make(1): Does not work like the other tests. Its Makefile is
self-testing.
m4(1): It uses complex voodo to test GNU m4(1) features.
To the new framework. I had worried about passing the binary data that
uudecode(1)'s test passes to diff(1) might give a user something nasty,
but this is unlikely to happen as even with an unmodified old nasty
diff(1) which doesn't recognise many binary files, these binary files
are recognised. Using $DIFF instead of `diff' in the library and making
it possible to override this with `cmp -s' might be nice some day, but
as of this second, there's no immediate need.
to handle the ones which output to stdout and have output in regress.$test.out,
etc. More freeform macros should and will be written, but these are the most
prominent and most straightforward sort of tests we have around, so it makes
sense to try to accomodate them.
expansion of embedded variables in the left-hand-side of an assignment
expression, using the simplest case - hiding recursion using nil-expanded
variables.
uuencode(1), and set a umask, so that the mode in the header is predictable.
If it varies, then the test is right to fail.
Remove the note about this test falsely failing, with that in mind.
and for proper behaviour of some sed functions given a nil pattern space,
as fixed in PR 34813.
The test for G was based on the test in the PR. The nil pattern space test
is slightly different as we need to get *some* output, as the core dump will
also produce no output (old behaviour) and turn up falsely that the utility
is working fine.
