Re: [Bug-ed] Error in POSIX ed address?

[Bob Proulx]

Hi Antonio,

Not seeing anyone else respond I thought I would make a comment.

Antonio Diaz Diaz wrote:
> I was fixing the ',,' address bug reported by Matthieu Felix when I
> found what I think is an error in the POSIX page for ed:

I think you did too.

> http://pubs.opengroup.org/onlinepubs/9699919799/utilities/ed.html
> 
> Address  Addr1  Addr2
> 7,+       7      8
> 7;+       7      8
>
> The first row should be equivalent to '7,.+', not to '7,8'. Or is this one
> of those historical quirks of ed?

The 7,+ address range is a state dependent range AFAIK.  It depends
upon the previous value state of dot.  If dot is 7 then the address
range is 7,8 because + by itself is the same as .+1 setting dot to 8.
But if dot is, say, 3, then 7,+ is the same as 7,4 which will print
line 4 due to the default action of using the last address if only one
is expected.

Therefore the Addr2 column for 7,+ should say ".+1".  Saying 8 there
is definitely not correct.

I know you know all of this following.  I am simply being verbose for
the email archive.

Using the Unix v7 ed sources to verify the traditional Unix ed
behavior let me poke around the corners a little bit.

$ seq 10 | sed 's/^/line /' > numbers.txt
$ v7ed numbers.txt
71
.=
10
7,+
?

That error ('?') is expected because 7,11 is the same as 11 and that
is beyond the end of the file.

3
line 3
.=
3
7,+
line 4
.=
4

Seeing 7,+ is so odd because it depends upon the previous value of dot.

7
line 7
7,+
line 8
7,+
line 9
7,+
line 10
7,+
?

Again it is moving dot along each time.  Which I had not realized
before but I guess can be used like this contrived example.

/pattern2/
?pattern1/,++s/foo/bar/

The /pattern2/ will set dot and then the ?pattern1? will be addr1 and
++ will be the dot (from pattern2) plus two lines as a range for the
command.  To exercise it with the test file:

/line 7/
line 7
?line 3?,++p
line 3
line 4
line 5
line 6
line 7
line 8
line 9

However all of the above is the Unix v7 ed.  I don't have easy access
to a SysV version of ed to test the above corners on but I think it
likely that it behaves the same way for this.

Hope this helps!

Bob