Re: [Boost-users] xpressive: set vs standard alternation compile problems

From: Eric Niebler (eric_at_[hidden])
Date: 2008-02-19 13:12:05

Zachary wrote:
> Hi,
>   I'm a new user of xpresssive and I'd like to first give a big thank 
> you for this wonderful tool.

Thanks!

>  I'm having problems with the syntax for static regexes.  The cheat 
> sheet in the documentation says that |(set= 'a','b','c')| and 
> |as_xpr('a') | 'b' |'c'| are equivalent yet the example
> 
>     sregex parentheses;
> 
>     parentheses = '(' >> *(keep(+~(boost::xpressive::set='(',')'))|by_ref(parentheses)) >> ')';
> 
> compiles while
> 
>     sregex parentheses;
> 
>     parentheses = '(' >> *(keep(+~(as_xpr('(')|')'))|by_ref(parentheses)) >> ')';
> 
> fails to do so (code and abbreviated errors attached).

I see how my docs led you astray. Although they match the same strings, 
there actually is a difference between (set='a','b','c') and 
(as_xpr('a')|'b'|'c'). The first is a set and can have its complement 
taken (e.g., operator~). The second is just a bunch of regexes in 
alternate, and cannot have its complement taken.

Consider this a doc bug, which I'll fix.

>  Additionally, 
> combining the original lines into one construction, like so:
> 
>     sregex parentheses = '(' >> *(keep(+~(boost::xpressive::set='(',')'))|by_ref(parentheses)) >> ')';
> 
> compiles but running the code fails with:
<snip>

This is an invalid use. It's no different than:

   int i = i + 1; // oops!

You may be surprised to find this compiles, but invokes undefined 
behavior. It's a sad fact in C++ that objects are in scope before they 
are initialized, leading to these sorts of nasty situations.

If you would like to put this all in one line, use xpressive::self, as:

   sregex parentheses = '(' >> *(keep(+~(set='(',')'))|self) >> ')';

HTH,

-- 
Eric Niebler
Boost Consulting
www.boost-consulting.com