Re: Unsatisfied %expectations

[Joel E. Denny]

On Fri, 18 Nov 2005, Akim Demaille wrote:

> I was just horrified to discover that unmet %expect are not errors.  I
> had an ambiguity that sneaked in and remain hidden in the lengthy
> compilation.  I propose the following patch, which is admittedly not
> backward compatible, but otherwise the feature is quite meaningless...
>
> An alternative would be introducing -Werror, but (i) that's more work
> :( and (ii) it still should be an error, not a warning!

I can understand your argument for making this an error in the case of 
relatively stable grammars.

However, consider how this will impact the development of new, 
experimental grammars. Every time the grammar is tweaked, the number of 
conflicts may change slightly.  The developer must then run bison, observe 
the number of conflicts reported in the error, update the bison spec with 
that number, and then try to compile again.  This could really slow down 
development especially when hacking a grammar for the sake of debugging. 
I can imagine this being incredibly frustrating for students (or anyone) 
learning bison for the first time.

Joel