What to do with extract/store floating()?

[Andrew Cagney <ac131313 at cygnus dot com>]

Hello,


At present GDB has the functions:

  DOUBLEST extract_floating (const void *in, int);
  void store_floating (void *, int, DOUBLEST);

where that ``int'' parameter is the length of the floating-point value 
being extracted/stored in bytes(1).

These functions then map that length onto a target type and from there 
do a floating point conversion.  This leads to a convoluted double flip:

gdbtypes.c:
	TYPE_LENGTH(type) = TARGET_DOUBLE_BIT / TARGET_CHAR_BIT;
client code:
	extract_floating (buf, TYPE_LENGTH (type));
doublest.c:
	if (length == TARGET_DOUBLE_BIT / TARGET_CHAR_BIT)
	  ..


Originally I was going to replace all these with:

	floatformat_to_doublest()
and 
floatformat_from_doublest()

however, in the light of the clarified meaning of ``struct type . 
length'', that initial change would be wrong (it is the total size of 
the type, the FP value is often smaller).


I'm now considering changing things to:

  DOUBLEST extract_floating (const void *in, struct type *type);
  void store_floating (void *, struct type *type, DOUBLEST);

store_floating(), for instance, would zero all TYPE_LENGTH(type) bytes 
and then store DOUBLEST based on TYPE_FLOATFORMAT().


A variation on this (more incremental, less risk) would be to introduce 
new functions with the above signatures (extract_doublest() and 
store_doublest()?) and then deprecate the old extract_floating() and 
store_floating() functions.  A quick grep shows ~34 calls, 4 in core 
code and one in stabsread.c!


Any preference?  I've a patch for the middle option but I'm not sure how 
correct it will be without testing.

	Andrew

(1) Please don't ask if they are target or host bytes.