_AS_ENSURE_STANDARD_FDS: use lsof and fstat when possiblezack/ensure-standard-fds

This follow-up patch adds additional ways of detecting whether fds 0,
1, 2 are closed, using the ‘lsof’ and ‘fstat’ utilities.  The former
is not a standard component of any OS, but is very widely installed
and can produce machine-parseable output; the latter ships with many
BSD variants but its output is in a fixed format not designed for
machine parsing.  Both of them may fail due to privilege restrictions.

The biggest problem with using these is that we have to run a program
and inspect its output.  I tried capturing the output using command
substitution, but that is often implemented using a pipe from the
child process to the parent shell; if fd 0 (for instance) started out
closed in the parent, the pipe is likely to occupy that fd number
right when lsof/fstat is inspecting the parent’s file table, causing a
false report that fd 0 is open.  The only alternatives I am aware of
are temporary files or named pipes, both of which involve creating
directory entries somewhere...and we can’t assume we have mktemp(1).
This patch uses a file in the current working directory, but that
breaks all of the tests that assume ‘configure --help’ will not write
to the current working directory (which seems like a reasonable
promise for us to make, tbh).

Better ideas solicited.

* lib/m4sugar/m4sh.m4 (_AS_ENSURE_STANDARD_FDS): If /proc/<pid>/fdinfo
  is not available, try to use lsof or fstat.

0 files changed, 0 insertions, 0 deletions