Re: [Xen-devel] [PATCH 2/2] libxl: Have flexarray using the GC

[Ian Jackson <Ian.Jackson@xxxxxxxxxxxxx>]

Anthony PERARD writes ("Re: [Xen-devel] [PATCH 2/2] libxl: Have flexarray using 
the GC"):
> On 09/28/2012 06:27 PM, Ian Jackson wrote:
> > If gc is NOGC then this does the wrong thing.  Callers should be able
> > to specify NOGC for a flexarray which they want to survive across
> > multiple calls into libxl.
> 
> I'm not sure I see your point here.  If a something needs a NOGC'ed
> flexarray, then flexarray_make(NOGC, x, x) will give this.

So then you put the flexarray in some long-term data structure, which
survives the call to libxl and therefore the gc.  Is it safe for
NOGC's gc*, derived from the now-destroyed actual gc, to be embedded
in the flexarray and reused later in a different libxl call with a
different gc (but the same ctx) ?

> > For this all to work correctly, including error handling, I think
> > flexarray_grow and its callers need to take a gc from the context.  It
> > would be wise to assert that the either 1. both the gc passed to make
> > and grow are NOGC or 2. they are the same.
> 
> All right, I'll add gc to _grow and it's caller. And will assert on both
> condition.

Feel free instead to explain to me why I'm wrong :-).

Ian.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel