Re: [PATCH 1/3] Properly expand @configure_input@ in config.status.

[Paul Eggert]

Paolo Bonzini <address@hidden> writes:

>> 3) echo 'abc' | sed 's1a\(b\)c1A\1C1g'
>> a) A1C
>> GNU
>> b) AbC
>> AIX, Darwin, FreeBSD, HP-UX, IRIX, NetBSD, OpenBSD, Solaris, Tru64
>
> Hmm, so on the RHS on the other hand the string is passed as-is.  It
> is interesting that for digits, GNU sed is the only one to follow
> Paul's interpretation ("\x produces literal character x") of
> SUSv3. :-)

I see what you're saying about the part I quoted being too strong.
But there's another part of the standard that suggests that AIX,
Darwin, etc., are conforming here, and that GNU is not conforming.
The standard also says:

   The special meaning of '&' in this context [i.e., in the
   replacement string] can be suppressed by preceding it by a
   backslash. The characters "\n", where n is a digit, shall be
   replaced by the text matched by the corresponding backreference
   expression. The special meaning of "\n" where n is a digit in this
   context, can be suppressed by preceding it by a backslash.

So, in 's&foo&\&&', the backslash suppresses the special meaning of
'&'.  But in 's1a\(b\)c1A\1C1g', the '\1' retains its special meaning
because it is not preceded by a backslash.  Admittedly this language
is not ironclad, but I think the intent is clear.

Anyway, regardless of the standards-lawyerese, it does appear that GNU
sed is incompatible with traditional sed in its behavior with the rhs;
I don't see the advantage of the incompatibility.