Separate compilation: Circumventing the speed hit of Texinfo-5

[Stefan Monnier]

As you're well aware, Texinfo-5's makeinfo is much slower than
Texinfo-4's.  In Emacs, this is perceived as a serious problem since
building the Elisp manual (3MB of Texinfo source) now takes
a substantial amount of time (~3min on my desktop), which is not too
problematic for a full Emacs build, but is unbearable when you just want
to check the result of a small fix.

Some of our developers (especially those who devote a lot of effort to
writing the Texinfo docs) end up sticking with Texinfo-4 just for that
reason, and others are strongly arguing in favor of switching to some
other documentation preparation system.

I'm not in love with Texinfo, but before considering another system, I'd
rather try and make sure this problem can't simply be fixed rather
than circumvented.  IIUC the current slow processing probably doesn't have
an easy fix and so it's very unlikely for it to get addressed directly
any time soon.  But AFAICT, that doesn't mean that we couldn't speed up
the particular case of "rebuild after a small change", using a separate
compilation approach (those 3MB of source are spread over many files,
currently the largest is about 300kB and we could easily split them
even further).

Maybe making separate compilation work 100% reliably in all cases is
hard and won't happen any time soon, in which case I'd be happy with
a "mostly working" separate compilation.


        Stefan