[readline-mingw] backspace key and TUI size

[readline-mingw] backspace key and TUI size

[Denis PILAT]

[-- Attachment #1: Type: text/plain, Size: 618 bytes --]

Hi,

When compiling gdb 6.4 for native window host (minGW) I get a gdb.exe 
and a gdbtui.exe where there were 2 problems.
- In the gdbtui.exe version, the TUI interface do not fit in the DOS 
command window.
- In the gdb.exe version, the backspace key do not behave correctly.

I did some little modification in readline to fix both problems. I 
attach the patch applied.
I've already submitted it to Chet Ramey (the readline maintainer at 
bash-maintainers@gnu.org) who accepted it.
I don't know the gdb policy with readline synchronization, but I would 
like it  to be integrated as well in gdb distribution.

Denis

[-- Attachment #2: 20060206-mingw-readline-tuisize-backspace.patch --]
[-- Type: text/plain, Size: 1408 bytes --]

Index: readline/terminal.c
===================================================================
--- readline/terminal.c	(revision 328)
+++ readline/terminal.c	(working copy)
@@ -70,6 +70,11 @@
 #include "rlshell.h"
 #include "xmalloc.h"
 
+#if defined (__MINGW32__)
+#  include <windows.h>
+#  include <wincon.h>
+#endif
+
 #define CUSTOM_REDISPLAY_FUNC() (rl_redisplay_function != rl_redisplay)
 #define CUSTOM_INPUT_FUNC() (rl_getc_function != rl_getc)
 
@@ -201,6 +206,15 @@
   struct winsize window_size;
 #endif /* TIOCGWINSZ */
 
+/* For mingw version, we get console size from windows API*/
+#if defined (__MINGW32__)
+  HANDLE hConOut = GetStdHandle(STD_OUTPUT_HANDLE);
+  CONSOLE_SCREEN_BUFFER_INFO scr;  
+  GetConsoleScreenBufferInfo(hConOut, &scr);
+  _rl_screenwidth = scr.dwSize.X;
+  _rl_screenheight = scr.srWindow.Bottom - scr.srWindow.Top + 1;
+#endif
+
 #if defined (TIOCGWINSZ)
   if (ioctl (tty, TIOCGWINSZ, &window_size) == 0)
     {
@@ -299,6 +313,7 @@
 void
 rl_resize_terminal ()
 {
+printf("rl_resize_terminal\n");
   if (readline_echoing_p)
     {
       _rl_get_screen_size (fileno (rl_instream), 1);
@@ -593,8 +608,8 @@
 {
   register int i;
 
-#ifndef __MSDOS__
-  if (_rl_term_backspace)
+#if !defined (__MSDOS__) && !defined (__MINGW32__)
+  if (_rl_term_backspace) 
     for (i = 0; i < count; i++)
       tputs (_rl_term_backspace, 1, _rl_output_character_function);
   else

Re: [readline-mingw] backspace key and TUI size

[Daniel Jacobowitz]

On Wed, Feb 08, 2006 at 09:02:57AM +0100, Denis PILAT wrote:
> Hi,
> 
> When compiling gdb 6.4 for native window host (minGW) I get a gdb.exe 
> and a gdbtui.exe where there were 2 problems.
> - In the gdbtui.exe version, the TUI interface do not fit in the DOS 
> command window.
> - In the gdb.exe version, the backspace key do not behave correctly.
> 
> I did some little modification in readline to fix both problems. I 
> attach the patch applied.
> I've already submitted it to Chet Ramey (the readline maintainer at 
> bash-maintainers@gnu.org) who accepted it.
> I don't know the gdb policy with readline synchronization, but I would 
> like it  to be integrated as well in gdb distribution.

We can import readline patches; I assume Chet didn't take this as-is
(it has a stray printf in it).  Chet, if you have a chance, could you
send us the version you've applied, so we can merge it?

Denis, I'm not sure why the backspace fix was necessary for you; my
backspace works fine.  What did this fix?

-- 
Daniel Jacobowitz
CodeSourcery

Re: [readline-mingw] backspace key and TUI size

[Denis PILAT]

Daniel Jacobowitz wrote:

>On Wed, Feb 08, 2006 at 09:02:57AM +0100, Denis PILAT wrote:
>  
>
>>Hi,
>>
>>When compiling gdb 6.4 for native window host (minGW) I get a gdb.exe 
>>and a gdbtui.exe where there were 2 problems.
>>- In the gdbtui.exe version, the TUI interface do not fit in the DOS 
>>command window.
>>- In the gdb.exe version, the backspace key do not behave correctly.
>>
>>I did some little modification in readline to fix both problems. I 
>>attach the patch applied.
>>I've already submitted it to Chet Ramey (the readline maintainer at 
>>bash-maintainers@gnu.org) who accepted it.
>>I don't know the gdb policy with readline synchronization, but I would 
>>like it  to be integrated as well in gdb distribution.
>>    
>>
>
>We can import readline patches; I assume Chet didn't take this as-is
>(it has a stray printf in it).  Chet, if you have a chance, could you
>send us the version you've applied, so we can merge it?
>
>Denis, I'm not sure why the backspace fix was necessary for you; my
>backspace works fine.  What did this fix?
>
>  
>
You're right, I forgot to remove a printf ...

Our host is i686-pc-mingw32, gdb is compiled under cygwin with 
"-mno-cygwin" flag to avoid cygwin dependencies.
We also use pdcurses library since ncurse is not available on windows.
I think that's why you don't see the problem. With this patch, I'm not 
using the tputs() function anymore on Windows, as it was not used for MSDOS.

Denis

Re: [readline-mingw] backspace key and TUI size

[Bob Rossi]

On Wed, Feb 08, 2006 at 05:03:33PM +0100, Denis PILAT wrote:
> Daniel Jacobowitz wrote:
> 
> >On Wed, Feb 08, 2006 at 09:02:57AM +0100, Denis PILAT wrote:
> > 
> >
> >>Hi,
> >>
> >>When compiling gdb 6.4 for native window host (minGW) I get a gdb.exe 
> >>and a gdbtui.exe where there were 2 problems.
> >>- In the gdbtui.exe version, the TUI interface do not fit in the DOS 
> >>command window.
> >>- In the gdb.exe version, the backspace key do not behave correctly.
> >>
> >>I did some little modification in readline to fix both problems. I 
> >>attach the patch applied.
> >>I've already submitted it to Chet Ramey (the readline maintainer at 
> >>bash-maintainers@gnu.org) who accepted it.
> >>I don't know the gdb policy with readline synchronization, but I would 
> >>like it  to be integrated as well in gdb distribution.
> >>   
> >>
> >
> >We can import readline patches; I assume Chet didn't take this as-is
> >(it has a stray printf in it).  Chet, if you have a chance, could you
> >send us the version you've applied, so we can merge it?
> >
> >Denis, I'm not sure why the backspace fix was necessary for you; my
> >backspace works fine.  What did this fix?
> >
> > 
> >
> You're right, I forgot to remove a printf ...
> 
> Our host is i686-pc-mingw32, gdb is compiled under cygwin with 
> "-mno-cygwin" flag to avoid cygwin dependencies.
> We also use pdcurses library since ncurse is not available on windows.
> I think that's why you don't see the problem. With this patch, I'm not 
> using the tputs() function anymore on Windows, as it was not used for MSDOS.

Have you considered using the ncurses port from Cygwin?

Bob Rossi

Re: [readline-mingw] backspace key and TUI size

[Daniel Jacobowitz]

On Wed, Feb 08, 2006 at 05:03:33PM +0100, Denis PILAT wrote:
> Our host is i686-pc-mingw32, gdb is compiled under cygwin with 
> "-mno-cygwin" flag to avoid cygwin dependencies.
> We also use pdcurses library since ncurse is not available on windows.
> I think that's why you don't see the problem. With this patch, I'm not 
> using the tputs() function anymore on Windows, as it was not used for MSDOS.

Ah - we (CodeSourcery) don't use a curses library at all; just the
standard Windows and MinGW DLLs.  See win32-termcap.c.  Are those
functions not being used for you?

I just don't think the patch could be right.  If you're using the same
termcap layer that my builds are using, it should have no effect; but
if you're using a termcap library that gives you a real backspace
capability string, and outputing that does the wrong thing, then either
your pdcurses library or your termcap entry for the Windows console
sounds pretty broken to me.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [readline-mingw] backspace key and TUI size

[Chet Ramey]

Daniel Jacobowitz wrote:

> 
> I just don't think the patch could be right.  If you're using the same
> termcap layer that my builds are using, it should have no effect; but
> if you're using a termcap library that gives you a real backspace
> capability string, and outputing that does the wrong thing, then either
> your pdcurses library or your termcap entry for the Windows console
> sounds pretty broken to me.

I agree.  If incorrect behavior happens only with that curses library
in place, then the termcap description is probably wrong.

Chet

-- 
``The lyf so short, the craft so long to lerne.'' - Chaucer
( ``Discere est Dolere'' -- chet )
						Live Strong.
Chet Ramey, ITS, CWRU    chet@case.edu    http://tiswww.tis.case.edu/~chet/

Re: [readline-mingw] backspace key and TUI size

[Andrew STUBBS]

Daniel Jacobowitz wrote:
> On Wed, Feb 08, 2006 at 05:03:33PM +0100, Denis PILAT wrote:
>> Our host is i686-pc-mingw32, gdb is compiled under cygwin with 
>> "-mno-cygwin" flag to avoid cygwin dependencies.
>> We also use pdcurses library since ncurse is not available on windows.
>> I think that's why you don't see the problem. With this patch, I'm not 
>> using the tputs() function anymore on Windows, as it was not used for MSDOS.
> 
> Ah - we (CodeSourcery) don't use a curses library at all; just the
> standard Windows and MinGW DLLs.  See win32-termcap.c.  Are those
> functions not being used for you?

No, we don't use those functions. The win32-termcap doesn't support the 
TUI. PDcurses does (mostly), but the symbols clash, so win32-termcap is 
disabled.

Andrew

Re: [readline-mingw] backspace key and TUI size

[Daniel Jacobowitz]

On Wed, Feb 08, 2006 at 04:45:15PM +0000, Andrew STUBBS wrote:
> Daniel Jacobowitz wrote:
> >On Wed, Feb 08, 2006 at 05:03:33PM +0100, Denis PILAT wrote:
> >>Our host is i686-pc-mingw32, gdb is compiled under cygwin with 
> >>"-mno-cygwin" flag to avoid cygwin dependencies.
> >>We also use pdcurses library since ncurse is not available on windows.
> >>I think that's why you don't see the problem. With this patch, I'm not 
> >>using the tputs() function anymore on Windows, as it was not used for 
> >>MSDOS.
> >
> >Ah - we (CodeSourcery) don't use a curses library at all; just the
> >standard Windows and MinGW DLLs.  See win32-termcap.c.  Are those
> >functions not being used for you?
> 
> No, we don't use those functions. The win32-termcap doesn't support the 
> TUI. PDcurses does (mostly), but the symbols clash, so win32-termcap is 
> disabled.

Please, guys, if you're going to submit patches that only matter with
other configuration changes and patches, say so!

I'd like to understand what goes wrong for you and why outputing
whatever pdcurses is giving you for backspace doesn't work.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [readline-mingw] backspace key and TUI size

[Chet Ramey]

Daniel Jacobowitz wrote:

> Please, guys, if you're going to submit patches that only matter with
> other configuration changes and patches, say so!
> 
> I'd like to understand what goes wrong for you and why outputing
> whatever pdcurses is giving you for backspace doesn't work.

Can we all agree that the screensize code is valuable, at least?

Chet

-- 
``The lyf so short, the craft so long to lerne.'' - Chaucer
( ``Discere est Dolere'' -- chet )
						Live Strong.
Chet Ramey, ITS, CWRU    chet@case.edu    http://tiswww.tis.case.edu/~chet/

Re: [readline-mingw] backspace key and TUI size

[Daniel Jacobowitz]

On Wed, Feb 08, 2006 at 12:03:32PM -0500, Chet Ramey wrote:
> Daniel Jacobowitz wrote:
> 
> > Please, guys, if you're going to submit patches that only matter with
> > other configuration changes and patches, say so!
> > 
> > I'd like to understand what goes wrong for you and why outputing
> > whatever pdcurses is giving you for backspace doesn't work.
> 
> Can we all agree that the screensize code is valuable, at least?

Absolutely!  Eli pointed out that it needs some error checking; this
is somewhat important, because stdin might not be connected to a
console (e.g. a pipe).  But other than that it looked right to me.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [readline-mingw] backspace key and TUI size

[Eli Zaretskii]

> Date: Wed, 8 Feb 2006 12:13:23 -0500
> From: Daniel Jacobowitz <drow@false.org>
> Cc: Andrew STUBBS <andrew.stubbs@st.com>, Denis PILAT <denis.pilat@st.com>, 	gdb-patches@sources.redhat.com, bash-maintainers@gnu.org, 	chet@case.edu
> 
> On Wed, Feb 08, 2006 at 12:03:32PM -0500, Chet Ramey wrote:
> > Daniel Jacobowitz wrote:
> > 
> > > Please, guys, if you're going to submit patches that only matter with
> > > other configuration changes and patches, say so!
> > > 
> > > I'd like to understand what goes wrong for you and why outputing
> > > whatever pdcurses is giving you for backspace doesn't work.
> > 
> > Can we all agree that the screensize code is valuable, at least?
> 
> Absolutely!  Eli pointed out that it needs some error checking; this
> is somewhat important, because stdin might not be connected to a
> console (e.g. a pipe).  But other than that it looked right to me.

Yes, apart of the bit with error checking, the MinGW screen-size code
is okay with me.

Re: [readline-mingw] backspace key and TUI size

[Eli Zaretskii]

> Date: Wed, 8 Feb 2006 11:51:14 -0500
> From: Daniel Jacobowitz <drow@false.org>
> Cc: Denis PILAT <denis.pilat@st.com>, gdb-patches@sources.redhat.com, 	bash-maintainers@gnu.org
> 
> I'd like to understand what goes wrong for you and why outputing
> whatever pdcurses is giving you for backspace doesn't work.

Seconded.  Please explain why Backspace doesn't work (if you don't
know, please debug this on your system and see what you come up with).
Thanks.

Re: [readline-mingw] backspace key and TUI size

[Andrew STUBBS]

Eli Zaretskii wrote:
>> Date: Wed, 8 Feb 2006 11:51:14 -0500
>> From: Daniel Jacobowitz <drow@false.org>
>> Cc: Denis PILAT <denis.pilat@st.com>, gdb-patches@sources.redhat.com, 	bash-maintainers@gnu.org
>>
>> I'd like to understand what goes wrong for you and why outputing
>> whatever pdcurses is giving you for backspace doesn't work.
> 
> Seconded.  Please explain why Backspace doesn't work (if you don't
> know, please debug this on your system and see what you come up with).
> Thanks.

It does work in MSYS terminals. In DOS/Windows or Cygwin terminals it 
moves the cursor right, instead of left, and leaves the deleted 
character on the screen. I assume that this curses just doesn't support 
those terminals in this respect.

I don't really understand why. I hope it will go away in future 
versions. I don't suppose anybody knows of a better/alternative curses 
implementation that doesn't require cygwin?

PDcurses does work for the TUI, which is nice. Except in Cygwin xterm, 
but you can't have everything.

I thought this patch was going to remain internal, at least until the 
MinGW.org GDB patches have been properly contributed.

Andrew

Re: [readline-mingw] backspace key and TUI size

[Daniel Jacobowitz]

On Wed, Feb 08, 2006 at 05:28:25PM +0000, Andrew STUBBS wrote:
> Eli Zaretskii wrote:
> >>Date: Wed, 8 Feb 2006 11:51:14 -0500
> >>From: Daniel Jacobowitz <drow@false.org>
> >>Cc: Denis PILAT <denis.pilat@st.com>, gdb-patches@sources.redhat.com,  
> >>bash-maintainers@gnu.org
> >>
> >>I'd like to understand what goes wrong for you and why outputing
> >>whatever pdcurses is giving you for backspace doesn't work.
> >
> >Seconded.  Please explain why Backspace doesn't work (if you don't
> >know, please debug this on your system and see what you come up with).
> >Thanks.
> 
> It does work in MSYS terminals. In DOS/Windows or Cygwin terminals it 
> moves the cursor right, instead of left, and leaves the deleted 
> character on the screen. I assume that this curses just doesn't support 
> those terminals in this respect.

Well that's why then.

The way a termcap library works is that it describes the capabilities
of each individual terminal that it supports; the control sequences it
outputs rely on knowing in advance how a particular terminal will
respond to various conditions.

If the mingw build of pdcurses works in an MSYS terminal and explodes
in a DOS console, then the MSYS terminal must implement something
different than the DOS console; if the curses library is not aware
of the difference, things will break.  Please don't adjust readline
for this; instead get pdcurses not to report the capability or to
report it correctly.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [readline-mingw] backspace key and TUI size

[Denis PILAT]

>
>
>If the mingw build of pdcurses works in an MSYS terminal and explodes
>in a DOS console, then the MSYS terminal must implement something
>different than the DOS console; if the curses library is not aware
>of the difference, things will break.  Please don't adjust readline
>for this; instead get pdcurses not to report the capability or to
>report it correctly.
>
>  
>
You're right Daniel, patching readline for that is not the best thing to 
do. I'm going to have a look at pdcurses sources instead.
Thanks anyway for your comments.
Regarding the TUI size, the approach seems to be fine, can I send you a 
new patch whith a ChangeLog for that ?

Denis

Re: [readline-mingw] backspace key and TUI size

[Daniel Jacobowitz]

On Thu, Feb 09, 2006 at 10:35:53AM +0100, Denis PILAT wrote:
> >
> >
> >If the mingw build of pdcurses works in an MSYS terminal and explodes
> >in a DOS console, then the MSYS terminal must implement something
> >different than the DOS console; if the curses library is not aware
> >of the difference, things will break.  Please don't adjust readline
> >for this; instead get pdcurses not to report the capability or to
> >report it correctly.
> >
> > 
> >
> You're right Daniel, patching readline for that is not the best thing to 
> do. I'm going to have a look at pdcurses sources instead.
> Thanks anyway for your comments.
> Regarding the TUI size, the approach seems to be fine, can I send you a 
> new patch whith a ChangeLog for that ?

Yes, please.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [readline-mingw] backspace key and TUI size

[Denis PILAT]

[-- Attachment #1: Type: text/plain, Size: 235 bytes --]



Daniel Jacobowitz wrote

>>Regarding the TUI size, the approach seems to be fine, can I send you a 
>>new patch whith a ChangeLog for that ?
>>    
>>
>
>Yes, please.
>
>  
>
Attached is the new proposal with only TUI stuffs.

Denis

[-- Attachment #2: readline.patch --]
[-- Type: text/plain, Size: 1037 bytes --]

Index: terminal.c
===================================================================
--- terminal.c	(revision 264)
+++ terminal.c	(working copy)
@@ -70,6 +70,11 @@
 #include "rlshell.h"
 #include "xmalloc.h"
 
+#if defined (__MINGW32__)
+#  include <windows.h>
+#  include <wincon.h>
+#endif
+
 #define CUSTOM_REDISPLAY_FUNC() (rl_redisplay_function != rl_redisplay)
 #define CUSTOM_INPUT_FUNC() (rl_getc_function != rl_getc)
 
@@ -201,6 +206,20 @@
   struct winsize window_size;
 #endif /* TIOCGWINSZ */
 
+/* For mingw version, we get console size from windows API*/
+#if defined (__MINGW32__)
+  HANDLE hConOut = GetStdHandle(STD_OUTPUT_HANDLE);
+  if ( hConOut != INVALID_HANDLE_VALUE ) 
+    {
+      CONSOLE_SCREEN_BUFFER_INFO scr;  
+      if ( GetConsoleScreenBufferInfo(hConOut, &scr) )
+        {
+          _rl_screenwidth = scr.dwSize.X;
+          _rl_screenheight = scr.srWindow.Bottom - scr.srWindow.Top + 1;
+        }
+    }
+#endif
+
 #if defined (TIOCGWINSZ)
   if (ioctl (tty, TIOCGWINSZ, &window_size) == 0)
     {

[-- Attachment #3: ChangeLog.gdb --]
[-- Type: text/plain, Size: 159 bytes --]

2005-02-10  Denis Pilat <denis.pilat@st.com>

	* readline/terminal.c
	(_rl_get_screen_size): we get console size from windows API
	when compiling with minGW. 

Re: [readline-mingw] backspace key and TUI size

[Eli Zaretskii]

> Date: Fri, 10 Feb 2006 16:41:29 +0100
> From: Denis PILAT <denis.pilat@st.com>
> Cc: Andrew STUBBS <andrew.stubbs@st.com>, Eli Zaretskii <eliz@gnu.org>,
>         gdb-patches@sources.redhat.com, bash-maintainers@gnu.org
> 
> Daniel Jacobowitz wrote
> 
> >>Regarding the TUI size, the approach seems to be fine, can I send you a 
> >>new patch whith a ChangeLog for that ?
> >>    
> >>
> >
> >Yes, please.
> >
> >  
> >
> Attached is the new proposal with only TUI stuffs.

This one is fine with me.  Thanks!

Re: [readline-mingw] backspace key and TUI size

[Daniel Jacobowitz]

On Fri, Feb 10, 2006 at 04:41:29PM +0100, Denis PILAT wrote:
> 
> 
> Daniel Jacobowitz wrote
> 
> >>Regarding the TUI size, the approach seems to be fine, can I send you a 
> >>new patch whith a ChangeLog for that ?
> >>   
> >>
> >
> >Yes, please.
> >
> > 
> >
> Attached is the new proposal with only TUI stuffs.

Just to close the loop on this patch:

This was fine with Eli and it's fine with me.  Chet, if you've picked
this up for readline, please let me know and I'll merge it to GDB's
copy.

Thanks Denis!

> 2005-02-10  Denis Pilat <denis.pilat@st.com>
> 
> 	* readline/terminal.c
> 	(_rl_get_screen_size): we get console size from windows API
> 	when compiling with minGW. 

FYI, the standard way to write this changelog entry would be:

2005-02-10  Denis Pilat  <denis.pilat@st.com>

	* readline/terminal.c (_rl_get_screen_size): Get console size from
	the Windows API when compiling with MinGW.


> Index: terminal.c
> ===================================================================
> --- terminal.c	(revision 264)
> +++ terminal.c	(working copy)
> @@ -70,6 +70,11 @@
>  #include "rlshell.h"
>  #include "xmalloc.h"
>  
> +#if defined (__MINGW32__)
> +#  include <windows.h>
> +#  include <wincon.h>
> +#endif
> +
>  #define CUSTOM_REDISPLAY_FUNC() (rl_redisplay_function != rl_redisplay)
>  #define CUSTOM_INPUT_FUNC() (rl_getc_function != rl_getc)
>  
> @@ -201,6 +206,20 @@
>    struct winsize window_size;
>  #endif /* TIOCGWINSZ */
>  
> +/* For mingw version, we get console size from windows API*/
> +#if defined (__MINGW32__)
> +  HANDLE hConOut = GetStdHandle(STD_OUTPUT_HANDLE);
> +  if ( hConOut != INVALID_HANDLE_VALUE ) 
> +    {
> +      CONSOLE_SCREEN_BUFFER_INFO scr;  
> +      if ( GetConsoleScreenBufferInfo(hConOut, &scr) )
> +        {
> +          _rl_screenwidth = scr.dwSize.X;
> +          _rl_screenheight = scr.srWindow.Bottom - scr.srWindow.Top + 1;
> +        }
> +    }
> +#endif
> +
>  #if defined (TIOCGWINSZ)
>    if (ioctl (tty, TIOCGWINSZ, &window_size) == 0)
>      {

-- 
Daniel Jacobowitz
CodeSourcery

Re: [readline-mingw] backspace key and TUI size

[Daniel Jacobowitz]

On Mon, Feb 20, 2006 at 10:36:47AM -0500, Daniel Jacobowitz wrote:
> FYI, the standard way to write this changelog entry would be:
> 
> 2005-02-10  Denis Pilat  <denis.pilat@st.com>
> 
> 	* readline/terminal.c (_rl_get_screen_size): Get console size from
> 	the Windows API when compiling with MinGW.

Chet confirmed that he has taken a functionally equivalent patch for
the next release of readline.  So, I've committed the reformatted
version of Denis's patch at the bottom of this message.

This gets show width and show height to work on Windows consoles, and
probably helps the TUI.  However, it does not fix paging.  I'm not
planning to fix it, either - I wasted all morning on it.  Here's the
story in case someone else goes down the same rathole someday.

First, GDB disables paging because tgetnum ("li") fails, so it assumes
we're in a non-console environment.  I could hack up win32-termcap.c
to provide "li" if we're on a console, but that's not all...

Next, readline sets cols to 79 instead of 80 because we don't have
both automatic margins ("am") and ignored newlines beyond the automatic
margins ("xn").  This causes utils.c to start printing the prompt
at the last column of the previous line instead of the first column
of the next line.  In fact, we do have automatic margins, but we don't
have "xn".  So even if I hack up win32-termcap.c further to report
"am", this still happens.  I think it'll happen on any terminal without
"xn", given the current logic in gdb and readline, whether or not the
terminal has "am".  But I don't know any other platform using such a
terminal at the moment, so I don't want to rip up the guts of either
readline or gdb to cope better.  It's just not worth it.

So since the only way I could get the paging prompt to work at all was
graphically pretty nasty, and this console has scrollback anyway, I'm
going to leave paging disabled.

-- 
Daniel Jacobowitz
CodeSourcery

Index: src/readline/terminal.c
===================================================================
--- src.orig/readline/terminal.c	2003-12-30 02:25:18.000000000 -0500
+++ src/readline/terminal.c	2006-02-23 10:44:00.000000000 -0500
@@ -70,6 +70,11 @@
 #include "rlshell.h"
 #include "xmalloc.h"
 
+#if defined (__MINGW32__)
+# include <windows.h>
+# include <wincon.h>
+#endif
+
 #define CUSTOM_REDISPLAY_FUNC() (rl_redisplay_function != rl_redisplay)
 #define CUSTOM_INPUT_FUNC() (rl_getc_function != rl_getc)
 
@@ -209,6 +214,20 @@ _rl_get_screen_size (tty, ignore_env)
     }
 #endif /* TIOCGWINSZ */
 
+  /* For MinGW, we get the console size from the Windows API.  */
+#if defined (__MINGW32__)
+  HANDLE hConOut = GetStdHandle (STD_OUTPUT_HANDLE);
+  if (hConOut != INVALID_HANDLE_VALUE)
+    {
+      CONSOLE_SCREEN_BUFFER_INFO scr;
+      if (GetConsoleScreenBufferInfo (hConOut, &scr))
+	{
+	  _rl_screenwidth = scr.dwSize.X;
+	  _rl_screenheight = scr.srWindow.Bottom - scr.srWindow.Top + 1;
+	}
+    }
+#endif
+
 #if defined (__EMX__)
   _emx_get_screensize (&_rl_screenwidth, &_rl_screenheight);
 #endif

Re: [readline-mingw] backspace key and TUI size

[Eli Zaretskii]

> Date: Thu, 23 Feb 2006 13:21:45 -0500
> From: Daniel Jacobowitz <drow@false.org>
> 
> This gets show width and show height to work on Windows consoles, and
> probably helps the TUI.  However, it does not fix paging.  I'm not
> planning to fix it, either - I wasted all morning on it.  Here's the
> story in case someone else goes down the same rathole someday.

FWIW, paging does work in the DJGPP port, so you may wish to take a
look at how it does that.

> First, GDB disables paging because tgetnum ("li") fails, so it assumes
> we're in a non-console environment.

You will see that the DJGPP port ifdef's around that code in utils.c.

> Next, readline sets cols to 79 instead of 80 because we don't have
> both automatic margins ("am") and ignored newlines beyond the automatic
> margins ("xn").  This causes utils.c to start printing the prompt
> at the last column of the previous line instead of the first column
> of the next line.

This part I don't understand: it doesn't happen with DJGPP, AFAICS,
but I don't see any ifdefs in the code to explain the difference.  Can
you post a short test case that exhibits this problem, and show the
code that causes utils.c to print the prompt at the wrong place?

> I think it'll happen on any terminal without "xn", given the current
> logic in gdb and readline, whether or not the terminal has "am".
> But I don't know any other platform using such a terminal at the
> moment

I think DJGPP is one such other platform.

Re: [readline-mingw] backspace key and TUI size

[Daniel Jacobowitz]

On Fri, Feb 24, 2006 at 10:34:51AM +0200, Eli Zaretskii wrote:
> > First, GDB disables paging because tgetnum ("li") fails, so it assumes
> > we're in a non-console environment.
> 
> You will see that the DJGPP port ifdef's around that code in utils.c.
> 
> > Next, readline sets cols to 79 instead of 80 because we don't have
> > both automatic margins ("am") and ignored newlines beyond the automatic
> > margins ("xn").  This causes utils.c to start printing the prompt
> > at the last column of the previous line instead of the first column
> > of the next line.
> 
> This part I don't understand: it doesn't happen with DJGPP, AFAICS,
> but I don't see any ifdefs in the code to explain the difference.  Can
> you post a short test case that exhibits this problem, and show the
> code that causes utils.c to print the prompt at the wrong place?

This happens because you bypass rl_get_screen_size.  Therefore you
use the "real" screen size, rather than readline's adjusted view of it.
Therefore you print the prompt when you've filled column 80, instead
of column 79.  The relevant bit of code in readline is:

  if (_rl_term_autowrap == 0)
    _rl_screenwidth--;

and:

  _rl_term_autowrap = tgetflag ("am") && tgetflag ("xn");

Readline decreases the screen width by one every time you cal
rl_set_screen_size, and supplies the decremented value back for
rl_get_screen_size; GDB relies on (A) having automatic margins,
and (B) having the full screen size.

I guess there are some ways I could get around this...

-- 
Daniel Jacobowitz
CodeSourcery

Re: [readline-mingw] backspace key and TUI size

[Eli Zaretskii]

> Date: Fri, 24 Feb 2006 09:01:45 -0500
> From: Daniel Jacobowitz <drow@false.org>
> Cc: Denis PILAT <denis.pilat@st.com>, Andrew STUBBS <andrew.stubbs@st.com>,
> 	gdb-patches@sources.redhat.com, bash-maintainers@gnu.org
> 
> This happens because you bypass rl_get_screen_size.  Therefore you
> use the "real" screen size, rather than readline's adjusted view of it.

Right, thanks.  So all you need to do is prevent readline from
decrementing _rl_screenwidth in the Windows case.

Re: [readline-mingw] backspace key and TUI size

[Daniel Jacobowitz]

On Fri, Feb 24, 2006 at 08:44:49PM +0200, Eli Zaretskii wrote:
> > Date: Fri, 24 Feb 2006 09:01:45 -0500
> > From: Daniel Jacobowitz <drow@false.org>
> > Cc: Denis PILAT <denis.pilat@st.com>, Andrew STUBBS <andrew.stubbs@st.com>,
> > 	gdb-patches@sources.redhat.com, bash-maintainers@gnu.org
> > 
> > This happens because you bypass rl_get_screen_size.  Therefore you
> > use the "real" screen size, rather than readline's adjusted view of it.
> 
> Right, thanks.  So all you need to do is prevent readline from
> decrementing _rl_screenwidth in the Windows case.

No, I'm pretty sure readline would misbehave if I did that - it checks
for am && xn capabilities, and my experiments show that the Windows
console does not have xn.  I'm not sure what in readline would go
wrong, though.

-- 
Daniel Jacobowitz
CodeSourcery

Re: [readline-mingw] backspace key and TUI size

[Eli Zaretskii]

> Date: Fri, 24 Feb 2006 13:48:52 -0500
> From: Daniel Jacobowitz <drow@false.org>
> Cc: gdb-patches@sources.redhat.com, bash-maintainers@gnu.org
> 
> > > This happens because you bypass rl_get_screen_size.  Therefore you
> > > use the "real" screen size, rather than readline's adjusted view of it.
> > 
> > Right, thanks.  So all you need to do is prevent readline from
> > decrementing _rl_screenwidth in the Windows case.
> 
> No, I'm pretty sure readline would misbehave if I did that - it checks
> for am && xn capabilities, and my experiments show that the Windows
> console does not have xn.

If readline does misbehave (I'm not sure it will), then you could do
in utils.c:init_page_info something similar to what DJGPP does (except
that it should be possible to reuse the results of readline's
initialization instead of calling the console primitives again).

Re: [readline-mingw] backspace key and TUI size

[Eli Zaretskii]

> Date: Wed, 8 Feb 2006 11:14:52 -0500
> From: Daniel Jacobowitz <drow@false.org>
> Cc: gdb-patches@sources.redhat.com, bash-maintainers@gnu.org
> 
> On Wed, Feb 08, 2006 at 05:03:33PM +0100, Denis PILAT wrote:
> > Our host is i686-pc-mingw32, gdb is compiled under cygwin with 
> > "-mno-cygwin" flag to avoid cygwin dependencies.
> > We also use pdcurses library since ncurse is not available on windows.
> > I think that's why you don't see the problem. With this patch, I'm not 
> > using the tputs() function anymore on Windows, as it was not used for MSDOS.
> 
> Ah - we (CodeSourcery) don't use a curses library at all; just the
> standard Windows and MinGW DLLs.  See win32-termcap.c.

But win32-termcap.c is not a general solution for readline, it's
private to GDB, right?  SHouldn't this problem be solved in readline
rather than in GDB (and in every other application that uses
readline)?

Re: [readline-mingw] backspace key and TUI size

[Daniel Jacobowitz]

On Wed, Feb 08, 2006 at 07:56:17PM +0200, Eli Zaretskii wrote:
> > Date: Wed, 8 Feb 2006 11:14:52 -0500
> > From: Daniel Jacobowitz <drow@false.org>
> > Cc: gdb-patches@sources.redhat.com, bash-maintainers@gnu.org
> > 
> > On Wed, Feb 08, 2006 at 05:03:33PM +0100, Denis PILAT wrote:
> > > Our host is i686-pc-mingw32, gdb is compiled under cygwin with 
> > > "-mno-cygwin" flag to avoid cygwin dependencies.
> > > We also use pdcurses library since ncurse is not available on windows.
> > > I think that's why you don't see the problem. With this patch, I'm not 
> > > using the tputs() function anymore on Windows, as it was not used for MSDOS.
> > 
> > Ah - we (CodeSourcery) don't use a curses library at all; just the
> > standard Windows and MinGW DLLs.  See win32-termcap.c.
> 
> But win32-termcap.c is not a general solution for readline, it's
> private to GDB, right?  SHouldn't this problem be solved in readline
> rather than in GDB (and in every other application that uses
> readline)?

I don't remember the reasoning, but this was discussed before, and
the current solution was preferred.  Either you give readline a termcap
library, or you supply stubs.  It sounds like there is a termcap
library we could make use of on Windows, but there isn't one
standard with the OS, and the one there is seems to have some
problems; I'm all for leaving alone one of the bits of this whole
mess which works OK...

-- 
Daniel Jacobowitz
CodeSourcery

Re: [readline-mingw] backspace key and TUI size

[Eli Zaretskii]

> Date: Wed, 08 Feb 2006 09:02:57 +0100
> From: Denis PILAT <denis.pilat@st.com>
> Cc: bash-maintainers@gnu.org
> 
> - In the gdb.exe version, the backspace key do not behave correctly.

Like Daniel, I don't see this problem on my system.

> +/* For mingw version, we get console size from windows API*/
> +#if defined (__MINGW32__)
> +  HANDLE hConOut = GetStdHandle(STD_OUTPUT_HANDLE);
> +  CONSOLE_SCREEN_BUFFER_INFO scr;  
> +  GetConsoleScreenBufferInfo(hConOut, &scr);
> +  _rl_screenwidth = scr.dwSize.X;
> +  _rl_screenheight = scr.srWindow.Bottom - scr.srWindow.Top + 1;
> +#endif

Shouldn't you make sure the handle returned by GetStdHandle is not
INVALID_HANDLE_VALUE, before you use it in a system call?

> -#ifndef __MSDOS__
> -  if (_rl_term_backspace)
> +#if !defined (__MSDOS__) && !defined (__MINGW32__)
> +  if (_rl_term_backspace) 

The idea is okay with me, but there;s a stray blank after the last
line above.  Please remove it.

And thanks for working on this.