Re: [Bug-ed] Problems in POSIX description of ed reported

[Antonio Diaz Diaz]

Hi Andrew,

Sorry for the late answer. I was on vacation. :-)

Andrew L. Moore wrote:
> > > Your rationale for rejecting 0c is based on implementation details.
> > No, it is not.
>
> Antonio, to quote your rationale:
>
>      ... accepting 0 as a valid address for the c command does not
>      make sense because the c command executes a d   command ...

The c command operates on existing lines; it replaces them with some 
other thing. Removing the lines is its job. Executing a d command is not 
an implementation detail. It is part of the c command description:

http://pubs.opengroup.org/onlinepubs/9699919799/utilities/ed.html
"The c command shall delete the addressed lines, then accept input text 
that replaces these lines".

IMO, commands operating on existing lines should not accept an invalid 
address, neither 0 nor $+1.


> Now let me take the naive user's perspective: Since the command
> `0a' means "insert before the first line" then it seems reasonable to
> expect the command `0i' to also "insert before the first line".  And
> to be consistent with modal commands `a' and `i', address `0'
> should be valid against command `c' as well.

This is where the naive user's reasoning fails. The c command uses input 
mode to collect the text that will replace the deleted lines, but this 
does not make it similar to a or i with respect to line addressing. The 
c command operates on existing lines, while the a and i commands just 
insert text at some place, which may be "before the first line", 
"between two lines", or "after the last line".


> The naive user's argument leaves something to be desired, but I
> believe the POSIX rationale is that, when possible, give the user the
> benefit of the doubt.

I think it is important for a program to behave correctly. If quirks are 
allowed, the naive user will require the same quirks to be ported to all 
implementations.


> POSIX singles out command `c' because it's one of the three modal
> commands - along with `a' and `i'.

I do not consider c, a and i to be in the same category. The fact that 
the three use input mode has nothing to do with the addresses that are 
valid for each of them.


> For what it's worth, I don't believe that `0c' was valid in Ken
> Thompson's ed, but I don't have access to an old UNIX(tm) system at
> the moment to confirm that. If so, you could use that in your
> argument. But striving to do something reasonable before throwing an
> error is part of the Unix philosophy, as outlined, e.g., in "The UNIX
> Programming Environment" by Kernighan and Pike.

Well, I do not think that accepting 0c is something reasonable. IMO 0a 
should be used instead.


Best regards,
Antonio.