Re: [PATCH 3/7] comctl32/monthcal: Use a single calendar for hittesting tests, make point calculation at runtime

[Paul Vriens]

On 09/29/2009 12:36 PM, Nikolay Sivov wrote:
> Changelog:
>     - be sure we have a single view calendar for hittesting

Hi Nikolay,

This one introduces a test failure on several platforms:

monthcal.c:1002: Test failed: Expected 10000, got 10002

I fixed it on my NT4 box by checking the point more to the left (see 
attached).

I somehow doubt it's the correct fix though. Aren't these tests with 
more-or-less fixed points error-prone (aka, locale/font/dpi dependent)?

[PATCH 3/7] comctl32/monthcal: Use a single calendar for hittesting tests, make point calculation at runtime

[Nikolay Sivov]

Paul Vriens wrote:
> On 09/29/2009 12:36 PM, Nikolay Sivov wrote:
> > Changelog:
> >     - be sure we have a single view calendar for hittesting
>
> Hi Nikolay,
>
> This one introduces a test failure on several platforms:
>
> monthcal.c:1002: Test failed: Expected 10000, got 10002
>
> I fixed it on my NT4 box by checking the point more to the left (see 
> attached).
>
> I somehow doubt it's the correct fix though. Aren't these tests with 
> more-or-less fixed points error-prone (aka, locale/font/dpi dependent)?

Hi, Paul.

I certainly noticed that. A problem for me is that it doesn't fail on 
mine XP. Could you please send me (offlist) a crossbuilt binary with 
this patch - I'm not able to built it myself at the moment.

About fragility of these test - it was a reason why I use dynamic point 
calculation here, it was fixed before. So now it should be less dpi/font 
dependent I hope. It isn't affected by locale too much I think.

[PATCH 3/7] comctl32/monthcal: Use a single calendar for hittesting tests, make point calculation at runtime

[Paul Vriens]

On 09/30/2009 12:10 PM, Nikolay Sivov wrote:
> Paul Vriens wrote:
> > On 09/29/2009 12:36 PM, Nikolay Sivov wrote:
> > > Changelog:
> > > - be sure we have a single view calendar for hittesting
> >
> > Hi Nikolay,
> >
> > This one introduces a test failure on several platforms:
> >
> > monthcal.c:1002: Test failed: Expected 10000, got 10002
> >
> > I fixed it on my NT4 box by checking the point more to the left (see
> > attached).
> >
> > I somehow doubt it's the correct fix though. Aren't these tests with
> > more-or-less fixed points error-prone (aka, locale/font/dpi dependent)?
>
> Hi, Paul.
>
> I certainly noticed that. A problem for me is that it doesn't fail on
> mine XP. Could you please send me (offlist) a crossbuilt binary with
> this patch - I'm not able to built it myself at the moment.
>
> About fragility of these test - it was a reason why I use dynamic point
> calculation here, it was fixed before. So now it should be less dpi/font
> dependent I hope. It isn't affected by locale too much I think.

Doing a little bit of testing (doing a hittest on the full range) shows 
subtle differences between for example my NT4 and W2K3 box. The W2K3 
calendar shows "April, 2009" where NT4 shows "April 2009".

Looking at 
http://test.winehq.org/data/3fe20bdc7ea59b3e5711bed26d86c433109b21e7/xp_af-x...
even seems to indicate that with this locale the year is shown before 
the month?

I'm wondering what a good test would be (that covers all cases) or 
whether most of these tests should be dropped?

A normal order should be for "April 2009":

- MCHT_TITLE
- MCHT_TITLEBTNPREV
- MCHT_TITLE
- MCHT_TITLEMONTH
- MCHT_TITLE
- MCHT_TITLEYEAR
- MCHT_TITLE
- MCHT_TITLEBTNNEXT
- MCHT_TITLE

(That box mentioned would have MCHT_TITLEMONTH and MCHT_TITLEYEAR the 
other way around).

Maybe testing the order (both variations) would be fine enough?

Any thoughts?

[PATCH 3/7] comctl32/monthcal: Use a single calendar for hittesting tests, make point calculation at runtime

[Nikolay Sivov]

Paul Vriens wrote:
> On 09/30/2009 12:10 PM, Nikolay Sivov wrote:
> > Paul Vriens wrote:
> > > On 09/29/2009 12:36 PM, Nikolay Sivov wrote:
> > > > Changelog:
> > > > - be sure we have a single view calendar for hittesting
> > >
> > > Hi Nikolay,
> > >
> > > This one introduces a test failure on several platforms:
> > >
> > > monthcal.c:1002: Test failed: Expected 10000, got 10002
> > >
> > > I fixed it on my NT4 box by checking the point more to the left (see
> > > attached).
> > >
> > > I somehow doubt it's the correct fix though. Aren't these tests with
> > > more-or-less fixed points error-prone (aka, locale/font/dpi dependent)?
> >
> > Hi, Paul.
> >
> > I certainly noticed that. A problem for me is that it doesn't fail on
> > mine XP. Could you please send me (offlist) a crossbuilt binary with
> > this patch - I'm not able to built it myself at the moment.
> >
> > About fragility of these test - it was a reason why I use dynamic point
> > calculation here, it was fixed before. So now it should be less dpi/font
> > dependent I hope. It isn't affected by locale too much I think.
>
> Doing a little bit of testing (doing a hittest on the full range) 
> shows subtle differences between for example my NT4 and W2K3 box. The 
> W2K3 calendar shows "April, 2009" where NT4 shows "April 2009".
>
> Looking at 
> http://test.winehq.org/data/3fe20bdc7ea59b3e5711bed26d86c433109b21e7/xp_af-x...
>
> even seems to indicate that with this locale the year is shown before 
> the month?
>
> I'm wondering what a good test would be (that covers all cases) or 
> whether most of these tests should be dropped?
>
> A normal order should be for "April 2009":
> - MCHT_TITLE
> - MCHT_TITLEBTNPREV
> - MCHT_TITLE
> - MCHT_TITLEMONTH
> - MCHT_TITLE
> - MCHT_TITLEYEAR
> - MCHT_TITLE
> - MCHT_TITLEBTNNEXT
> - MCHT_TITLE
>
> (That box mentioned would have MCHT_TITLEMONTH and MCHT_TITLEYEAR the 
> other way around).
>
> Maybe testing the order (both variations) would be fine enough?
>
> Any thoughts?

If some locales do this rectangles permutation it's another bug in Wine 
code, we currently don't depend on locale at this point.

Maybe you're right and we should add alternate hittesting codes for a 
test results structure and make it differ on month/year cases.
When Wine got support for this it will fail without such changes on it 
as well. So patch it please if you have time now.

P.S. how could I change a locale on my Debian system for particular Wine 
run? Could you help me with that?

[PATCH 3/7] comctl32/monthcal: Use a single calendar for hittesting tests, make point calculation at runtime

[Paul Vriens]

On 10/01/2009 09:58 AM, Nikolay Sivov wrote:
> Paul Vriens wrote:
> > On 09/30/2009 12:10 PM, Nikolay Sivov wrote:
> > > Paul Vriens wrote:
> > > > On 09/29/2009 12:36 PM, Nikolay Sivov wrote:
> > > > > Changelog:
> > > > > - be sure we have a single view calendar for hittesting
> > > >
> > > > Hi Nikolay,
> > > >
> > > > This one introduces a test failure on several platforms:
> > > >
> > > > monthcal.c:1002: Test failed: Expected 10000, got 10002
> > > >
> > > > I fixed it on my NT4 box by checking the point more to the left (see
> > > > attached).
> > > >
> > > > I somehow doubt it's the correct fix though. Aren't these tests with
> > > > more-or-less fixed points error-prone (aka, locale/font/dpi dependent)?
> > >
> > > Hi, Paul.
> > >
> > > I certainly noticed that. A problem for me is that it doesn't fail on
> > > mine XP. Could you please send me (offlist) a crossbuilt binary with
> > > this patch - I'm not able to built it myself at the moment.
> > >
> > > About fragility of these test - it was a reason why I use dynamic point
> > > calculation here, it was fixed before. So now it should be less dpi/font
> > > dependent I hope. It isn't affected by locale too much I think.
> >
> > Doing a little bit of testing (doing a hittest on the full range)
> > shows subtle differences between for example my NT4 and W2K3 box. The
> > W2K3 calendar shows "April, 2009" where NT4 shows "April 2009".
> >
> > Looking at
> > http://test.winehq.org/data/3fe20bdc7ea59b3e5711bed26d86c433109b21e7/xp_af-x...
> >
> > even seems to indicate that with this locale the year is shown before
> > the month?
> >
> > I'm wondering what a good test would be (that covers all cases) or
> > whether most of these tests should be dropped?
> >
> > A normal order should be for "April 2009":
> > - MCHT_TITLE
> > - MCHT_TITLEBTNPREV
> > - MCHT_TITLE
> > - MCHT_TITLEMONTH
> > - MCHT_TITLE
> > - MCHT_TITLEYEAR
> > - MCHT_TITLE
> > - MCHT_TITLEBTNNEXT
> > - MCHT_TITLE
> >
> > (That box mentioned would have MCHT_TITLEMONTH and MCHT_TITLEYEAR the
> > other way around).
> >
> > Maybe testing the order (both variations) would be fine enough?
> >
> > Any thoughts?
>
> If some locales do this rectangles permutation it's another bug in Wine
> code, we currently don't depend on locale at this point.
>
> Maybe you're right and we should add alternate hittesting codes for a
> test results structure and make it differ on month/year cases.
> When Wine got support for this it will fail without such changes on it
> as well. So patch it please if you have time now.

I'll see what I can come up with, although probably later today.
> P.S. how could I change a locale on my Debian system for particular Wine
> run? Could you help me with that?

What I usually do is something like

LC_ALL=nl_NL.UTF-8 wine ....
or
LC_ALL=nl_NL.UTF-8 make monthcal.ok

More lang references can be found at http://wiki.winehq.org/TestingLanguages

[PATCH 3/7] comctl32/monthcal: Use a single calendar for hittesting tests, make point calculation at runtime

[Nikolay Sivov]

Paul Vriens wrote:
> On 10/01/2009 09:58 AM, Nikolay Sivov wrote:
> > P.S. how could I change a locale on my Debian system for particular Wine
> > run? Could you help me with that?
>
> What I usually do is something like
>
> LC_ALL=nl_NL.UTF-8 wine ....
> or
> LC_ALL=nl_NL.UTF-8 make monthcal.ok
>
> More lang references can be found at 
> http://wiki.winehq.org/TestingLanguages

Strange it doesn't work for me. Both LC_ALL and LANG don't change 
anything. Maybe I've got a wrong configuration on my system, how could I 
reset it to default or reinitialize in some way?