Re: ACTION-128: Draft @summary voting text in conjunction with PF

On Jul 8, 2009, at 12:00 AM, Henri Sivonen wrote:

> On Jul 7, 2009, at 03:10, Maciej Stachowiak wrote:
>
>> The actual material difference between different diagnostic classes  
>> is not friendliness in the validator, but rather whether authors  
>> can still use the feature if they have a requirement (self-imposed  
>> or otherwise) to produce fully conforming content.
>>
>> Thus, I think a suggested non-error diagnostic would be better than  
>> a requirement for a down-played error, in that it would give the  
>> advanced experts who choose to disregard the warning the  
>> opportunity to have their content be conforming.
>
>
> I don't really like spec-prescribed warnings, either, because I fear  
> that if we start doing normative warnings, people start wanting to  
> make failures to adhere to their code style aesthetics as normative  
> warnings when the discussion won't affect valid/invalid.

I think that would be a bad way to use warnings, and I hope that won't  
happen. If we do introduce them, they should be reserved for things  
where there is a strong reason to discourage a practice or alert  
authors to pitfalls, but there are sufficient exceptions that making  
such content blanket invalid would be excessive. Basically the rough  
equivalent of a SHOULD NOT. Matters of aesthetics would not qualify. I  
would hope there is broad agreement on this.

I also think validators should be free to have ways to turn off some  
of the otherwise mandatory or spec-suggested warnings, or alternately  
to treat warnings as fatal at the request of the user - the rough  
equivalent of -Wno-foo and -Werror in gcc. I think this will further  
reduce the temptation to use warnings for inappropriate purposes.

Regards,
Maciej