Right way to cope with user error in make test?

[Alexandre Julliard]

Jeremy White <jwhite at winehq.org> writes:

>> Alexandre has said in the past that test failures for an incorrectly
>> or incompletely built wine tree should result in test failures and I
>> agree with this.
>
> Yah; I think to some extent we need to wait for Alexandre to express
> an opinion on how, if at all, he'd like to address this.
>
> We seem to have come up with about 4 approaches:
>   1.  Skip the test.  Rob thinks Alexandre will reject this
>   2.  Make WINE_NOTICE_WITH be default error; i.e. require
>       an explicit --without in order to skip a package you lack
>   3.  Create some sort of config record; a config.id if you will.
>       This could then be read by dotests and/or winetests
>       to not transmit the results.
>       (As a side note, this might be handy place to put
>       a git HEAD which might allow my winetest patches to go in,
>       thereby eliminating the need for an out of tree dotests.  But no bias here <grin>).
>   4.  Have dotests scan the existing config.log file.
>
> Alexandre, do you have a preference?

I don't like any of those. What I would expect is if a library is needed
for a given dll to work, that dll would fail to load when the library is
missing, and winetest should handle that the same way it does for
missing Win9x dlls. If it's only a small part of the dll that is broken,
then it's perfectly fine for the test to fail.

-- 
Alexandre Julliard
julliard at winehq.org