Re: [PATCH] --warnings=no-XXX [RFC] --warnings categories

[Akim Demaille]

Le 22 nov. 08 à 05:13, Di-an JAN a écrit :

> On Fri, 21 Nov 2008, Akim Demaille wrote:
> > Di-an, could you please push in master the patch of yours that  
> > warns about the missing ; and adds them?  We can have it in 2.5 to  
> > drop the ; feature completely starting with 2.6 or so.
>
> Do you want to keep the warning forever or just for 2.5?

I was thinking about 2.5 only.

> Should it be just for default language, default skeleton, and --yacc?
> It should work for C, C++, and Java.

No, it should be for C only, same cases as for Joel's patch.

> Thinking about it some more, I agree with Joel[2] that this patch[1]  
> is
> too complicated to be worthwhile.

I agree too (and discussed about it privately with Joel, I'm not just  
overruling him here), I very much dislike all that quantity of code  
for just a single semicolon.  Yet backward compatibility does matter.   
I have no idea how many people are suffering now from this change in  
Bison's behavior.  They have better things to do in there lives that  
having to understand that someone left out a semicolon in a Bison  
grammar.

That's why I believe that we need to send the grammar maintainers a  
strong signal: that what your patch does.  Since it's only a warning,  
it should protect users from broken tarballs, and hopefully, it should  
give a chance to maintainers to fix their packages.

We could leave the change forever too.  I just don't like the idea of  
propagating a historical mistake.  This means Bison will have this  
forever, and I don't feel confortable with this.  Yet this might be  
the most pragmatic approach.

<sigh>

I don't know.  I have no strong opinion now that Joel's patch fixes  
most cases while preserving the cases that need not be broken by  
this.  I just like the idea of being able to get rid of this in the  
future, thanks to a temporary warning.

> I mean, trying to handle preprocessing
> directives... I didn't even thought about it until the testsuite  
> failed.
> Bison really shouldn't do the job of the compiler, whenever  
> possible[3].
> I think we can do without the warning as there will be a compiler
> error whenever the warning issued correctly.

I'm not that sure.  Beware that users who install tarballs need not be  
package maintainers.  They just want "./configure && make && make  
check && sudo make install" to DTRT.  And I hate wasting their time,  
they have better things to live than tracking missing semicolons.

> The only reason to bother is for bug-compatibility, and the reported
> problem[4] is GCC's sample frontend treelang, which is not supposed to
> be actually used, and is removed from GCC 4.4 (in development)[5].

Interesting.  I wasn't aware of this, thanks for the pointers.