Re: Advice needed, please

[Hans Aberg]

On 8 Jun 2009, at 22:37, xelanoimis wrote:

> All right, I changed the grammar to be more like this ansi c here:
> http://www.quut.com/c/ANSI-C-grammar-y.html
> And I got rid of the conflicts :) All except the "if else" which I
> understand is OK to have.

If it is the dangling "else", you can get rid of that, too, by setting  
%nonassoc on the "then" and "else" tokens.

> The issue with the %%error_verbose was that I had bison v1.24 which  
> didn't
> support it.

It is ancient.

> For the sake of more info, I installed v2.4 and managed to compile my
> grammar with it. I'm not sure what is the standard procedure for  
> correct
> installing, but I copied the bison.exe and m4.exe (whatever it is)  
> in my
> working folder and then started to gather required dll-s from the  
> net. Also
> I had to copy some bison/share... folder in my c:/programfiles. I  
> saw the
> path hardcoded in bison.exe. Is this normal? Can't I have them  
> somewhere
> else?

It depends how you compiler treats it, but on Posix computers (don't  
know about your platform), the paths become hardcoded by ./configure  
before the compile in the sources. Then at runtime, 'bison' must find  
'm4', which must be sufficiently recent.

> Now I have two more questions:
>
> 1. As I said, I split the expression rule to ensure correct  
> precedences
> (like in ansi C exp_add, exp_mul, exp_shift,...). How is this  
> affecting the
> resulting parser? I guess the functionality it's the same as having  
> a single
> expression rule and the tokens arranged correctly when declared. I  
> think the
> parser is a little bigger with more rules (not really sure). But  
> which one
> is it faster?

Most time is typically spent in the lexer and the actions, so just  
write the grammar to be correct and well structured. If the lexer is  
generated by Flex, it should not gulp too much in each match - buffer  
problem.

> 2. Is it worth for me to use the new bison version over the old  
> 1.24? The
> output parser.cpp file was smaller before (70KB vs 100KB). Is the  
> new one
> significantly faster or it just reports errors better?

The old one is so ancient, nobody here can remember :-). Only the  
latest version is supported.

  Hans