[racket] Continuation marks

The 'preserves-marks flag makes sense only in the context of the JIT's
implementation.

At the Racket level, there's no way to write a function `f` so that

 (with-continuation-mark
  'x
  'my-value
  (begin
   (f)
   (continuation-mark-set-first #f 'x)))

produces anything other than 'my-value. Even if `f` contains a
`with-continuation-mark` form to set a mark with the key 'x, returning
from `f` necessarily means leaving the dynamic extent of that form, so
the mark will not be visible.

It turns out that the JITted form of `with-continuation-mark` is not
responsible for popping a mark stack as it returns. Instead, every
non-tail position is wrapped with a kind of push and pop of the
continuation-mark stack. Since `(f)` appears in a non-tail position
within with `begin` form, the call with be wrapped.

If `f` is known to not have a `with-continuation-mark` as a tail form,
however, then the push and pop of the continuation-mark stack will not
be necessary around the `(f)` call. The 'preserves-marks flag tells the
JIT that it can make that assumption. (In retrospect, a better name
would be 'has-no-wcm-in-tail-position.)


At Sat, 6 Sep 2014 11:36:37 -0300, Gustavo Massaccesi wrote:
> I agree that to get this output, using continuation marks is a bad
> idea. It'd be much better to use a parameter (or set!ing a global
> variable at a last resort).
> 
> I'm trying to understand continuations marks. I'm reading some parts
> of the Racket compiler, and internally the functions have some flags
> that are not visible from the Racket code.
> 
> One of the flags is 'preserves-marks. From
> http://docs.racket-lang.org/raco/decompile.html
> 
> > The preserves-marks? field is true if calling the function is expected to 
> leave continuation marks unchanged by the time it returns.
> 
> So: Can a function change the continuation marks after it returns?
> What does that exactly mean? Can I use that to complete the code in my
> original e-mail? Is that a good idea for a real program?
> 
> (The last question is easy: No. :) )
> 
> 
> I also read the answer from Matthias Felleisen. I'm not sure that I
> understood it completely, but somehow it give me a new idea to solve
> this. But I need to try it before I'm sure it works.
> 
> 
> Gustavo
> 
> On Fri, Sep 5, 2014 at 7:04 PM, John Clements <clements at brinckerhoff.org> wrote:
> >
> > On Sep 5, 2014, at 2:24 PM, Gustavo Massaccesi <gustavo at oma.org.ar> wrote:
> >
> >> I'm trying to add a continuation mark inside a function, but I want
> >> that the new mark to be visible after the functions returns. (Mostly
> >> for curiosity, not an actual problem.)
> >>
> >> I want something like this:
> >>
> >> ;--
> >> #lang racket/base
> >>
> >> (define (display-cm-x)
> >>  (displayln
> >>   (continuation-mark-set->list
> >>    (current-continuation-marks) 'x)))
> >>
> >> (define (add-cm-x v)
> >>  '???)
> >>
> >> (define-syntax-rule (with-cm-x v body ...)
> >>  (with-continuation-mark 'x v
> >>    body ...))
> >>
> >> (let ()
> >>  (display-cm-x) ;==>()
> >>  (with-cm-x 7
> >>    (display-cm-x)) ;==>(7)
> >>  (display-cm-x) ;==>()
> >>  (add-cm-x 7)
> >>  (display-cm-x) ;actual==>(), expected==>(7)
> >>  )
> >> ;--
> >>
> >> The last line prints (), but I wish it prints (7).
> >
> > This is going to sound flip, but … it sounds like you don’t want continuation 
> marks. The whole point of continuation marks is that they’re associated with 
> continuations, and will disappear when those continuations are gone.
> >
> > The example you give isn’t quite enough to figure out exactly what you want; 
> it sounds like some combination of fluid-let and … something else.
> >
> > Maybe you can describe why it is that you want this behavior?
> >
> > John
> >
> 
> ____________________
>   Racket Users list:
>   http://lists.racket-lang.org/users