Re: lalr1.cc: use %printer in the syntax error messages

[Akim Demaille]

Hi Joel,

Unwinding my todo stack :)

Le 27 août 09 à 09:11, Joel E. Denny a écrit :

> > (Oh, and by the way, the error messages presented here are  
> > completely wrong.
> > I have no idea why they are, and I hoped that Joel's changes would  
> > solve them,
> > but they do not: there are many many other tokens that can follow  
> > the initial
> > float, arithmetical operators for instance.)
>
> I've checked it out, and the problem is fixed by canonical LR.

I did expect this, but I didn't understand why I had the problem,  
since the state lists in the *.output file correctly listed the  
expected tokens (I'm referring to the lists of tokens to shift, those  
for reduction, and a default reduction).  Of course the problem was  
the default reduction, which I had overlooked.  The state in which the  
error is actually diagnosed is reached a couple of default reductions  
later, and in that state, indeed, the only valid token is EOF.

I played with canonical-lr, but it is way too expensive for me, both  
in terms of size, and duration of compilation (maybe in a release  
mode, but...).  But toying with lr.default-reductions (still in lalr)  
gives very good results.

I have 382 LR(0) states, and 12941 is canonical LR(1).

Here are the size of the parser files.

address@hidden ~/src/gnet/kernel $ ls -l _build/*/ 
ugrammar 
.cc                                                                  
12:40:05
-rw-r--r--  1 akim  akim   321287 Sep  2 12:10 _build/accepting/ 
ugrammar.cc
-rw-r--r--  1 akim  akim   170357 Sep  2 12:13 _build/all/ugrammar.cc
-rw-r--r--  1 akim  akim  8913314 Sep  2 11:52 _build/canonical-lr/ 
ugrammar.cc
-rw-r--r--  1 akim  akim   263165 Sep  2 12:06 _build/consistent/ 
ugrammar.cc

The syntax error on "1 1" behaves as expected with both accepting and  
consistent (i.e., mute since there are too many possibilities ;).

So I will proceed with default-reductions=all on space-starving  
targets (on which anyway I use simple error messages to get rid of the  
tables), and "accepting" otherwise.  The documentation for error- 
verbose should probably promote the use of accepting/consistent.