| Age | Commit message (Collapse) | Author |
|---|
|
|
|
isn't portable (with ppc 4xx as an example of a platform it doesn't work
on). This is why va_copy exists.
|
|
|
|
compiler so much.
|
|
|
|
|
|
|
|
index blocks).
|
|
|
|
|
|
CFG_TOYSH_DEBUG to shut up the spurious "gcc can't tell that this is never
actually used uninitialized because gcc is stupid" warnings.
|
|
|
|
|
|
doesn't work on some devices, and we can't always tell _when_ it failed), so
go to the binary search for now.
|
|
|
|
|
|
|
|
|
|
4 bytes and reduces running time by one half of one percent.
|
|
based handling of a 6-byte header with with two 24-bit integer reads. Use
xmalloc() and xzalloc().
|
|
shouldn't be a problem if we register signal handlers with sigaction(SA_RESTART)
Straighten out count and len (I generally consistently use "count" for the
current progress and "len" for the total, but this time I got them backwards
for some reason and don't want to confuse myself in future.)
|
|
for limit[] to move a test out of a loop. Unroll a single-bit get_bits()
to avoid a function call in the common case on a hot path. And one more
application of the old "two tests in one via typecasting and/or math" trick.
|
|
|
|
|
|
|
|
but it looks cleaner.
|
|
|
|
|
|
out how to turn an if/else into a multiply and subtract (saving 2 bytes).
|
|
out the bunzip2 library.
|
|
|
|
|
|
|
|
|
|
need 3. Doing varargs for this doesn't really appeal to me (bugs in waiting)
so I made an xcreate() that takes 3 args, and had xopen() call it with 0 for
the third argument. That way, if we feed O_CREAT to xopen() the permission
000 result should be easy to spot.
|
|
|
|
|
|
unrolling a small memmove.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
existing applets. Still a bit buggy, but bits of it work.
|
|
functions in main.c with USE() macros so --gc-sections can strip them.
|
|
|
|
|
|
|
