This matches the existing use of "got -r path cmd" and "git_commit path args".

As reported by naddy, this is a non-portable Korn shell construct. Instead, use the one argument array then make faux space-delimited string "arrays" for the other ids; they can still be indexed with pop_idx. improved by + ok naddy@

This begins enabling the use of keywords in got wherever commit ids or references are used, with more work intended to expand support across all such instances (e.g., branch, checkout, etc.), and add more keywords. The keywords ":base" and ":head" can be passed to 'got {diff,log,update} -c' commands as a substitute for the corresponding commit hash id. Keywords and references can also be modified by appending a ':+' or ':-' and an optional integer N to specify by first parent traversal the Nth generation descendant or antecedent, respectively. If N is omitted, a '1' is implicitly appended. tweaks + ok op and stsp

patch from Josiah Frentsos, thanks!

Problem marked with a TODO comment by me ages ago and rediscovered by naddy@ ok op@

Now that the use of sed(1) in tests has been replaced with ed(1), the sed wrapper can be retired. There are existing tests which use sed, but they're modifying a stream, rather than using -i, so there's no portability concerns.

When looking for a suitable date wrapper, don't use the platform name, and rather use date(1)'s own parsing to choose BSD date, or gdate (GNU). Additionally, only declare the date() wrapper once. From Christian "naddy" Weisgerber

ok stsp

Check if date(1) supports -r and fallback to gdate if that fails, via a shell-wrapper. From Christian "naddy" Weisgerber.

GNU Coreutils is wrapped for some commands which don't offer a portable means to run those commands on non-BSD systems. ln(1) was being used with a non-portable option (-h) which has now been replaced upstream. Therefore, we can remove this wrapper. Patch from Christian "naddy" Weisgerber.

ok stsp

This allows to run only the test cases named on the command line.

ok jamsek

otherwise most test fails because of double "/" ending in paths and tricking some check into thinking that's failed. ok jamsek stsp

fixes test_rebase_no_author_info when a valid ~/.gitconfig is found. ok stsp@

When making sed's command-line flags portable, don't split on $@ before parsing it, as this won't preserve the original string which needs passing to sed unmodified if it's never had to be changed.

When massaging sed command strings, to handle the differences between "-i ''" not meaning the same thing on non-BSD systems, this previously used a bashism to determine the positional arguments. Instead, defer to using a loop which doesn't rely on bashisms.

Remove erroneously-placed backticks when describing a lack of GNU coreutils.

Don't rely on $OSTYPE being available, which it isn't in strict POSIX mode. Ubuntu's sh is often alised to dash, and hence this fails. Instead, default to using the output from 'uname'.

When running tests on non-Linux systems, it was just assumed that GNU coreutils had been installed. This meant that the wrapper commands for date/ln on those systems always had gdate and gln installed. Rather than assume that, check to see if the GNU coreutils versions as present, and if not echo a warning.

PLATFORM is populated when running via `./configure && make` but this isn't guaranteed if an individual test is run directly, such as `regress/cmdline/tag.sh`. In such cases, PLATFORM will be empty, but we still want to use it. Since we test for non-linux values, only set PLATFORM if we're running on Linu so that the correct commands are used.

This brings the rest of the regression test scripts in line with patch.sh.

GNU sed doesn't like "-i ''" which on BSD means to not keep backups around. Scan for "-i ''" and replace with just "-i" which will use the filename given to sed just fine.

Ubuntu's default shell is dash, when using /bin/sh. The portable nature of got is such that "$OSTYPE" to determine the host type (linux, bsd, etc.) is bash-specific and is not part of POSIX. autotools already provide a mechanism for determining the underlying platform type, and PLATFORM is already a subst value. Therefore, let the -portable parts of the regress test-suite use $PLATFORM to look at the underlying OS type. The tests themselves already expect /bin/sh to be POSIX-compliant in all other areas, so there should be no need to change the #! lines.