Hi Martijn,
i think your patch does achieve the desired effect, but i'd prefer
the following one instead. There is really no need to malloc(3) an
empty string, only to later translate and free it.
My version does not seems very dangerous. All it changes is that,
if "b" or "t" is followed by optional whitespace, and then immediately
by a semicolon, branching is set up to the end of the script -
in exactly the same way as for "b" at the end of an input line -
instead of to a label with an empty name, which never exists.
Nothing else changes, in particular not parsing of subsequent input.
The test suite still succeeds. Additional tests:
$ echo "A\nB" | sed '1b;='
A
2
B
$ echo "A\nB" | sed 's/A/C/;t;='
C
2
B
$ echo "A\nB" | sed '1bL;=;:L'
A
2
B
$ echo "1b\n=" > tmp.sed; echo "A\nB" | sed -f tmp.sed
A
2
B
Yours,
Ingo
Index: compile.c
===================================================================
RCS file: /cvs/src/usr.bin/sed/compile.c,v
retrieving revision 1.49
diff -u -r1.49 compile.c
--- compile.c 14 Aug 2018 18:10:09 -0000 1.49
+++ compile.c 6 Dec 2018 19:07:31 -0000
@@ -284,7 +284,7 @@
case BRANCH: /* b t */
p++;
EATSPACE();
- if (*p == '\0')
+ if (*p == '\0' || *p == ';')
cmd->t = NULL;
else
cmd->t = duptoeol(p, "branch", &p);
> Index: compile.c
> ===================================================================
> RCS file: /cvs/src/usr.bin/sed/compile.c,v
> retrieving revision 1.49
> diff -u -p -r1.49 compile.c
> --- compile.c 14 Aug 2018 18:10:09 -0000 1.49
> +++ compile.c 5 Dec 2018 06:02:15 -0000
> @@ -797,7 +797,8 @@ fixuplabel(struct s_command *cp, struct
> cp->u.c = NULL;
> break;
> }
> - if ((cp->u.c = findlabel(cp->t)) == NULL)
> + if ((cp->u.c = findlabel(cp->t)) == NULL &&
> + *cp->t != '\0')
> error(COMPILE, "undefined label '%s'", cp->t);
> free(cp->t);
> break;
>
