Re: [Boost-users] Boost.xpressive: compiled program crashes :-(

From: George Petasis (petasisg_at_[hidden])
Date: 2008-01-11 12:21:03

O/H Eric Niebler έγραψε:
> George Petasis wrote:
>> Actually, xpressive throws an exception:
>>
>> xpressive error: error_stack: There was insufficient memory to determine 
>> whether the regular expression could match the specified character 
>> sequence. (12)
>>
>> Any ideas on that? can I increase the stack size under windows?
> 
> That's some grammar! I'm amazed you actually got it to compile. :-P As 
> you've already figured out, you're blowing your stack. I can see some 
> ways to simplify your grammar, but I think your best bet would be to 
> switch to a parser generator like Boost.Spirit. Xpressive's exhaustive 
> backtracking behavior makes it use stack space more greedily than Spirit.

I have used spirit in the past. I am trying to use xpressive, because:

a) it has backtracking (which is really cool :-) )
b) it is easier to get it compile, at least with VC++ 2005 (I get far 
more failures with spirit)

But of course, backtracking comes with a cost :-) Stack exhaustion...

> 
> If you want to try sticking with Xpressive, here are some things to try:
> 
> - Increasing your program's stack size. You're using MSVC? See the 
> /STACK switch for the linker.
I have tested up to 262 MB, but no change :-) My grammar is probably too 
complex :-). How exactly xpressive uses the stack? Is it a recursive 
descendant parser?
> 
> - Avoid using by_ref(a_sregex) where it's not really necessary. For 
> instance, you have lots of regexes like this:
> 
>> s_scl__com_ =
>>   (as_xpr(","));
>>
>> s_scl__pvl_ =
>>   (as_xpr("-"));
>>
>> s_scl__sls_ =
>>   (as_xpr("/"));
> 
> Embedded regexes use more stack space and incur a perf hit that you 
> wouldn't get if you had used ordinary strings for this instead.
Its in my plans to fix this, and detect only the cases references are 
used...
> 
> - Use keep() for the parts of your grammar that do not need exhaustive 
> backtracking semantics. keep(<some regex>) matches <some regex> and then 
> reclaims the stack space used by <some regex>. But then <some regex> can 
> only match in one way, and no alternatives are tried.
I didn't know about keep(). I will read the manuals...
> 
> - If you have patterns like (string1 | string2 | string3 ...) where 
> stringN are simple string literals, you're better off using a symbol 
> parser (assuming you're using xpressive 2.0). See 
> <http://tinyurl.com/2rd5d7>.
To say the truth, I tried to use a symbol table, but I didn't manage to 
get it to compile. I always got an error when I was using the table in 
the grammar rules. How can I check what version of xpressive I am using?

Initially, I downloaded the latest release of the boost library. But the 
file regex_actions.hpp was missing, so I downloaded an newer release of 
xpressive (2.0.1 I think)...
> 
> Best of luck!
> 
Thanks :-)

Regards,

George