Marshal TKIND_DISPATCH interfaces just like TKIND_INTERFACE.

actually support them. Small other cleanups.

waveOutOpen problem so that we can remove the todo_wine.
Win9x does not support WAVE_FORMAT_DIRECT.
Don't check the format if the waveOutOpen command failed.
Use trace, not winetest_trace!

directly, on Mingw we use it to build the .a library.

to switch it off.

much as possible in the buffer even if it is too small. This is
undocumented and does not match the Win9x behavior.
Skip Unicode tests on Win9x.
Fix usage of memset/sizeof/COUNTOF.
Fix signed/unsigned warnings (in MSVC) by using lstrlenA instead of
strlen.

ERROR_INSUFFICIENT_BUFFER if the buffer is too small.

Test Get{System,Windows}Directory{A,W}(NULL, 0). This is a more
standard way to get len_with_null too.
Adapt the error code checks to take into account variations between
Win9x and NT.

Adapt the DeleteFileA error code checks to take into account variations
between Win9x and NT.
Test DeleteFile(NULL).
Add tests for DeleteFileW.
On NT, calling _lclose on an already closed handle will cause memory
corruption and thus sometimes crash -> removed the relevant test.
Skip the Unicode tests when on Win9x.

Made the -i option support removing ignored symbols too.
Removed the -noimport flag in spec files.

MultiByteToWideChar but not lstrcmpW! So I implemented our own so
that we can have a meaningful test.

generated file.

- Use new time functions to reimplement FileTimeToLocalTime and
  LocalTimeToFileTime.

Simplify $infile existence test.

path.
Merge winetest_set_trace_location with winetest_set_ok_location.

GetShortPathNameA behaves differently on Win9x and NT: on NT it
succeeds even if not all path components exist, as long as they are
already in the 8.3 format.
Wine apparently implements the NT behavior thus many todo_wine went
away.
Fixed some error code checks to take into account all possible return
values.
GetTempFileNameA appears to only use the lower 16bits of the id on
Win95 (and never returns more than 16 bits on other platforms).
GetLongPathNameA is missing on some Windows versions.
Modified the GetTempPath tests to make sure they return the expected
value. Removed the redundant tests (e.g. if buf[0]==0 then buf!="foo",
no need to test both).
The 'len_with_null - 1' case is not testable as the Windows behavior
varies too much between versions.
We cannot check whether Windows touches the buffer either as this
heavily depends on the Windows version and specific circumstances of
the call.
Finally NT4 sometimes exaggerates the required buffer size.

that STRICT works correctly; moved some definitions back to windef.h
where they belong, and removed a couple of definitions that don't
exist on Windows.