Just as an example of extending sed, I introduced "sed -i".

Yes, it doesn't require much maintenance, but sometimes you can be surprised. I started maintaining GNU grep 3 years ago because it was in a really sorry state. Some parts were almost rewritten to make it faster and more correct.

I never use sed -i because I can't rely on that being a standard feature as I move from system to system. With a utility as basic as sed that is found on almost every UNIX (and hence often used in a system's startup or configuration scripts, etc.), I want my own sed scripts to work consistently across all systems I might use.

If I really want to create temp files I can use ed or use sed with shell redirection. Were it for some reason a requirement, I can avoid temp files, at least for the substitution (s) command, as follows:

  sed -a '"$1"';H;$!d;g;w'$2
  where $1 is some sed s commands and $2 is a file

This is not "perfect" as there will blank lines, but it does the job without temp files (if that were really a concern).

If GNU grep was in need of repair and you fixed it, then I thank you. But I'm not clear on how that justifies any "extensions" to GNU grep.

Maintenance, at least to me, means fixing things, not extending them and adding more complexity and things that can potentially break or create incompatibilities across different UNIX's.

Both GNU sed and BSD sed have -i, what other kinds of sed are you using?

/usr/bin/sed on Solaris 10 and earlier does not support -i. I don't know if OpenIndiana / Solaris 11's sed support -i, but OmniOS uses gnu sed by default, so OI might also.

As I recall, AIX's default sed also does not support -i, but I no longer have access to AIX systems to test against.

Honestly, sed -i is so useful that I'd file a feature request if a sed didn't support it.

Its easier to just not use sed and use perl -pie instead. Ironically its much more portable to non gnu systems.