[Vm-dev] socketRecordSize is probably not 64-bit friendly

[David T. Lewis]

On Sun, Mar 22, 2015 at 11:27:05AM -0700, Eliot Miranda wrote:
> 
> Hi David,
> 
> On Mar 22, 2015, at 10:58 AM, "David T. Lewis" <lewis at mail.msen.com> wrote:
> 
> > 
> > On Sun, Mar 22, 2015 at 05:04:12PM +0100, Nicolas Cellier wrote:
> >> 
> >> Hi,
> >> I noticed that SocketPlugin>>socketRecordSize was so defined:
> >> socketRecordSize
> >>    "Return the size of a Smalltalk socket record in bytes."
> >> 
> >>    ^ self cCode: 'sizeof(SQSocket)' inSmalltalk: [12]
> >> 
> >> 
> >> Shouldn't it have been like its brothers in OSProcessPlugin and AioPlugin:
> >> 
> >> socketRecordSize
> >>    "Return the size of a Smalltalk socket record in bytes."
> >> 
> >>    ^ self cCode: 'sizeof(SQSocket)' inSmalltalk: [Smalltalk wordSize * 3]
> >> 
> >> 
> >> I presume we don't use self sizeof: because the Smalltalk side does not
> >> have any idea of what a SQSocket is... (an opaque struct/handle)
> > 
> > On my machine, sizeof(SQSocket) is 16 for both 32 bits and 64 bits. I would
> > have expected it to be 12 for the 32-bit case, but maybe the compiler is
> > aligning it differently for some reason.
> > 
> > In any case, "Smalltalk wordSize * 3" would not be right, because the
> > actual data structure is this:
> > 
> >    typedef struct
> >    {
> >      int   sessionID;
> >      int   socketType;  /* 0 = TCP, 1 = UDP */
> >      void  *privateSocketPtr;
> >    }  SQSocket, *SocketPtr;
> > 
> > 
> > In order to cover the common cases, it might be best to just change
> > the implementation to this:
> > 
> >     ^ self cCode: 'sizeof(SQSocket)' inSmalltalk: [16]
> 
> No, no and thrice no!  :). It should /always/ be written 
> 
>      ^self sizeof: #SQSocket
> 
> and the size answered in a sizeof: implementation marked <doNotGenerate>.  That way 
> - there is one def not one in potentially multiple clients
> - you don't even need the sizeOfSQSocket method in the first place.
> 
> Please see VMMaker.oscog
> 

Ok.

I note for the record that SocketPlugin>>socketRecordSize has a time stamp
of 2/22/2000 ;-)

Dave