Faster than the bit twiddling, given the types of data and programs we
have now. Also means we can drop the distinction between 'safe' and
'unsafe' indexing when bounds checking is or is not required. Also now
gives consistent results with orca JS for 'strange' symbols like :, ;,
etc.
Static visibility probably useless given how we're invoking the
compilers (except maybe tcc.) But the 'noinline' attribute speeds up
compilation slightly for gcc, and significantly for clang. No useful
difference in the generated code -- most of this is cold stuff in the
UI. (gcc generated binary might get slightly larger depending on lto
choices now, but only by a few hundred bytes, probably.)
Save a level of nesting and mental bookkeeping. The control flow was
already jumping within and outside of the loop, so we aren't making this
any more complicated than it already was -- in fact, it's more simple,
because now we also don't have to remember that we're in a 'for' loop.
Was just too long and made navigation difficult. There's no actual
requirement for this procedure to be split out. But it lets us drop 2
levels of indentation and makes navigating up and down main() easier. On
the downside, it breaks up control flow.
cancel