Re: [PATCH] Add new lisp function length= with bytecode support

[Stefan Monnier <monnier@iro.umontreal.ca>]

> When the bytecode interpreter encounters a length bytecode with a list
> argument followed by a comparison bytecode it defers to the new special
> purpose length comparison functions.
> * src/bytecode.c (exec_byte_code): Change the Blength bytecode case and add
> the new functions.
> * lisp/emacs-lisp/byte-opt.el (byte-optimize-binary-predicate,
> byte-optimize-predicate): Make the byte-compiler put the length and
> comparison bytecodes next to each other when possible.
> * src/lisp.h (length_Beqlsign, length_Bgtr, length_Blss, length_Bleq,
> length_Bgeq, length_Beq): Declare new C functions.

I like that.

> +	      switch (PEEK)
> +		{
> +		case Beqlsign:
> +		  op = FETCH;
> +		  v1 = POP;
> +		  TOP = length_Beqlsign (TOP, v1);
> +		  break;
> +
> +		case Bgtr:
> +		  op = FETCH;
> +		  v1 = POP;
> +		  TOP = length_Bgtr (TOP, v1);
> +		  break;
> +
> +		case Blss:
> +		  op = FETCH;
> +		  v1 = POP;
> +		  TOP = length_Blss (TOP, v1);
> +		  break;
> +
> +		case Bleq:
> +		  op = FETCH;
> +		  v1 = POP;
> +		  TOP = length_Bleq (TOP, v1);
> +		  break;
> +
> +		case Bgeq:
> +		  op = FETCH;
> +		  v1 = POP;
> +		  TOP = length_Bgeq (TOP, v1);
> +		  break;
> +
> +		case Beq:
> +		case Bequal:
> +		  op = FETCH;
> +		  v1 = POP;
> +		  TOP = length_Beq (TOP, v1);
> +		  break;
> +
> +		default:
> +		  TOP = Flength (TOP);
> +		}
> +	    }

Please move most of that to a separate function (which I guess will take
the list, the op and the value to which to compare the list).

> +/* The following are used above in the Blength case. Each assumes s1
> +   is a number or marker and s2 is a list. */
> +
> +Lisp_Object
> +length_Beqlsign (Lisp_Object s1, Lisp_Object s2)
> +{
> +  Lisp_Object val = Qnil;
> +
> +  CHECK_NUMBER_OR_FLOAT_COERCE_MARKER (s1);
> +
> +  if (__builtin_expect (FLOATP (s1), 0))
> +    {
> +      s2 = Flength(s2);
> +      val = arithcompare (s1, s2, ARITH_EQUAL);
> +    }
> +  else
> +    {
> +      intptr_t n = XINT (s1);
> +      intptr_t i = 0;
> +      FOR_EACH_TAIL (s2)
> +	{
> +	  i++;
> +	  if (i > n)
> +	    return val;
> +	}
> +      CHECK_LIST_END (s2, s2);
> +      if (i == n)
> +	val = Qt;
> +    }
> +
> +  return val;
> +}
> +
> +Lisp_Object
> +length_Bgtr (Lisp_Object s1, Lisp_Object s2)
> +{
> +  Lisp_Object val = Qnil;
> +
> +  CHECK_NUMBER_OR_FLOAT_COERCE_MARKER (s1);
> +
> +  if (__builtin_expect (FLOATP (s1), 0))
> +    {
> +      s2 = Flength(s2);
> +      val = arithcompare (s1, s2, ARITH_GRTR);
> +    }
> +  else
> +    {
> +      intptr_t n = XINT (s1);
> +      intptr_t i = 0;
> +      FOR_EACH_TAIL (s2)
> +	{
> +	  i++;
> +	  if (i >= n)
> +	    return val;
> +	}
> +      CHECK_LIST_END (s2, s2);
> +      if (i < n)
> +	val = Qt;
> +    }
> +
> +  return val;
> +}

Similarly, here (and below), I'm wondering if we can't reduce the code
duplication.  Furthermore, you might like to declare them static, so the
compiler is more likely to inline them.

> +Lisp_Object length_Beqlsign (Lisp_Object, Lisp_Object);
> +Lisp_Object length_Bgtr (Lisp_Object, Lisp_Object);
> +Lisp_Object length_Blss (Lisp_Object, Lisp_Object);
> +Lisp_Object length_Bleq (Lisp_Object, Lisp_Object);
> +Lisp_Object length_Bgeq (Lisp_Object, Lisp_Object);
> +Lisp_Object length_Beq (Lisp_Object, Lisp_Object);
 
Don't expose them in lisp.h, since they're only used in bytecode.c.


        Stefan