Re: ED on austingroupbugs

[Andrew L. Moore]

For a given pattern delimiter, d, POSIX specifies that the delimiter can 
be included in the pattern if it is backslash-escaped (i.e., \d).  So in 
the case d == '\\', the ed command `s\\\' should be equivalent to 
`s/\\//p'.  My current implementation treated this as `s///' instead.
The following commit fixes this (tests needed):
https://github.com/slewsys/ed/commit/cd2c4db24927df7b0009f5a9f1d4338d72cdfd11

GNU ed has an underlying bug:  the command `s/pattern` is incorrectly 
flagged as an error due to a missing delimiter, whereas it has always 
been equivalent to `s/pattern//p'. Though not spelled out explicitly by 
POSIX, it is traditional behavior - and consistent with the line 
addressing expression `/pattern' (without terminating delimiter) which 
GNU ed does correctly handle.
-AM

On 4/23/23 10:01, Andrew L. Moore wrote:
> On 4/22/23 02:59, vico@tuta.io wrote:
> > Maybe you haven't read it and you'll be interested:
> > https://austingroupbugs.net/view.php?id=1662
>
> POSIX does not allow a backslash delimiter in ex, but does not disallow 
> it for ed.  And indeed, The Open Group testsuite for ed does not 
> complain if a backslash delimiter is supported.
>
> I'm sharing this with GNU ed maintainer, Antonio Diaz, since a single 
> backslash (\) is permitted as a (admittedly weird) pattern delimiter in 
> GNU, macOS and *BSD ed as well, since they all share the same codebase.
>
> Thank you for the heads up!
> -AM