sed: do not ignore 'g' modifier when match starts with ^

It is perfectly valid to start a regex with ^ and have other patterns
with \| that can match more than once, e.g. the following example
should print ca, as illustrated with gnu sed:
$ echo 'abca' | sed -e 's/^a\|b//g'
ca

busybox before patch:
$ echo 'abca' | busybox sed -e 's/^a\|b//g'
bca

busybox after patch:
$ echo 'abca' | ./busybox sed -e 's/^a\|b//g'
ca

regcomp handles ^ perfectly well as illustrated with the second 'a' that
did not match in the example, we ca leave the non-repeating to it if
appropriate.
The check had been added before using regcomp and was required at the
time (f36635cec6da) but no longer makes sense now.

(tested with glibc and musl libc)

function                                             old     new   delta
add_cmd                                             1189    1176     -13

Signed-off-by: Dominique Martinet <asmadeus@codewreck.org>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

2 files changed, 7 insertions, 2 deletions

diff --git a/editors/sed.c b/editors/sed.c
index a6845a979..e8c82ac63 100644
--- a/editors/sed.c
+++ b/editors/sed.c
@@ -435,8 +435,7 @@ static int parse_subst_cmd(sed_cmd_t *sed_cmd, const char *substr)
 		switch (substr[idx]) {
 		/* Replace all occurrences */
 		case 'g':
-			if (match[0] != '^')
-				sed_cmd->which_match = 0;
+			sed_cmd->which_match = 0;
 			break;
 		/* Print pattern space */
 		case 'p':
diff --git a/testsuite/sed.tests b/testsuite/sed.tests
index 67ff87e93..2b78c9b12 100755
--- a/testsuite/sed.tests
+++ b/testsuite/sed.tests
@@ -399,6 +399,12 @@ testing "sed uses previous regexp" \
 	"" \
 	"q\nw\ne\nr\n"
 
+testing "sed ^ OR not^" \
+	"sed -e 's/^a\|b//g'" \
+	"ca\n" \
+	"" \
+	"abca\n"
+
 # testing "description" "commands" "result" "infile" "stdin"
 
 exit $FAILCOUNT