Fixing the conformance tests

[Dan Kegel]

Paul V. wrote:
> As we are slowly progressing towards a more green
> test.winehq.org there are still some conformance tests
> that fail either on all platforms or with a specific configuration (like IE7).

Here's some data.
On my new box (CPU: Core 2 Duo e8400,
graphics: nVidia Corporation GeForce 8500 GT (rev a1)
window manager: metacity 2.25.0 built from source)
distro: 32 bit Hardy), out of 212 full runs of 'make -k test',
here's a count of failures by test name:

6 d3d9 device
6 comdlg32 printdlg
8 wininet http
14 kernel32 thread
19 comctl32 tooltips
24 ddraw visual
32 riched20 editor
46 user32 win
137 user32 input
215 user32 msg
216 dsound ds3d
216 winmm wave
216 urlmon protocol
216 dsound ds3d8

In my perfect world, those tests would all pass, every time, even if
you ran them 100 times.
That would make patchwatcher a lot more useful, since
then it wouldn't have to blacklist these tests.

Here's a little script to tickle just the flaky tests I found:

#!/bin/sh
try()
{
    cd dlls/$1/tests
    rm $2.ok
    make $2.ok
    cd ../../..
}
set -e
set -x
while true
do
    try comctl32 tooltips
    try comdlg32 printdlg
    try d3d9 device
    try ddraw visual
    try dsound ds3d8
    try kernel32 thread
    try riched20 editor
    try user32 input
    try user32 win
    try wininet http
done

This aborts for me in just five seconds.   Can *anyone* run it for
an hour without failure?  If so, how can I duplicate your setup?
- Dan