[ragel-users] Parsing From Ring Buffers

[Adrian Thurston]

Yeah, that shouldn't be too hard. In fact it is already done for java and
ruby code generators due to the lack of pointers. These languages only
have table-based drivers though and since you are after speed you probably
want to make it work with -G2. Should be straightforward. Send the results
here if you try it :)

Adrian


On Thu, 31 Jul 2008, William Ahern wrote:

>
> Cursorily looking at the generated code, it appears to be deceptively easy
> to turn `p', `pe', and `eof' into indexes, rather than pointers, given the
> `fpc' macro. Perhaps by simply adding a second macro to increment and
> decrement the index?
>
> Am I missing anything that would throw a monkey wrench into such a patch?
> This could make scanners more efficient, because there would be no need to
> use memmove().
>
> It might even work without a patch, but technically it's undefined behavior
> in C to derive pointer values outside arrays or malloc'd blocks. Compilers
> like TinyCC have the ability to catch such errors, which can be useful when
> tracking down bugs. But to keep from having to re-adjust `ts' and `te', and
> so that Ragel sees a linear input space during scanning, the indexing would
> have to extend beyond the ring buffer (and fpc would just be a simple
> modulus operation).
>
>
> >