Re: analyzing warning messages from log file

[David Kastrup]

Graham Percival <address@hidden> writes:

> On Sun, Jan 16, 2011 at 07:04:02AM +0100, Werner LEMBERG wrote:
>> 
>> while walking through the log file I see many warnings for lilypond
>> input files.  Some of them are part of the regression test,
>
> IMO those should be moved into a separate directory, intended for
> such warnings.
>
>> What about adding
>> 
>>   #(ly:warning (_ "Expect one warning"))
>> 
>> to files where a warning is expected?
>
> The last time this was suggested (about 5 years ago?), somebody
> (maybe Han-Wen?) pointed out that #(ly:warning ...) would be
> processed during the parsing stage, whereas most warnings occurred
> during the engraving stage.  In a file with multiple expected
> warnings, they wouldn't match up nicely.  The idea was dropped.
>
> I personally would like some functions to print output at the
> relevant stage:
>   #(ly:warning_parser (_ "blah"))
>   #(ly:warning_translator (_ "blah"))
>   #(ly:warning_event_class (_ "blah"))
>   #(ly:warning_engraver (_ "blah"))
>   #(ly:warning_performer (_ "blah"))
>   #(ly:warning_whatever_else_we_have (_ "blah"))
>
> (or not necessarily "warning"; just a "print_during" -- it would
> actually be useful for one of my research projects!)
>
> I'd feel better about decorating regtests if we could truly target
> the message to the warning.  (I still think that most regtests
> should produce no warnings, and any exception should be in a
> different directory)

I'd rather we just use
#(regtest:expext_warning (_ "Some particular warning"))

before the line being the eventual cause of some expected warning, and
ly:warning get hijacked, matching the expected warnings to the given
warnings, and complaining at the end of the run in case any expected
warnings did not turn up.

-- 
David Kastrup