
> 3) The cygwin program which call console API directly does not work.
Understood, I think this is an acceptable trade-off.

>> using attached program sgr.java
>> run program, check coloured output of middle characters
>> ✓works (did not in 3.1.6)
> This may work by chance. Escape sequences in native console apps
> are supported only with ENABLE_VIRTUAL_TERMINAL_PROCESSING.
So this seems to be enabled by the Java runtime. Would be nice if there 
were a Windows option (exe file property) to enforce the mode even for 
programs that do not actively invoke it. I've filed an issue 
(https://github.com/microsoft/terminal/issues/6973) to this aim.

Kind regards,
Thomas