On Tue, Dec 22, 2009 at 02:32:24AM +0000, Matthew Szudzik wrote:
> On Mon, Dec 21, 2009 at 08:37:06PM -0500, Hugo Villeneuve wrote:
> > OpenBSD's sed technicaly supports command concatation with ";" on
> > the command line but it breaks "branch" without label. Which correctly
I should have left more quoted text about tutorial from other sed
implementation.
I had been using gnu sed for a while and encountered this issue
going with OpenBSD.
>
> Read the caveat at the end of the sed man page:
>
> "The use of semicolons to separate multiple commands is not permitted
> for the following commands: a, b, c, i, r, t, w, :, and #."
I should read better. Since it's documented, it's not a bug.
Of course that sentence is false. ":" correctly interpret semicolon
as next command. "b" and "t" has partial support (only when the
label is not empty). It is true for the a c i r w #.
Is this one of those few case where I should obey the man page
rather than the code?
>
> Either use newlines instead of semicolons, or give each line of the
> script separately with the -e option. For example, instead of
>
> sed 's/a/b/ ; t ; s/c/d/'
>
> use
>
> sed -e 's/a/b/' -e 't' -e 's/c/d/'
It's a cute work around. My point is: if a label can't contain
semicolons, why support "tlabel;" and not "t;"? I don't understand
the logic of it.
Don't trust this diff but it make my few test works:
Index: compile.c
===================================================================
RCS file: /cvs/src/usr.bin/sed/compile.c,v
retrieving revision 1.28
diff -u -r1.28 compile.c
--- compile.c 16 Oct 2008 16:34:32 -0000 1.28
+++ compile.c 25 Dec 2009 17:46:49 -0000
@@ -283,7 +283,7 @@
case BRANCH: /* b t */
p++;
EATSPACE();
- if (*p == '\0')
+ if (*p == '\0' || *p == ';' )
cmd->t = NULL;
else
cmd->t = duptoeol(p, "branch", &p);
