Re: [Chicken-users] hash-table-walk + hash-table-delete!

[Kon Lovett]

On Mar 5, 2007, at 10:26 AM, Graham Fawcett wrote:

> On 3/5/07, Kon Lovett <address@hidden> wrote:
> > On Mar 5, 2007, at 8:47 AM, Zbigniew wrote:
> >
> > > SRFI-69 appears to provide no guarantee of safety, so for maximum
> > > portability (including to future implementations) you might want to
> > > iterate manually over a copy of the keys obtained via hash-table- 
> > keys.
>
> > Zbigniew is right but you could conditionally expand:
> >
> > (cond-expand
> >    (chicken
> >      ... hash-table-walk delete ... )
> >    (else
> >      ... safe delete ... ) )
>
> Right. (Kon, can I take this to mean that you also think a
> "hash-table-walk delete" is safe in the current Chicken
> implementation?)

Yes.

##sys#for-each doesn't care about the previous pair in the chain when  
looping, hash-table-walk passes the entire bucket chain on to for- 
each & the hash vector is not resized. (I think you already pointed  
this out.)

The only issue I see is if someone does a hash-table-delete! during a  
hash-table-walk where the bucket is the same as the currently  
"walked" bucket & the deleted entry is "ahead" of the current bucket  
link. i.e. deleting an arbitrary entry conditional on the currently  
"walked" entry.

However I think a '(hash-table-remove! hash-table (-> key value  
boolean))' procedure should be added to "extras.scm" just to be  
"safe". I suppose it could fluid-bind the hash-table-delete!  
procedure to avoid someone doing the above; although they could still  
cache the binding beforehand - perversity tends to the maximum.

Best Wishes,
Kon

> Zbignew wrote:
> > I think this is an oversight in SRFI-69.
>
> I agree; it's a surprising omission.
>
> Graham