[ragel-users] Re: fexec bug in v5.15?

[Adrian Thurston]

As separate statements it doable but it would require control flow analysis 
of the host language in order to determine if one statement comes after the 
other. Control flow analysis is a huge undertaking, one that I would have to 
say is too much for Ragel.

But combination statements that hold and return would work.

-Adrian

Carlos Antunes wrote:
> On 11/6/06, Adrian Thurston <thurs... at cs.queensu.ca> wrote:
>> As it turns out, it's not safe to manipulate the current pointer in
>> pattern actions. I had realized this recently and in 5.15 an error is
>> emitted if you use fhold in a pattern action. It didn't occur to me that
>> I have to do the same for fexec. I should also put a warning in the docs
>> about directly manipulating p in pattern actions.
>>
>> The reason is that as a part of restarting the scanner, it's sometimes
>> necessary to manipulate p. Sometimes you need to shift it back one.
>> Other times you need to set it to tokend. Other times you don't need to
>> do anything at all. This all depends on the contents of the patterns. So
>> if you change p first, you mess the scanner up in ways that are hard to
>> predict.
>>
> 
> Ok, I see. Maybe fhold and fexec could be allowed in those cases where
> the scanner doesn't need to restart? For example { fret; fhold; } or {
> fnext whatever; fexec tokstart; }? Basically, fhold and fexec could be
> specified in all situations where the scanner knows it wouldn't be
> restarted. Is this doable? Obviously, it would also be nice to specify
> as an example { fhold; fret } and get the same behavior as { fret;
> fhold; }.
> 
> Carlos
>