lj_str_new() had a separate fast-path and slow-path. This was bad
because (a) the fast-path was complex and (b) the fast-path was
actually slower than the slow-path in practice and (c) in practice it
could cause confusing performance problems depending on the memory
alignment of any often-reused string buffers in a program.
This change specifically makes the 'life' benchmark faster and more
robust to memory layout.
This requires shuffling a bunch of stuff around to get a CLI lua
executable that uses an automation-like environment, but that's
something that'll be nice to have in the future anyway.
Busted indirectly depends on lfs and we currently can't use external
binary modules, so add a copy of lfs to the repo and build it with
Aegisub.
Rather than having it be a configurable option, simply build in final
release mode if Aegisub is being built from a signed tag, and get the
version information from that tag.
Drop the version number from the OS X app bundle when not building from
a tag as it done everywhere else. Drop the arch suffix since ppc hasn't
been tested in years and isn't worth testing as it's only going to get
less relevant as time goes on.
Remove some redundant definitions in the build system.