git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH] t4014: shell portability fix

From
Jeff King <peff@peff.net>
Date
Jun 1, 2016, 05:33 UTC
Message-ID
<20160601053325.GA20797@sigill.intra.peff.net>
In-Reply-To
<20160601034413.GC5411@sigill.intra.peff.net>
On Tue, May 31, 2016 at 11:44:13PM -0400, Jeff King wrote:
> So this is gross, but I think it actually _is_ portable, with the
> exception of the "is it exported" check.

Hmm. So after thinking on this, I realized we don't have to do the clean-up ourselves at all, if we simply operate in a subshell. That means that any shell functions we call couldn't mutate the state, but that's probably an acceptable compromise, if the goal is to behave like env except for being able to call shell functions.

Here is the "final" version of the more complicated scheme I came up with. That I think should be fairly portable, but the subshell thing is probably way less gross.

-- >8 --
test_var_is_set () {
	eval "test -n \"\${$1+set}\""
}
test_var_is_exported () {
	sh -c "test -n \"\${$1+set}\""
}
# set var named by $1 to contents of $2
test_set_var () {
	eval "$1=\$2"
}
# set var named by $1 to contents of var named by $2
test_copy_var () {
	eval "test_set_var \$1 \"\$$2\""
}
# just a syntactic convenience
add_to () {
	eval "$1=\"\$$1
		\$2\""
}
test_env () {
	test_env_restore_=
	while test $# -gt 0
	do
		case "$1" in
		*=*)
			# this whole thing is not safe when the var name has
			# spaces or other meta-characters, but since the names
			# all come from our test scripts, that should be OK
			test_env_var_=${1%%=*}
			test_env_orig_=test_env_orig_$test_env_var_
			test_copy_var $test_env_orig_ $test_env_var_
			test_env_val_=${1#*=}
			shift
			# unset value and clear export flag...
			add_to test_env_restore_ "unset $test_env_var_"
			# ...and then restore what was there, if anything
			if test_var_is_set "$test_env_var_"
			then
				add_to test_env_restore_ \
					"test_copy_var $test_env_var_ $test_env_orig_"
				if test_var_is_exported "$test_env_var_"
				then
					add_to test_env_restore_ \
						"export $test_env_var_"
				fi
			fi
			# and then clean up our temp variable
			add_to test_env_restore_ "unset $test_env_orig_"
			test_set_var $test_env_var_ "$test_env_val_"
			eval "export $test_env_var_"
			;;
		*)
			"$@"
			test_env_ret_=$?
			eval "$test_env_restore_"
			return $test_env_ret_
			;;
		esac
	done
}
# simple exercise
show () {
	echo >&2 "$1: here=$myvar ${myvar+(set)} sub=$(sh -c 'echo "$myvar ${myvar+(set)}"')"
}
doit () {
	echo "==> $1"
	show before
	test_env myvar="temporary $myvar" show during
	show after
}

unset myvar doit unset myvar="horrible \"\\' mess" doit with-value export myvar doit exported-with-value myvar= export myvar doit exported-but-blank

Previous: Jeff KingNext: Jeff King
Message 8 of 20 in “t4014: shell portability fix”
  1. t4014: shell portability fixJunio C Hamano, May 31, 2016
  2. Jeff KingMay 31, 2016
  3. Eric SunshineJun 1, 2016
  4. Jeff KingJun 1, 2016
  5. Jeff KingJun 1, 2016
  6. Jeff KingJun 1, 2016
  7. Jeff KingJun 1, 2016
  8. Jeff KingJun 1, 2016
  9. Jeff KingJun 1, 2016
  10. Junio C HamanoJun 1, 2016
  11. Junio C HamanoJun 1, 2016
  12. Jeff KingJun 1, 2016
  13. Junio C HamanoJun 1, 2016
  14. Jeff KingJun 1, 2016
  15. Junio C HamanoJun 1, 2016
  16. Jeff KingJun 1, 2016
  17. Johannes SixtJun 1, 2016
  18. Jeff KingJun 1, 2016
  19. Eric SunshineJun 1, 2016
  20. Ramsay JonesJun 1, 2016

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.