Re: "notes" mechanism (fwd)

[Joel E. Denny]

This message was rejected by bison-patches the last time I tried to send 
it, so here it is again.

---------- Forwarded message ----------
Date: Tue, 28 Jul 2009 08:19:16 -0400 (EDT)
From: Joel E. Denny <address@hidden>
To: Alex Rozenman <address@hidden>
Cc: Akim Demaille <address@hidden>, address@hidden
Subject: Re: "notes" mechanism

On Tue, 28 Jul 2009, Alex Rozenman wrote:

> > Can you explain your motivation a little more?  Is it that you want to
> > reuse the same *_at invocations to list variants in both the warning and
> > complaint cases?  You could add to note_at a bool argument that specifies
> > whether to invoke warn_at or complain_at.
> >
> My motivation was the following:
> 1) Keep overall *number* of errors/warnings issued by bison reasonable. When
> bison issues an "ambiguity" error with 10 sub-messages, overall number of
> errors (if one does "grep" on output, or, alternatively bison itself will
> count them and print a summary message) should be one.

That begs another question: why do we need to count the errors or 
warnings?  Maybe there's a good reason, but I don't know it yet.

> 2) Not to see the "warning" word in each sub-message.

I don't see the appeal of that change either, but maybe I've just grown 
accustomed to the current practice.  I just checked gcc 4.2.4, and it also 
prints "warning:" on every submessage.  (Actually, gcc also prints 
"error:" for errors in the same manner.)

Unless there are other practical reasons for these changes, I think we 
need more opinions.  Akim says he'll be back in a couple of weeks.  He 
usually has a better sense of whether it's ok to change long-standing 
practices in Bison.