[Haskell-cafe] pointer equality

[Bertram Felgenhauer]

Carl Howells wrote:
> On Tue, Jul 19, 2011 at 11:14 PM, yi huang <yi.codeplayer at gmail.com> wrote:
> > 2011/7/20 Eugene Kirpichov <ekirpichov at gmail.com>
> >>
> >> reallyUnsafePointerEq#, and it really is as unsafe as it sounds :)
> >>
> > Why is it so unsafe? i can't find any documentation on it.
> > I think always compare pointer first is a good optimization.
> 
> False positives and false negatives are both possible, depending on GC
> timing.

At the moment, as implemented in ghc, false positives are not possible,
because GC only happens on allocation [*], and there is no allocation
happening in that primitive operation. I don't think this is going
to change without a total rewrite of ghc, since allowing GC (i.e.,
moving pointers) at arbitrary times would be a fundamental change to
the STG execution model.

Pretty much everything else imaginable can happen; in particular, if two
variables a and b compared equal at one point, they may later become
different pointers again. In the parallel RTS, if you're unlucky, this
may even be a permanent effect.

Best regards,

Bertram

[*] we'll have thread-local GC for the first generation soon, but a lot
of effort went into ensuring consistentcy of pointers seen by other
threads.