When dealing with network protocols and file formats, it's common to
have to read or write 16-, 32-, or 64-bit datatypes in signed or
unsigned flavors. Common Lisp sort of supports this by specifying
:element-type for streams, but that facility is underspecified and
there's nothing similar for read/write from octet vectors. What most
people wind up doing is rolling their own small facility for their
particular needs and calling it a day.
This library attempts to be comprehensive and centralize such
facilities. Functions to read 16-, 32-, and 64-bit quantities from
octet vectors in signed or unsigned flavors are provided; these
functions are also SETFable. Since it's sometimes desirable to
read/write directly from streams, functions for doing so are also
provided. Reading/writing IEEE singles/doubles (i.e. single-float
and double-float) is also supported, with limitations on platforms
that do not fully implement IEEE floating point.
In addition to centralizing such facilities, NIBBLES also aspires to become a place where compiler optimizations can be written once and used everywhere. The intention is that (eventually):
(nibbles:sb32ref/le vector index)will compile (with any necessary safety checks) to a MOVSX
instruction on an x86oid processor in SBCL (or other implementations)
if vector and index are of appropriate types.
I remember reading a post on comp.lang.lisp that suggested the designers of Common Lisp ignored the realities of octets and endianness and so forth. This library is a small step towards remedying that deficiency.
