Portability of mknod

[Daniel E. Macks]

A bunch of info/t tests in texinfo-6.3 were SKIPped on my OS X 10.11
box, all as a result of 'mknod' being detected but then failing in
Init-inter.inc. I do have /sbin/mknod, but it's not GNUish. In
particular, it doesn't support the "p" token to cause creation of a
named FIFO pipe. If I install GNU coreutils, I do get a mknod that
behaves the way the tests want, and the tests succeed. But darwin does
have 'mkfifo', and if I use that instead of mknod, the tests also
succeed.

It's not as simple as setting an autoconf variable whose value could
be "mknod" or "mkfifo", because mkfifo implicitly knows to create
named pipes (no token to pass). That is, GNU wants:

  mknod FOO p

but darwin needs:

  mkfifo FOO

so it might be cleaner to have autoconf propagate a boolean
HAVE_GNU_MKNOD to control which commands are used. Or else autodetect
mkfifo and then propagate MKFIFO via TESTS_ENVIRONMENT (using it if
detected, else using mknod)? I can work up a patch if someone says
what sort of test/default/fallback behavior is preferred.

dan

--
Daniel Macks
address@hidden