rocky / elisp-bytecode Goto Github PK

Will probably require building older versions of Emacs

The byte-code manual lists goto as having a stack effect of -1+0. However looking at the source this doesn't seem to correct. It does not pop at all. Unless I am missing something?

	CASE (Bgoto):
	  op = FETCH2;
	op_branch:
	  op -= pc - bytestr_data;
	op_relative_branch:
	  if (BYTE_CODE_SAFE
	      && ! (bytestr_data - pc <= op
		    && op < bytestr_data + bytestr_length - pc))
	    emacs_abort ();
	  quitcounter += op < 0;
	  if (!quitcounter)
	    {
	      quitcounter = 1;
	      maybe_gc ();
	      maybe_quit ();
	    }
	  pc += op;
	  NEXT;

Many opcodes are Emacs Lisp functions. Perhaps we should link back into the texinfo for those.

This is great! Thank you for doing it.

It would be even better if it was part of the documentation distributed with Emacs, no?

Is that the plan?

I think the attribution for the quote found lisp/emacs-lisp/byte-opt.el to Eric Naggum isn't correct. The bit about racehorses and pigs is from East of Eden but more significantly, I think Jamie Zawinski may be the one who added it and the bit about turbocharged VW bugs below it (in Lucid Emacs, presumably) since it's present in the initial checkin of the file by Jim Blandy circa Emacs 18/19.

Are BRgoto etc. ever emitted? If no, were they, and in which versions?

I can't find any live code that would emit them, and I don't think they're disassembled properly. On the other hand, bytecode.c does seem to handle them.

Hi,

switch-like jump table.  Top of stack is a variable reference.  Below that is ahash table mapping compared values to instructions offsets.

The hash table values aren't offsets, they're pc values.

So it's not:

pc = pc + ht.lookup(TOS);

It's:

pc = ht.lookup(TOS);

This is seen in bytecode.c:

	op_branch:
	  op -= pc - bytestr_data;
	op_relative_branch:
	  pc += op;
	  NEXT;

Inlining the computation we get pc = pc + op - pc - byte_strdata == op - byte_strdata.

It doesn't appear to do anything, and it's not actually emitted...

Any chance of adding page numbers on the ready-made PDF?

Thanks.

In the document:

1 byte for call[0] .. call[4]; 2 bytes for call[5], 8-bit operand; 3 bytes for call[6], 16-bit operand.

Apparently call[0] to call[5] should be 1 byte, call[6] is 2 bytes and call[7] is 3 byte

"Stephan Monier" should be "Stefan Monnier". One 'f', Two 'n's.

(Assuming there aren't two people with extraordinarily-similar names with sufficient knowledge to have contributed to this particular project.)

What kinds of optimization are performed? When were they introduced? Examples of optimizations.