GNU bug report logs - #10397
[PATCH] Under Remote Desktop, NUMCOLORS is unreliable; workaround

Previous Next

To add a comment to this bug, you must first unarchive it, by sending
a message to control AT debbugs.gnu.org, with unarchive 10397 in the body.
You can then email your comments to 10397 AT debbugs.gnu.org in the normal way.

Toggle the display of automated, internal messages from the tracker.

Message #5 received at submit <at> debbugs.gnu.org (full text, mbox):

From: Daniel Colascione <dancol <at> dancol.org>
To: bug-gnu-emacs <at> gnu.org
Subject: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable; workaround
Date: Thu, 29 Dec 2011 06:05:53 -0800

Under remote desktop, Windows returns the wrong number of colors from
GetDeviceCaps (hdc, NUMCOLORS).  I hit this bug myself, and MSDN
comments seem to indicate that others hit it as well.  The workaround
seems harmless: on non-palettized displays, calculating the number of
display colors based on display bitness should produce good results.
---
 src/w32fns.c |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletions(-)

diff --git a/src/w32fns.c b/src/w32fns.c
index 822e353..4b94f16 100644
--- a/src/w32fns.c
+++ b/src/w32fns.c
@@ -4510,7 +4510,10 @@ If omitted or nil, that stands for the selected frame's display.  */)
   if (dpyinfo->has_palette)
     cap = GetDeviceCaps (hdc, SIZEPALETTE);
   else
-    cap = GetDeviceCaps (hdc, NUMCOLORS);
+    // GetDeviceCaps (NUMCOLORS) is buggy under remote desktop and sometimes
+    // returns the number of system reserved colors (20) instead of
+    // the actual number of available colors.
+    cap = -1;
 
   /* We force 24+ bit depths to 24-bit, both to prevent an overflow
      and because probably is more meaningful on Windows anyway */
-- 
1.7.5.1

Message #8 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Juanma Barranquero <lekktu <at> gmail.com>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Thu, 29 Dec 2011 17:13:25 +0100

On Thu, Dec 29, 2011 at 15:05, Daniel Colascione <dancol <at> dancol.org> wrote:

> The workaround
> seems harmless: on non-palettized displays, calculating the number of
> display colors based on display bitness should produce good results.

Even so, why fix what is not broken? Why can't you just do

=== modified file 'src/w32fns.c'
--- src/w32fns.c	2011-12-04 08:02:42 +0000
+++ src/w32fns.c	2011-12-29 16:10:33 +0000
@@ -4511,5 +4511,12 @@
     cap = GetDeviceCaps (hdc, SIZEPALETTE);
   else
-    cap = GetDeviceCaps (hdc, NUMCOLORS);
+    {
+      cap = GetDeviceCaps (hdc, NUMCOLORS);
+      /* GetDeviceCaps (NUMCOLORS) is buggy under remote desktop and
+         sometimes returns the number of system reserved colors (20)
+         instead of the actual number of available colors.  */
+      if (cap == 20)
+        cap = -1;
+    }

   /* We force 24+ bit depths to 24-bit, both to prevent an overflow

> +    // GetDeviceCaps (NUMCOLORS) is buggy under remote desktop and sometimes
> +    // returns the number of system reserved colors (20) instead of
> +    // the actual number of available colors.

Please, don't use "C++ / modern C" style comments; use /* */ instead.

    Juanma

Message #11 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Daniel Colascione <dancol <at> dancol.org>
To: Juanma Barranquero <lekktu <at> gmail.com>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable; 
	workaround
Date: Thu, 29 Dec 2011 08:23:50 -0800

[Message part 1 (text/plain, inline)]

On 12/29/11 8:13 AM, Juanma Barranquero wrote:
> On Thu, Dec 29, 2011 at 15:05, Daniel Colascione <dancol <at> dancol.org> wrote:
> 
>> The workaround
>> seems harmless: on non-palettized displays, calculating the number of
>> display colors based on display bitness should produce good results.
> 
> Even so, why fix what is not broken? Why can't you just do
> 
> === modified file 'src/w32fns.c'
> --- src/w32fns.c	2011-12-04 08:02:42 +0000
> +++ src/w32fns.c	2011-12-29 16:10:33 +0000
> @@ -4511,5 +4511,12 @@
>      cap = GetDeviceCaps (hdc, SIZEPALETTE);
>    else
> -    cap = GetDeviceCaps (hdc, NUMCOLORS);
> +    {
> +      cap = GetDeviceCaps (hdc, NUMCOLORS);
> +      /* GetDeviceCaps (NUMCOLORS) is buggy under remote desktop and
> +         sometimes returns the number of system reserved colors (20)
> +         instead of the actual number of available colors.  */
> +      if (cap == 20)
> +        cap = -1;
> +    }

Why? What's the point of adding the extra complexity?
Setting cap to -1 leads to this line

    1 << min (dpyinfo->n_planes * dpyinfo->n_cbits, 24);

which produces a reasonable result for direct color displays.
Why keep using NUMCOLORS, which we know to be broken?

>    /* We force 24+ bit depths to 24-bit, both to prevent an overflow
> 
>> +    // GetDeviceCaps (NUMCOLORS) is buggy under remote desktop and sometimes
>> +    // returns the number of system reserved colors (20) instead of
>> +    // the actual number of available colors.
> 
> Please, don't use "C++ / modern C" style comments; use /* */ instead.
> 

That block slipped through.

[signature.asc (application/pgp-signature, attachment)]

Message #14 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Juanma Barranquero <lekktu <at> gmail.com>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Thu, 29 Dec 2011 17:27:48 +0100

On Thu, Dec 29, 2011 at 17:23, Daniel Colascione <dancol <at> dancol.org> wrote:

> Why? What's the point of adding the extra complexity?
> Setting cap to -1 leads to this line
>
>    1 << min (dpyinfo->n_planes * dpyinfo->n_cbits, 24);
>
> which produces a reasonable result for direct color displays.
> Why keep using NUMCOLORS, which we know to be broken?

No, you said that NUMCOLORS is known to be broken in a very specific
case. In the general, non-RemoteDektop case, it works. Can you
guarantee that for every non-palettized display, it will produce the
same number? Because otherwise you're changing the current behavior
for people who's not affected by the RemoteDeskopt-related bug.

    Juanma

Message #17 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Daniel Colascione <dancol <at> dancol.org>
To: Juanma Barranquero <lekktu <at> gmail.com>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable; 
	workaround
Date: Thu, 29 Dec 2011 08:42:37 -0800

[Message part 1 (text/plain, inline)]

On 12/29/11 8:27 AM, Juanma Barranquero wrote:
> On Thu, Dec 29, 2011 at 17:23, Daniel Colascione <dancol <at> dancol.org> wrote:
> 
>> Why? What's the point of adding the extra complexity?
>> Setting cap to -1 leads to this line
>>
>>    1 << min (dpyinfo->n_planes * dpyinfo->n_cbits, 24);
>>
>> which produces a reasonable result for direct color displays.
>> Why keep using NUMCOLORS, which we know to be broken?
> 
> No, you said that NUMCOLORS is known to be broken in a very specific
> case. In the general, non-RemoteDektop case, it works. 

I'm not convinced there aren't other bugs lurking in the code backing
NUMCOLORS; after all, it's doing the same simple calculation we are, and
it's somehow doing it wrong.

http://msdn.microsoft.com/en-us/library/dd144877%28v=VS.85%29.aspx#3
suggests that NUMCOLORS is generally flaky.

> Can you
> guarantee that for every non-palettized display, it will produce the
> same number? Because otherwise you're changing the current behavior
> for people who's not affected by the RemoteDeskopt-related bug.

No, I can't guarantee that my original change always produces the same
results: it might fix other bugs.

[signature.asc (application/pgp-signature, attachment)]

Message #20 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Juanma Barranquero <lekktu <at> gmail.com>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Thu, 29 Dec 2011 17:45:44 +0100

On Thu, Dec 29, 2011 at 17:42, Daniel Colascione <dancol <at> dancol.org> wrote:

> I'm not convinced there aren't other bugs lurking in the code backing
> NUMCOLORS; after all, it's doing the same simple calculation we are, and
> it's somehow doing it wrong.

You have also no proof. I don't see why should we try to fix bugs we
don't even know are real. Better to fiix the one you know it *is*
real, IMO.

> No, I can't guarantee that my original change always produces the same
> results: it might fix

or introduce

> other bugs.

    Juanma

Message #23 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Daniel Colascione <dancol <at> dancol.org>
To: Juanma Barranquero <lekktu <at> gmail.com>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable; 
	workaround
Date: Thu, 29 Dec 2011 14:59:11 -0800

[Message part 1 (text/plain, inline)]

On 12/29/11 8:45 AM, Juanma Barranquero wrote:
> On Thu, Dec 29, 2011 at 17:42, Daniel Colascione <dancol <at> dancol.org> wrote:
> 
>> I'm not convinced there aren't other bugs lurking in the code backing
>> NUMCOLORS; after all, it's doing the same simple calculation we are, and
>> it's somehow doing it wrong.
> 
> You have also no proof. I don't see why should we try to fix bugs we
> don't even know are real. Better to fiix the one you know it *is*
> real, IMO.

What about this: we'll distrust any NUMCOLORS response less than 256.
You'll never use direct color with a bit depth that small, so any answer
in that range must be bogus.  This approach would address my concerns
about wrong values other than exactly 20 being returned.

[signature.asc (application/pgp-signature, attachment)]

Message #26 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Juanma Barranquero <lekktu <at> gmail.com>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Fri, 30 Dec 2011 00:10:56 +0100

On Thu, Dec 29, 2011 at 23:59, Daniel Colascione <dancol <at> dancol.org> wrote:

> What about this: we'll distrust any NUMCOLORS response less than 256.
> You'll never use direct color with a bit depth that small, so any answer
> in that range must be bogus.  This approach would address my concerns
> about wrong values other than exactly 20 being returned.

OK, assuming you're talking post-24.1.

    Juanma

Message #29 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Daniel Colascione <dancol <at> dancol.org>
To: Juanma Barranquero <lekktu <at> gmail.com>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable; 
	workaround
Date: Thu, 29 Dec 2011 15:13:02 -0800

[Message part 1 (text/plain, inline)]

On 12/29/11 3:10 PM, Juanma Barranquero wrote:
> On Thu, Dec 29, 2011 at 23:59, Daniel Colascione <dancol <at> dancol.org> wrote:
> 
>> What about this: we'll distrust any NUMCOLORS response less than 256.
>> You'll never use direct color with a bit depth that small, so any answer
>> in that range must be bogus.  This approach would address my concerns
>> about wrong values other than exactly 20 being returned.
> 
> OK, assuming you're talking post-24.1.
> 
>     Juanma
> 

Would it be possible to get the more limited (exactly 20) fix into 24.1?

[signature.asc (application/pgp-signature, attachment)]

Message #32 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Juanma Barranquero <lekktu <at> gmail.com>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Fri, 30 Dec 2011 00:18:32 +0100

On Fri, Dec 30, 2011 at 00:13, Daniel Colascione <dancol <at> dancol.org> wrote:

> Would it be possible to get the more limited (exactly 20) fix into 24.1?

The "20-check" is a bug fix against a specific bug. OOH, it's not a
regression, OTOH, its impact is limited to Windows. You'll have to ask
Chong and/or Stefan; it's their decision.

    Juanma

Message #35 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Jason Rumney <jasonr <at> gnu.org>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Fri, 30 Dec 2011 09:04:16 +0800

Daniel Colascione <dancol <at> dancol.org> writes:

> Under remote desktop, Windows returns the wrong number of colors from
> GetDeviceCaps (hdc, NUMCOLORS).  I hit this bug myself, and MSDN
> comments seem to indicate that others hit it as well.  The workaround
> seems harmless: on non-palettized displays, calculating the number of
> display colors based on display bitness should produce good results.

I've always been under the impression that this is deliberate, and
related to the bandwidth that is available, so at least applications
that want to improve performance over low bandwidth links can restrict
their use of colors.

Message #38 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Daniel Colascione <dancol <at> dancol.org>
To: Jason Rumney <jasonr <at> gnu.org>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable; 
	workaround
Date: Thu, 29 Dec 2011 17:10:44 -0800

[Message part 1 (text/plain, inline)]

On 12/29/11 5:04 PM, Jason Rumney wrote:
> Daniel Colascione <dancol <at> dancol.org> writes:
> 
>> Under remote desktop, Windows returns the wrong number of colors from
>> GetDeviceCaps (hdc, NUMCOLORS).  I hit this bug myself, and MSDN
>> comments seem to indicate that others hit it as well.  The workaround
>> seems harmless: on non-palettized displays, calculating the number of
>> display colors based on display bitness should produce good results.
> 
> I've always been under the impression that this is deliberate, and
> related to the bandwidth that is available, so at least applications
> that want to improve performance over low bandwidth links can restrict
> their use of colors.

A remote desktop user can change the depth of the virtual display
presented to applications on the server.  If a user wants to trade
fidelity for bandwidth, he can configure his client to use an 8bpp
visual.  Some users (me) configure their clients for a relatively high
bit depth, but find that the OS lies to applications some of the time
(NUMCOLORS is wrong, but the display bitness is accurate).  I think the
NUMCOLORS behavior is a real bug; if it weren't, the lie would be more
consistently presented.

[signature.asc (application/pgp-signature, attachment)]

Message #41 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Juanma Barranquero <lekktu <at> gmail.com>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Fri, 30 Dec 2011 04:07:48 +0100

On Thu, Dec 29, 2011 at 23:59, Daniel Colascione <dancol <at> dancol.org> wrote:

> What about this: we'll distrust any NUMCOLORS response less than 256.
> You'll never use direct color with a bit depth that small, so any answer
> in that range must be bogus.

Hmm. Shouldn't in fact GetDeviceCaps (hdc, NUMCOLORS) always be <= 256?

According to http://msdn.microsoft.com/en-us/library/dd144877(v=vs.85).aspx

  NUMCOLORS
  Number of entries in the device's color table, if the device has a
color depth of no more than 8 bits per pixel. For devices with greater
color depths, 1 is returned.

(It says "1", but it's a typo for "-1".)

    Juanma

Message #44 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Daniel Colascione <dancol <at> dancol.org>
To: Juanma Barranquero <lekktu <at> gmail.com>
Cc: 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable; 
	workaround
Date: Thu, 29 Dec 2011 19:18:13 -0800

[Message part 1 (text/plain, inline)]

On 12/29/11 7:07 PM, Juanma Barranquero wrote:
> On Thu, Dec 29, 2011 at 23:59, Daniel Colascione <dancol <at> dancol.org> wrote:
> 
>> What about this: we'll distrust any NUMCOLORS response less than 256.
>> You'll never use direct color with a bit depth that small, so any answer
>> in that range must be bogus.
> 
> Hmm. Shouldn't in fact GetDeviceCaps (hdc, NUMCOLORS) always be <= 256?
> 
> According to http://msdn.microsoft.com/en-us/library/dd144877(v=vs.85).aspx
> 
>   NUMCOLORS
>   Number of entries in the device's color table, if the device has a
> color depth of no more than 8 bits per pixel. For devices with greater
> color depths, 1 is returned.
> 
> (It says "1", but it's a typo for "-1".)

Good catch. What about this (untested) code?

  hdc = GetDC (dpyinfo->root_window);
  if (dpyinfo->has_palette)
    cap = GetDeviceCaps (hdc, SIZEPALETTE);
  else if (dpyinfo->n_cbits <= 8)
    /* According to the MSDN, GetDeviceCaps (NUMCOLORS) is valid only
       for devices with at most eight bits per pixel.  It's supposed
       to return -1 for other displays, but because it actually
       returns other, incorrect values under some conditions (e.g.,
       remote desktop), only use it when we know it's valid.  */
    cap = GetDeviceCaps (hdc, NUMCOLORS);
  else
    cap = -1;

[signature.asc (application/pgp-signature, attachment)]

Message #47 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Eli Zaretskii <eliz <at> gnu.org>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: lekktu <at> gmail.com, 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Fri, 30 Dec 2011 10:51:40 +0200

> Date: Thu, 29 Dec 2011 19:18:13 -0800
> From: Daniel Colascione <dancol <at> dancol.org>
> Cc: 10397 <at> debbugs.gnu.org
> 
> > Hmm. Shouldn't in fact GetDeviceCaps (hdc, NUMCOLORS) always be <= 256?
> > 
> > According to http://msdn.microsoft.com/en-us/library/dd144877(v=vs.85).aspx
> > 
> >   NUMCOLORS
> >   Number of entries in the device's color table, if the device has a
> > color depth of no more than 8 bits per pixel. For devices with greater
> > color depths, 1 is returned.
> > 
> > (It says "1", but it's a typo for "-1".)
> 
> Good catch. What about this (untested) code?
> 
>   hdc = GetDC (dpyinfo->root_window);
>   if (dpyinfo->has_palette)
>     cap = GetDeviceCaps (hdc, SIZEPALETTE);
>   else if (dpyinfo->n_cbits <= 8)
>     /* According to the MSDN, GetDeviceCaps (NUMCOLORS) is valid only
>        for devices with at most eight bits per pixel.  It's supposed
>        to return -1 for other displays, but because it actually
>        returns other, incorrect values under some conditions (e.g.,
>        remote desktop), only use it when we know it's valid.  */
>     cap = GetDeviceCaps (hdc, NUMCOLORS);
>   else
>     cap = -1;

There's a comment near the end of the GetDeviceCaps documentation
saying this:

  NUMCOLORS doesn't always return one with more than 256 colors

  The documentation for NUMCOLORS is wrong, devices that support more
  than 256 colors often return -1 and some virtual machines can return
  the number of Windows reserved colors (i.e. 20), even in high color
  mode.  Combine the NUMCOLORS return value with BITSPIXEL and PLANES
  to reliably detect indexed color.

So how about using `1 << (n_planes * n_cbits)' (which we compute when
NUMCOLORS call returns -1) unconditionally?  IOW, why do we need to
call GetDeviceCaps(..., NUMCOLORS) here in the first place?

Message #50 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Eli Zaretskii <eliz <at> gnu.org>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: lekktu <at> gmail.com, 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Fri, 30 Dec 2011 11:00:20 +0200

> Date: Thu, 29 Dec 2011 19:18:13 -0800
> From: Daniel Colascione <dancol <at> dancol.org>
> Cc: 10397 <at> debbugs.gnu.org
> 
> >   NUMCOLORS
> >   Number of entries in the device's color table, if the device has a
> > color depth of no more than 8 bits per pixel. For devices with greater
> > color depths, 1 is returned.
> > 
> > (It says "1", but it's a typo for "-1".)
> 
> Good catch. What about this (untested) code?
> 
>   hdc = GetDC (dpyinfo->root_window);
>   if (dpyinfo->has_palette)
>     cap = GetDeviceCaps (hdc, SIZEPALETTE);
>   else if (dpyinfo->n_cbits <= 8)
>     /* According to the MSDN, GetDeviceCaps (NUMCOLORS) is valid only
>        for devices with at most eight bits per pixel.  It's supposed
>        to return -1 for other displays, but because it actually
>        returns other, incorrect values under some conditions (e.g.,
>        remote desktop), only use it when we know it's valid.  */
>     cap = GetDeviceCaps (hdc, NUMCOLORS);
>   else
>     cap = -1;

Looks good to me.  I think this could go into 24.1, unless Jason
(or someone else) objects.

As I wrote elsewhere, past 24.1, we could explore the possibility of
not calling GetDeviceCaps here at all (assuming that using the number
of planes and bits per plane does the job even when the latter is 8 or
less).

Message #53 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Juanma Barranquero <lekktu <at> gmail.com>
To: Eli Zaretskii <eliz <at> gnu.org>
Cc: Daniel Colascione <dancol <at> dancol.org>, 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Fri, 30 Dec 2011 13:24:52 +0100

On Fri, Dec 30, 2011 at 10:00, Eli Zaretskii <eliz <at> gnu.org> wrote:

> Looks good to me.  I think this could go into 24.1, unless Jason
> (or someone else) objects.
>
> As I wrote elsewhere, past 24.1, we could explore the possibility of
> not calling GetDeviceCaps here at all (assuming that using the number
> of planes and bits per plane does the job even when the latter is 8 or
> less).

Agreed.

    Juanma

Message #56 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Eli Zaretskii <eliz <at> gnu.org>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: Juanma Barranquero <lekktu <at> gmail.com>, 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Tue, 07 Aug 2012 20:13:14 +0300

> Date: Tue, 07 Aug 2012 01:19:27 -0700
> From: Daniel Colascione <dancol <at> dancol.org>
> 
> Under remote desktop, Windows returns the wrong number of colors from
> GetDeviceCaps (hdc, NUMCOLORS).  I hit this bug myself, and MSDN
> comments seem to indicate that others hit it as well.  The workaround
> seems harmless: on non-palettized displays, calculating the number of
> display colors based on display bitness should produce good results.
> ---
>  src/w32fns.c |    9 ++++++++-
>  1 files changed, 8 insertions(+), 1 deletions(-)
> 
> diff --git a/src/w32fns.c b/src/w32fns.c
> index b82d4bc..7fc5cf5 100644
> --- a/src/w32fns.c
> +++ b/src/w32fns.c
> @@ -4513,8 +4513,15 @@ If omitted or nil, that stands for the selected frame's display.  */)
>    hdc = GetDC (dpyinfo->root_window);
>    if (dpyinfo->has_palette)
>      cap = GetDeviceCaps (hdc, SIZEPALETTE);
> -  else
> +  else if (dpyinfo->n_cbits <= 8)
> +    /* According to the MSDN, GetDeviceCaps (NUMCOLORS) is valid only
> +       for devices with at most eight bits per pixel.  It's supposed
> +       to return -1 for other displays, but because it actually
> +       returns other, incorrect values under some conditions (e.g.,
> +       remote desktop), only use it when we know it's valid.  */
>      cap = GetDeviceCaps (hdc, NUMCOLORS);
> +  else
> +    cap = -1;
>  
>    /* We force 24+ bit depths to 24-bit, both to prevent an overflow
>       and because probably is more meaningful on Windows anyway */

Last time we spoke, the conclusion (at least mine ;-) was that it
might be better not to call GetDeviceCaps at all, and instead reuse
the code below this, which uses the number of planes and bits per
plane.  If you agree with that reasoning, could you please see if that
change solves your problem?

In any case, let's separate between this patch and the rest of "w32
GUI with Cygwin" patches, as they are really unrelated.  In
particular, as soon as we agree on this one, you cam go ahead and
commit it regardless of the rest.

Thanks.

Message #59 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Daniel Colascione <dancol <at> dancol.org>
To: Eli Zaretskii <eliz <at> gnu.org>
Cc: Juanma Barranquero <lekktu <at> gmail.com>, 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable; 
	workaround
Date: Tue, 07 Aug 2012 10:33:23 -0700

[Message part 1 (text/plain, inline)]

On 8/7/12 10:13 AM, Eli Zaretskii wrote:
>> Date: Tue, 07 Aug 2012 01:19:27 -0700
>> From: Daniel Colascione <dancol <at> dancol.org>
>>
>> Under remote desktop, Windows returns the wrong number of colors from
>> GetDeviceCaps (hdc, NUMCOLORS).  I hit this bug myself, and MSDN
>> comments seem to indicate that others hit it as well.  The workaround
>> seems harmless: on non-palettized displays, calculating the number of
>> display colors based on display bitness should produce good results.
>> ---
>>  src/w32fns.c |    9 ++++++++-
>>  1 files changed, 8 insertions(+), 1 deletions(-)
>>
>> diff --git a/src/w32fns.c b/src/w32fns.c
>> index b82d4bc..7fc5cf5 100644
>> --- a/src/w32fns.c
>> +++ b/src/w32fns.c
>> @@ -4513,8 +4513,15 @@ If omitted or nil, that stands for the selected frame's display.  */)
>>    hdc = GetDC (dpyinfo->root_window);
>>    if (dpyinfo->has_palette)
>>      cap = GetDeviceCaps (hdc, SIZEPALETTE);
>> -  else
>> +  else if (dpyinfo->n_cbits <= 8)
>> +    /* According to the MSDN, GetDeviceCaps (NUMCOLORS) is valid only
>> +       for devices with at most eight bits per pixel.  It's supposed
>> +       to return -1 for other displays, but because it actually
>> +       returns other, incorrect values under some conditions (e.g.,
>> +       remote desktop), only use it when we know it's valid.  */
>>      cap = GetDeviceCaps (hdc, NUMCOLORS);
>> +  else
>> +    cap = -1;
>>  
>>    /* We force 24+ bit depths to 24-bit, both to prevent an overflow
>>       and because probably is more meaningful on Windows anyway */
> 
> Last time we spoke, the conclusion (at least mine ;-) was that it
> might be better not to call GetDeviceCaps at all, and instead reuse
> the code below this, which uses the number of planes and bits per
> plane.  If you agree with that reasoning, could you please see if that
> change solves your problem?

Sorry about that --- I'm bringing a lot of this up form very cold
mental storage. It's been a little while since I've had a chance to do
any Emacs hacking.

I'm perfectly happy using the planes-and-bits code instead of calling
GetDeviceCaps. I'll remove this patch from the cygw32 changeset and
check the (now, much simpler) fix for the colors issue into the trunk,
if that's all right.

[signature.asc (application/pgp-signature, attachment)]

Message #62 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Eli Zaretskii <eliz <at> gnu.org>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: lekktu <at> gmail.com, 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
	workaround
Date: Tue, 07 Aug 2012 21:07:02 +0300

> Date: Tue, 07 Aug 2012 10:33:23 -0700
> From: Daniel Colascione <dancol <at> dancol.org>
> CC: Juanma Barranquero <lekktu <at> gmail.com>, 10397 <at> debbugs.gnu.org
> 
> > Last time we spoke, the conclusion (at least mine ;-) was that it
> > might be better not to call GetDeviceCaps at all, and instead reuse
> > the code below this, which uses the number of planes and bits per
> > plane.  If you agree with that reasoning, could you please see if that
> > change solves your problem?
> 
> Sorry about that --- I'm bringing a lot of this up form very cold
> mental storage. It's been a little while since I've had a chance to do
> any Emacs hacking.

No need to apologize.  It's not that my memory is better than yours, I
just looked up the bug report and re-read the entire thread...

> I'm perfectly happy using the planes-and-bits code instead of calling
> GetDeviceCaps. I'll remove this patch from the cygw32 changeset and
> check the (now, much simpler) fix for the colors issue into the trunk,
> if that's all right.

Please go ahead, but I hope you have a way to check the change you
will commit with remote desktop, because I don't.

Message #65 received at 10397 <at> debbugs.gnu.org (full text, mbox):

From: Lars Ingebrigtsen <larsi <at> gnus.org>
To: Daniel Colascione <dancol <at> dancol.org>
Cc: Juanma Barranquero <lekktu <at> gmail.com>, Eli Zaretskii <eliz <at> gnu.org>,
 10397 <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
 workaround
Date: Thu, 25 Feb 2016 16:55:15 +1030

Daniel Colascione <dancol <at> dancol.org> writes:

>>>    hdc = GetDC (dpyinfo->root_window);
>>>    if (dpyinfo->has_palette)
>>>      cap = GetDeviceCaps (hdc, SIZEPALETTE);
>>> -  else
>>> +  else if (dpyinfo->n_cbits <= 8)
>>> +    /* According to the MSDN, GetDeviceCaps (NUMCOLORS) is valid only
>>> +       for devices with at most eight bits per pixel.  It's supposed
>>> +       to return -1 for other displays, but because it actually
>>> +       returns other, incorrect values under some conditions (e.g.,
>>> +       remote desktop), only use it when we know it's valid.  */
>>>      cap = GetDeviceCaps (hdc, NUMCOLORS);
>>> +  else
>>> +    cap = -1;

Looking at w32fns.c, I can't really find anything that resembles the
surrounding code in this patch.  Instead there seems to be newer code
that does...  stuff...  to palettes.

So is this all outdated now, and this stuff works fine?
-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no

Message #70 received at 10397-done <at> debbugs.gnu.org (full text, mbox):

From: Glenn Morris <rgm <at> gnu.org>
To: 10397-done <at> debbugs.gnu.org
Subject: Re: bug#10397: [PATCH] Under Remote Desktop, NUMCOLORS is unreliable;
 workaround
Date: Mon, 12 Dec 2016 19:04:42 -0500

Version: 24.3

I see this was fixed in emacs-24.2-3079-g821812e.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.