[m-rev.] for review: fix silent failure of making .int file

[Julien Fischer]

On Fri, 17 Dec 2021, Zoltan Somogyi wrote:

> 2021-12-17 12:26 GMT+11:00 "Julien Fischer" <jfischer at opturion.com>:
>>>   Instead of generating error_specs that are conditional on
>>>   --warn-unused-imports being OFF, generate them condition on
>>
>> s/condition/condtional/
>>
>> That's fine otherwise.
>
> Thanks for both reviews.
>
> I think this diff brings up a followup issue, which is: what should we do
> with warnings that the compiler generates while making interface files?
> At the moment, we print them, and --halt-at-warn turns them into errors
> that cause the creation of the .int file to fail, but I see at least two alternatives
> that would give programmers more freedom to choose *when* they fix
> the warnings.
>
> 1 When making interface files, we could simply filter out error_specs
>   with severity_warning the same way that this diff filters out error_specs
>   that don't print anything.
>
> 2. When making interface files, we could decide whether to treat warnings
>   as errors not based on the value of --halt-at-warn, but based on the value
>   of a new option, maybe called --interface-halt-at-warn, which could be
>   set independently. With this option, warnings could still be printed
>   at interface generation time, but would not prevent programmers
>   prioritizing fixing issues in other modules first.
>
> I myself would prefer 2 over both 1 and the status quo. Any other opinions?

I prefer option 2.

Julien.