Wine Test Bot 1.7.14 analysis

[Jeremy White]

I wrote a script to analyze winetestbot results on all of the testbot 
vms except the win8 vms (they are just too broken to try to analyze 
right now).

I'm trying to get a handle on the nature of the bot failures; this 
current script looks for consistent failures (partly because a 
consistent failure that goes green is a win, and I badly want to track 
wins).

My results are here:
   http://www.winehq.org/~jwhite/ecd24b5a874e.html

The short summary is that we have 38 current failures.  But we appear to 
have fixed 3 failures (at least on the testbot vms).  We also have 14 
that seem consistent (i.e. that should be more tractable).

I intend to use this to track progress and show status.  This result is 
arguably not that interesting; I'm mostly posting it now so I can be 
publicly shamed if I fail to follow through.

Cheers,

Jeremy

[Nikolay Sivov]

On 3/10/2014 06:13, Jeremy White wrote:
> msxml3:saxreader.html 
> <http://test.winehq.org/data/tests/msxml3:saxreader.html> 
> xp_newtb-wxppro(0 
> <http://test.winehq.org/data/ecd24b5a874ead368c8f6e9d6981bb0e02472f9d/xp_newtb-wxppro/msxml3:saxreader.html>,1 
> <http://test.winehq.org/data/630e8d92578b347d6e94db097c05572bb416bb2e/xp_newtb-wxppro/msxml3:saxreader.html>,2 
> <http://test.winehq.org/data/49f3b4282d29890fe3c213d096571af4de2c45cd/xp_newtb-wxppro/msxml3:saxreader.html>,3 
> <http://test.winehq.org/data/9c5c3a81ceab5362513de6eb81cee921dcc52c14/xp_newtb-wxppro/msxml3:saxreader.html>,4 
> <http://test.winehq.org/data/049f08f4cda090189ae57d4ba58906d891ac3d4c/xp_newtb-wxppro/msxml3:saxreader.html>,5 
> <http://test.winehq.org/data/376953e00a97cc6ff5e18e8a8e0cd7fb70b15629/xp_newtb-wxppro/msxml3:saxreader.html>,6 
> <http://test.winehq.org/data/0eb626587b2f75e9904ba827eec1cd8a7f5789a2/xp_newtb-wxppro/msxml3:saxreader.html>,7 
> <http://test.winehq.org/data/fcae01672f2d480597a40850ff0386268b24791d/xp_newtb-wxppro/msxml3:saxreader.html>,8 
> <http://test.winehq.org/data/ccd8daf0f8564949e0811decf6a110b95be1a57a/xp_newtb-wxppro/msxml3:saxreader.html>,9 
> <http://test.winehq.org/data/37e0a1a5d4977a5f017709109dd6cf7a948b78e8/xp_newt...>) 
> 2000_newtb-w2000pro(1 
> <http://test.winehq.org/data/630e8d92578b347d6e94db097c05572bb416bb2e/2000_newtb-w2000pro/msxml3:saxreader.html>,2 
> <http://test.winehq.org/data/49f3b4282d29890fe3c213d096571af4de2c45cd/2000_newtb-w2000pro/msxml3:saxreader.html>,4 
> <http://test.winehq.org/data/049f08f4cda090189ae57d4ba58906d891ac3d4c/2000_newtb-w2000pro/msxml3:saxreader.html>,5 
> <http://test.winehq.org/data/376953e00a97cc6ff5e18e8a8e0cd7fb70b15629/2000_newtb-w2000pro/msxml3:saxreader.html>,7 
> <http://test.winehq.org/data/fcae01672f2d480597a40850ff0386268b24791d/2000_newtb-w2000pro/msxml3:saxreader.html>,9 
> <http://test.winehq.org/data/37e0a1a5d4977a5f017709109dd6cf7a948b78e8/2000_ne...>)
> Here's a can of green paint for this:

http://www.winehq.org/pipermail/wine-patches/2014-March/131010.html

[Nikolay Sivov]

On 3/10/2014 06:13, Jeremy White wrote:

Regarding Mac failures, it looks like this problem:

http://test.winehq.org/data/ecd24b5a874ead368c8f6e9d6981bb0e02472f9d/mac_fg-...

is about missing/old libxslt

[Francois Gouget]

On Mon, 10 Mar 2014, Nikolay Sivov wrote:
> On 3/10/2014 06:13, Jeremy White wrote:
>
> Regarding Mac failures, it looks like this problem:
>
> http://test.winehq.org/data/ecd24b5a874ead368c8f6e9d6981bb0e02472f9d/mac_fg-...
>
> is about missing/old libxslt

What's interesting is that the macdrv tests do not run into this issue.

The difference is that for the macdrv tests I set 
DYLD_FALLBACK_LIBRARY_PATH="/opt/local/lib" which causes them to use 
what I believe to be the MacPorts libxslt.dylib library.

For the x11drv tests I set DYLD_FALLBACK_LIBRARY_PATH="/opt/X11/lib" 
because the libX11.dylib library does not (or did not) play well with 
XQuartz. As a result the X11 tests use /usr/lib/libxslt.dylib which is 
the library shipped with Snow Leopard.

I'll retry running the x11drv tests with /opt/local/lib.


However given that MacPorts is not part of Mac OS X that raises the 
question of what sort of system tweaks make sense in order to run the 
Wine conformance tests without error.

-- 
Francois Gouget <fgouget(a)free.fr>              http://fgouget.free.fr/
      Broadcast message : fin du monde dans cinq minutes, repentez vous !

[Nikolay Sivov]

On 3/10/2014 19:10, Francois Gouget wrote:
> On Mon, 10 Mar 2014, Nikolay Sivov wrote:
> > On 3/10/2014 06:13, Jeremy White wrote:
> >
> > Regarding Mac failures, it looks like this problem:
> >
> > http://test.winehq.org/data/ecd24b5a874ead368c8f6e9d6981bb0e02472f9d/mac_fg-...
> >
> > is about missing/old libxslt
> > What's interesting is that the macdrv tests do not run into this issue.
>
> The difference is that for the macdrv tests I set
> DYLD_FALLBACK_LIBRARY_PATH="/opt/local/lib" which causes them to use
> what I believe to be the MacPorts libxslt.dylib library.
>
> For the x11drv tests I set DYLD_FALLBACK_LIBRARY_PATH="/opt/X11/lib"
> because the libX11.dylib library does not (or did not) play well with
> XQuartz. As a result the X11 tests use /usr/lib/libxslt.dylib which is
> the library shipped with Snow Leopard.
>
> I'll retry running the x11drv tests with /opt/local/lib.
>
> However given that MacPorts is not part of Mac OS X that raises the
> question of what sort of system tweaks make sense in order to run the
> Wine conformance tests without error.
> So it failed while running with system shipped version? I think we 
> should use a version that we expect users to use, and I don't know which 
> one is that. If system version is old enough to cause troubles then we 
> should be using something more up-to-date from MacPorts. System lib is 
> unlikely to be updated with system update, right? Especially for system 
> that are not supported anymore (not sure if Snow Leopard still gets 
> updates).

If it's decided to use system lib no matter what then test should be 
improved to skip in such cases.

[Stefan Dösinger]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Am 2014-03-10 03:13, schrieb Jeremy White:
> * ddraw:ddraw4.html 2000_newtb-w2000pro(7,9)
> * ddraw:ddraw7.html 2000_newtb-w2000pro(7,9) win7_newtb-w7u(7,8,9)
> Some of the lines in them are fairly easy to fix, I'll send a patch.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
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=+s7i
-----END PGP SIGNATURE-----

[Stefan Dösinger]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Am 2014-03-10 11:04, schrieb Stefan Dösinger:
> Am 2014-03-10 03:13, schrieb Jeremy White:
> > * ddraw:ddraw4.html 2000_newtb-w2000pro(7,9) * ddraw:ddraw7.html
> > 2000_newtb-w2000pro(7,9) win7_newtb-w7u(7,8,9)
> > Some of the lines in them are fairly easy to fix, I'll send a
> > patch.
> > Actually the machines I looked at weren't from the testbot at all, but
> > fgouget's VMware machine. Still at least one failure on them should be
> > fixable.

Francois, can you test the attached patch on your
fg-win7u64-* VM? It doesn't matter which one, except for the 0sp one,
which seems to time out. This patch should fix this kind of error in
ddraw4 and ddraw7:

ddraw4.c:4190: Test failed: Got unexpected hr 0x88760104 for format
D3DFMT_YUY2, resource type videomemory overlay, size 1x1, expected 0.

There are still other failures left in those tests, so they still
won't pass. The error count should be considerably lower though.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
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=VrUP
-----END PGP SIGNATURE-----

[Francois Gouget]

On Mon, 10 Mar 2014, Stefan Dösinger wrote:
[...]
> Francois, can you test the attached patch on your
> fg-win7u64-* VM? It doesn't matter which one, except for the 0sp one,
> which seems to time out. This patch should fix this kind of error in
> ddraw4 and ddraw7:
>
> ddraw4.c:4190: Test failed: Got unexpected hr 0x88760104 for format
> D3DFMT_YUY2, resource type videomemory overlay, size 1x1, expected 0.

Yep. I tested it in fg-win7u64-1spie9 and the patch fixes these 
failures. It leaves 16 and 18 test failures for ddraw4 and ddraw7 
respectively, which is a big improvement over the 143 and 145 the VM had 
before.


-- 
Francois Gouget <fgouget(a)free.fr>              http://fgouget.free.fr/
May your Tongue stick to the Roof of your Mouth with the Force of a Thousand Caramels.

[Henri Verbeet]

On 10 March 2014 03:13, Jeremy White <jwhite(a)codeweavers.com> wrote:
> I wrote a script to analyze winetestbot results on all of the testbot vms
> except the win8 vms (they are just too broken to try to analyze right now).
>
> I'm trying to get a handle on the nature of the bot failures; this current
> script looks for consistent failures (partly because a consistent failure
> that goes green is a win, and I badly want to track wins).
>
> My results are here:
>   http://www.winehq.org/~jwhite/ecd24b5a874e.html

Note that e.g. the win2000 testbot doesn't have results for all runs.
It looks like this causes the script to classify some failures that
should be "fixed" as intermittent failures. That might in turn cause
someone to draw wrong conclusions about e.g. the ddraw tests, if they
didn't pay enough attention to wine-patches.

[Jeremy White]

On 03/10/2014 06:08 AM, Henri Verbeet wrote:
> On 10 March 2014 03:13, Jeremy White <jwhite(a)codeweavers.com> wrote:
> > I wrote a script to analyze winetestbot results on all of the testbot vms
> > except the win8 vms (they are just too broken to try to analyze right now).
> >
> > I'm trying to get a handle on the nature of the bot failures; this current
> > script looks for consistent failures (partly because a consistent failure
> > that goes green is a win, and I badly want to track wins).
> >
> > My results are here:
> >    http://www.winehq.org/~jwhite/ecd24b5a874e.html
>
> Note that e.g. the win2000 testbot doesn't have results for all runs.
> It looks like this causes the script to classify some failures that
> should be "fixed" as intermittent failures. That might in turn cause
> someone to draw wrong conclusions about e.g. the ddraw tests, if they
> didn't pay enough attention to wine-patches.

Yes; not just win2k, but the win7u bot is unreliable, and one of the 
other win7 bots and one of the vista bots have a few drop outs as well.

But my code, in theory, skips holes in the data, so long as the data 
stays in line.

In other words, a pattern like this:
   SS-FFFF-FF
where S is success, F is failure, and - is missing data, is considered
'fixed'.  A pattern like this:
   F-F--F-F-F
is considered 'consistently failing'.  All other patterns are considered
intermittent.

Note that it's only against the newtb vms; so you'll see the claim that 
d3d9:stateblock is fixed, but there is one non newtb machine where it 
still fails.

Cheers,

Jeremy

[Henri Verbeet]

On 10 March 2014 13:26, Jeremy White <jwhite(a)codeweavers.com> wrote:
> But my code, in theory, skips holes in the data, so long as the data stays
> in line.
>
> In other words, a pattern like this:
>   SS-FFFF-FF
> where S is success, F is failure, and - is missing data, is considered
> 'fixed'.  A pattern like this:
>   F-F--F-F-F
> is considered 'consistently failing'.  All other patterns are considered
> intermittent.

The data for ddraw7 on Windows 2000 for example is "-SS-SS-F-F".

[Jeremy White]

On 03/10/2014 07:55 AM, Henri Verbeet wrote:
> On 10 March 2014 13:26, Jeremy White <jwhite(a)codeweavers.com> wrote:
> > But my code, in theory, skips holes in the data, so long as the data stays
> > in line.
> >
> > In other words, a pattern like this:
> >    SS-FFFF-FF
> > where S is success, F is failure, and - is missing data, is considered
> > 'fixed'.  A pattern like this:
> >    F-F--F-F-F
> > is considered 'consistently failing'.  All other patterns are considered
> > intermittent.
>
> The data for ddraw7 on Windows 2000 for example is "-SS-SS-F-F".

Yeah, I explained it incorrectly (and the code is rough, and quite 
possibly wrong).  A pattern requires exact edges to be considered fixed; 
so 'SS-FFFF-FF' would be considered indeterminate.  'SSF-FFF-FF' would 
be considered fixed.

I'll see about tweaking that  (I changed it to prevent '-FFFFFFFFF' from 
being considered 'fixed' <grin>.  But I can fix that a different way).

Cheers,

Jeremy

[Jeremy White]

> I'll see about tweaking that  (I changed it to prevent '-FFFFFFFFF' from
> being considered 'fixed' <grin>.  But I can fix that a different way).

I've updated it.

   http://www.winehq.org/~jwhite/ecd24b5a874e.html

We now have 10 purported fixes!  Woohoo!

I've also added annotations, which was the main feature I was planning, 
so I could work the list more intelligently.  I've hopefully added in 
the relevant notes from Francois.

My main focus is on the test bot vms, as those match the automated patch 
screen.  (Doesn't mean we shouldn't fix some of the other interesting 
tests; I'm just trying to limit the scope).

Cheers,

Jeremy

[Jeremy White]

I've updated it slightly to show the S/-/F indicators, and run it 
against the latest Wine:
   http://www.winehq.org/~jwhite/770213e16c69.html

The good news is that Nikolay has, apparently, painted msxml3:saxreader 
a nice shade of green.

A few other tests are now being considered intermittent 
(advapi:eventlog, urlmon:url, and comdlg32:filedlg).  That's not a 
material change; mostly just highlights flaws in the previous analysis.

So, 1 down, 32 to go...

Cheers,

Jeremy

[Francois Gouget]

* ntdll:exception (All QEmu VMs)
  These are caused by a known QEmu bug. That bug got fixed^H^H^Hreplaced 
  by another bug in 1.7.0. See:
  https://bugs.launchpad.net/qemu/+bug/1119686


Also a test may have multiple independent failures. So it's important to 
look at the individual test failures and gather clues from the tests 
they fail on.

For instance:
user32:msg.html win7_newtb-w7u(0,1,2,7,8,9) xp_newtb-wxppro(4)
- There are 49 failures on the newtb-w7u VM and under a dozen on the 
  other TestBot VMs. The reason is that the newtb-w7u VM is set up with 
  a Japanse locale which causes extra failures. Note that one can get 
  more information about a given VM by clicking on the 'info' link in 
  the test results. Here:
  http://test.winehq.org/data/ecd24b5a874ead368c8f6e9d6981bb0e02472f9d/win7_ne...

  These extra user32:msg test failures in Japanese locales have been 
  documented: http://bugs.winehq.org/show_bug.cgi?id=35611

- The same test also exhibits specific errors in the Hebrew locale. We 
  don't have Hebrew or other LTR VMs in the TestBot but my Windows 7 VM 
  has one such test configuration: fg-win7u64-he.
  http://bugs.winehq.org/show_bug.cgi?id=35610

- The remaining Windows 7 VMs have a group of 3 'region' failures, 
  and another unrelated group of 3 'message 31f' failures which appears 
  to be somewhat random (or at least does not affect all VMs). 

- My Windows XP VMs have a totally unrelated set of 3 'minimum timeout' 
  failures, which sometimes appear on Windows 2003 and Windows 8.
  http://bugs.winehq.org/show_bug.cgi?id=34915

So as one can see things can be quite complex under the hood.


Here are some test failures I looked into and tried to diagnose a bit 
(all VMs, not just the WineTestBot ones):

* Bug 35573 - gdi32:fonts test_stock_fonts() fails on Windows 7 in the 
              Japanese and Hebrew locales
  http://bugs.winehq.org/show_bug.cgi?id=35573

* Bug 35760 - gdi32:font test_fullname2() fails on Windows 7 in the 
              French locale
  http://bugs.winehq.org/show_bug.cgi?id=35760

* Bug 33720 - user32:menu
  This one is intermittent.
  http://bugs.winehq.org/show_bug.cgi?id=33720

* Bug 33718 - comctl32:propsheet Add button test failure
  http://bugs.winehq.org/show_bug.cgi?id=33718

* Bug 33719 - comctl32:propsheet custom window proc test failure
  http://bugs.winehq.org/show_bug.cgi?id=33719


And for some Windows 8 issues:

* Bug 35575 - gdi32:font Windows 8.1 failures
  http://bugs.winehq.org/show_bug.cgi?id=35575

* Bug 34830 - rpcrt4:cstub fails and crashes on Windows 8
  http://bugs.winehq.org/show_bug.cgi?id=34830

* Bug 34829 - wintrust:softpub crashes on Windows 8
  http://bugs.winehq.org/show_bug.cgi?id=34829


-- 
Francois Gouget <fgouget(a)codeweavers.com>