sed command in busybox

[John Kelly]

On Sun, 20 Nov 2005 13:08:26 +1100, Glenn L McGrath
<bug1 at iinet.net.au> wrote:

>> You can search the comp.unix.shell archives for:
 
>>     retain_quoted v0.004

>Found it, i got an error trying to run it under busybox sed, "bad
>option in substitution expression", it has commands like "s/foo/bar/
>#baz" i think technically it should have a semicolon in there,
>i.e. "s/foo/bar/;#baz"

>"Command verbs other than {, a, b, c, i, r, t, w, :, and # can be
>followed by a semicolon, optional <blank>s, and another command verb."

Maybe that needs cleaning up.  But since it works with GNU, I have
little incentive.


>I got it to run, and i see its output is different than GNU sed, I will
>see if i can derive a simpler test from it.

It's different because busybox sed does not recognize the \n escape
within the bracket expressions, and that causes the matches to fail.

I discovered that by changing all the occurrences of \n to an unusual
character like @ and omitting that character from my test input.  And
since the G commands put newlines in the pattern space, I also had to
add s commands to replace the newlines with @, after each G command.

But once I solved that problem, further testing with various inputs
showed that some inputs caused busybox sed to go hawire and produce
crazy output.  At that point, I concluded busybox sed is not up to the
rigors of my script.  Admittedly, it's not your typical sed script.
The long regex near the top:

    ^(([^"'\n]*([^\"'\n]|\\"|\\'))*)("|')(.*)


was difficult to derive and is difficult to understand, unless you
convert it to the symbolic form I used during development and testing

    ^(([^QqN]*([^bQqN]|\bQ|\bq))*)(Q|q)(.*)


where you can feed it symbolic input from the command line to see how
it works.  That helped me appreciate the programmers who wrote the GNU
sed C code which can properly handle it.