Re: [Vala] The state of closures

["Sam Liddicott" <sam liddicott com>]

On 28/04/11 16:44, Luca Bruno wrote:
> On Thu, Apr 28, 2011 at 04:38:43PM +0100, Sam Liddicott wrote:
> > On 28/04/11 16:08, Luca Bruno wrote:
> > > On Thu, Apr 28, 2011 at 03:58:32PM +0100, Sam Liddicott wrote:
> > > > This feature may be less useful for vala if it does stack-tear-down
> > > > for interruptable functions, but perhaps more suited if it uses
> > > > co-routines with different stacks so that the stack is switched and
> > > > not torn-down when an interruptable function is interrupted.
> > > Yes that's the problem, with vala you may pass the closure to another
> > > function and exit from the current scope, while with gcc nested function
> > > the stack is lost once you exit the outer function.
> > And it's hard to control this or know when it is happening by accident.
> >
> > I wonder if it is worth considering libpcl or other co-routine
> > libraries - it would take away the need for a lot of effort to get
> > fuller closure support.
> >
> > I've also been thinking of ways to detect when vala closures go
> > stale - one way is for the data pointer to be indirected via a heap
> > allocated pointer. This could be set to null by the outer function
> > when it exits, and so any use after this point can be detected - as
> > when the wrapper trampoline function is executed it will detect that
> > the data pointer points to NULL instead of pointing to the actual
> > data pointer.
> I don't get what's the problem with actual vala closures that you need to
> find an alternative
yes.. it's been a few years since I piped up.

Part of the problem is (or was) with interruptable functions which 
tear-down the stack and thus invalidate the closure for interruptable 
functions.

Another is that last time I checked closures still could not access 
local variables in the enclosing function.

I was suggesting that if GCC closure support were used and co-routines 
for interruptable functions that then closures could access local 
variables of the enclosing functions quite well. Otherwise

This stemmed from a time when I was trying to use vala to write samba 4 
modules where the closure support (with access to variables of the 
enclosing function) would have been very useful if it could survive the 
tear-down of the stack (which libpcl co-routines provide).

Sam