🚀 go-pugleaf

In article <TIM.91Jun8191942@kahlo.cstr.ed.ac.uk> tim@cstr.ed.ac.uk (Tim Bradshaw) wrote:
> 
> Also: I've seen at least 2 new byte-compilers discussed recently:
> which is the best (or are either of them better than the original?)?

The first is one hacked up by Hallvard Furuseth and Dale Worley, which is
slightly faster than the standard bytecomp; the primary speedup is in
load-time for some packages (like gnus) because it compiles more forms than
the standard bytecomp does, and merges some adjascent top-level forms into 
one call to byte-code.  Also, it doesn't compile trivially small functions
which would run faster interpreted.

Someone once made a post claiming that this sped up gnus by 50%, and this
has been repeated several times, but it's just not true.

The version of bytecomp that is in emacs19 is roughly the same as this one.

I have my own version of the byte compiler (which Hallvard has helped with a
lot) which does the above and much more, including removal of unreachable
code, removal of calls to known side-effectless functions whose return value
is not used, compile-time evaluation of known constant forms (like (ash 1 6),
etc), open-coding of literal lambda expressions, peephole optimization of
emitted code, and support for inline functions (defsubsts).  It also produces
compile-time warning messages for mismatched calls and arglists, macros being
used before they are defined, and various syntax errors.

It's not done yet.  It hasn't been tested thoroughly, and I haven't done any
timings, but it reduces code size by 10-20%.  I'll post it when it's ready.
(This might not be for a while, though, due to my rapidly diminishing free
time...)

I don't know how much it will speed up existing code, but I have found that
making use of the inline function facility can easilly speed things up by an
order of magnitude.  Elisp function calls are a lot more expensive that you
might think.

	-- Jamie