Re: Boost.Parser parsing length-prefixed strings efficiently

From: Zach Laine (whatwasthataddress_at_[hidden])
Date: 2024-03-01 01:25:05

On Thu, Feb 29, 2024 at 2:13 PM Mohammad Nejati [ashtum] via Boost
<boost_at_[hidden]> wrote:
>
> Hi everyone,
>
> I have two questions about the following code, which attempts to parse
> length-prefixed strings using the proposed Boost.Parser:
>
> ```
> std::string input = "[5]HELLO[5]WORLD";
>
> unsigned int size = 0;
> auto action = [&size](auto & ctx) { size = bp::_attr(ctx); };
>
> auto lps_parser = '[' >> bp::uint_[action] >> ']' >>
> bp::repeat(std::ref(size))[bp::char_];
>
> auto result = bp::parse(input, *lps_parser);
> ```
>
> 1. Is it possible to eliminate the need for the action and size
> variable by using some kind of placeholder in bp::repeat that utilizes
> the attribute of the first parser?

Well, if you put everything into a rule, you could have a locals
struct that had room for that size.  Then you could refer to that
local in the parameter to repeat().  I'll be adding a more explicit
example for locals and parameters soon.  In the meantime, you can look
at the tests to see them in action.

> 2. How can we make this part of the parser more efficient:
> bp::repeat(std::ref(size))[bp::char_]
> The current implementation seems to loop and execute bp::char_ parser
> on each character, while all it needs to do is chunk a portion of the
> input string.

Huh.  Yeah, I definitely don't provide a way for a parser to bump the
input position forward.  You could do it in a custom parser you wrote
yourself.  Each parser has a mutable reference to the current iterator
position; you could write one that parsed a number N in brackets, and
then did "first += N" if first was random_access.  It's at odds with
how the iterators are assumed to work -- Parser only assumes
forward_ranges, except for the string_view[] directive, which requires
contiguous_ranges.

Zach