On 14/07/2015 10:19, Ulrike Fischer wrote:
> Am Tue, 14 Jul 2015 09:42:13 +0100 schrieb Joseph Wright:
> 
>> I was thinking of
>>
>>     \keys_define:nn {test}
>>       {
>>         testa .code:n = {\tl_show:n {#1}},
>>         testa .value_required:,
>>         testb .code:n = {\tl_show:n {#1}},
>>         testb .value_required:,
>>         testb .default:n = {}
>>     }
>>
>>     \keys_set:nn {test} {testa,testb}
>>
>> where one might expect to get an error only in the first case. That's
>> not currently the situation!
> 
> Hm. My first reaction was that it works as expected. If I explictly
> require that the *user* gives a value, a *package* default doesn't
> make sense and shouldn't replace the user action.

That's entirely logical and as you say what currently happens, it's just
it's not documented (or I think tested). I'll address that so we have a
clear position :-)

> It is also consistent with the other way round:
> 
> \documentclass{article}
> \usepackage{expl3}
> \begin{document}
> \ExplSyntaxOn
>   \keys_define:nn {test}
>       {
>         testb .code:n = {#1},
>         testb .value_forbidden:,
>         testb .default:n = {xxx}
>     }
> 
>    \keys_set:nn {test} {testb}
> 
> \end{document} 

Again, probably should document this.

Time-allowing I will do an update to l3keys today to address all of the
points raised.
--
Joseph Wright