using the appropriate (unsigned) format specification. This prevents
integer overflow when ULLONG_MAX and (on some architectures) ULONG_MAX
are used to initialize an intmax_t and then displayed as the signed
value -1. (A different approach was suggested in the bug report,
which I did not use.) If other limits are defined to be unsigned,
they could be moved here.
PR: 164049
Reported by: Marcus Reid
intent of the Standard.
- Make getconf able to distinguish between configuration variables which
are entirely unknown and those which are merely not defined in the
compilation environment. The latter now get a more appropriate
"undefined\n" result rather than a diagnostic. This may not be
exactly right, but it's closer to the intent of the Standard than
the previous behavior.
- Support ``programming environments'' by validating that the environment
requested with the `-v' flag is the one-and-only execution environment.
(If more environments are supported for some platforms in the future,
multiple getconf(1) executables will be required, but a simple edit in
progenv.gperf will enable automatic support for it.) Document POSIX
standard programming environments.
- Add all of the 1003.1-2001 configuration variables. FreeBSD does not
support all of these (including some that are mandatory); getconf will
later be fixed to break the world should a required variable not be
defined.
As a result of all these changes, gperf is no longer adequate. Keep the
overall format and names of the files for now, to preserve revision history.
Use an awk script to process the .gperf files into C source, which does a
few things that gperf, as a more general tool, cannot do. The keyword
recognition function is no longer a perfect hash function.
This may obviate the need for gperf in the source tree.
- Add a small compile-time regression test to break the build if any of the
.gperf files declare conflicting token sets. (gperf itself would have done
this for the simple case of duplicate tokens in the same input file.)
