byteSizeOf: is that correct for 64bit images? Also what about the (sqInt)(long)

Ok, I have to ask, is this correct?    Er 4 or 8 or does it matter, do we have other '4' values embedded in the interpreter?
byteSizeOf: oop
        | slots |
self flag: #Dan.
        (self isIntegerObject: oop) ifTrue:[^0].
        slots := self slotSizeOf: oop.
        (self isBytesNonInt: oop)
                ifTrue:[^slots]
                ifFalse:[^slots * 4]




Also what about
        byteSize := interpreterProxy byteSizeOf: fontFilePath cPtrAsOop.

where we have, note the (sqInt)(long)

generateCPtrAsOop: aNode on: aStream indent: anInteger

        aStream nextPutAll: '((sqInt)(long)('.
        self emitCExpression: aNode receiver on: aStream.
        aStream nextPutAll: ') - ';
                nextPutAll: ObjectMemory baseHeaderSize printString;
                nextPut: $).

Translates to
        byteSize = interpreterProxy->byteSizeOf(((sqInt)(long)(fontFilePath) - 4));

which of course is wrong.

See http://bugs.squeak.org/view.php?id=7431

--
===========================================================================
John M. McIntosh <[hidden email]>   Twitter:  squeaker68882
Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
===========================================================================

 
On Sun, Dec 20, 2009 at 02:45:59PM -0800, John M McIntosh wrote:
No, this is not correct, good catch. It should be:

  byteSizeOf: oop
  | slots |
  self flag: #Dan.
  (self isIntegerObject: oop) ifTrue:[^0].
  slots := self slotSizeOf: oop.
  (self isBytesNonInt: oop)
  ifTrue:[^slots]
  ifFalse:[^slots * ObjectMemory bytesPerWord]

I've clean a lot of this crud out of the interpreter, but there are bound
to be a few more that we have not spotted yet.

Dave

On 21.12.2009, at 00:44, David T. Lewis wrote:
Does "ObjectMemory bytesPerWord" compile to a constant?

- Bert -

 
On Mon, Dec 21, 2009 at 12:56:01AM +0100, Bert Freudenberg wrote:
Bert,

Obviously you are not using SlangBrowser ;-)

Yes, it answers a constant. The implementation is just this:

  /* Answer the size of an object memory word in bytes. */
  sqInt bytesPerWord(void) {
  return BytesPerWord;
  }

The method is inlined, so it basically translates into the CPP macro
BytesPerWord. Thus #bytesSizeOf: is translated as follows (this is with
MemoryAccess enabled so you can see the the full C code, not just the
macros):

  sqInt byteSizeOf(sqInt oop) {
      sqInt slots;
      sqInt header;
      sqInt sz;
 
  flag("Dan");
  if ((oop & 1)) {
  return 0;
  }
  /* begin slotSizeOf: */
  if ((oop & 1)) {
  slots = 0;
  goto l1;
  }
  /* begin lengthOf: */
  header = ((sqInt) ((((sqInt *) ((sqMemoryBase) + oop)))[0]));
  /* begin lengthOf:baseHeader:format: */
  if ((header & TypeMask) == HeaderTypeSizeAndClass) {
  sz = (((sqInt) ((((sqInt *) ((sqMemoryBase) + (oop - (BytesPerWord * 2)))))[0]))) & LongSizeMask;
  } else {
  sz = header & SizeMask;
  }
  sz -= header & Size4Bit;
  if (((((usqInt) header) >> 8) & 15) <= 4) {
  slots = ((usqInt) (sz - BaseHeaderSize)) >> ShiftForWord;
  goto l2;
  }
  if (((((usqInt) header) >> 8) & 15) < 8) {
  slots = ((usqInt) (sz - BaseHeaderSize)) >> 2;
  goto l2;
  } else {
  slots = (sz - BaseHeaderSize) - (((((usqInt) header) >> 8) & 15) & 3);
  goto l2;
  }
  slots = null;
  l2: /* end lengthOf: */;
  l1: /* end slotSizeOf: */;
  if (((((usqInt) (((sqInt) ((((sqInt *) ((sqMemoryBase) + oop)))[0])))) >> 8) & 15) >= 8) {
  return slots;
  } else {
  return slots * ( 4);
  }
  }

Dave
 

On 21.12.2009, at 01:17, David T. Lewis wrote:
Thanks :)

I'd find it more readable and more revealing to use BytesPerWord directly, as is done in almost all other places:

        ... ifFalse:[^slots * BytesPerWord]

- Bert -

 
On Mon, Dec 21, 2009 at 01:32:52AM +0100, Bert Freudenberg wrote:
You're right, I should have done it that way. I'll make a note to fix it
next time I commit something to SqS.

Dave
 

Careful, this is a class var.

This does not work on external plugins, so I think you'll find the proper pattern is

self bytesPerWord   which does the right thing if the code is in a plugin versus  ObjectMemory or subclass of.


On 2009-12-20, at 5:42 PM, David T. Lewis wrote:

>> BytesPerWord

--
===========================================================================
John M. McIntosh <[hidden email]>   Twitter:  squeaker68882
Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
===========================================================================

 
On Sun, Dec 20, 2009 at 06:11:11PM -0800, John M McIntosh wrote:
Indeed. Thanks John.