[2013-04-02 15:36] Sam Watkins <s...@nipl.net> > I reckon by far the most common thing people do with sed is to replace > one (sort of) string with another. But even this "core sed business" > is ridiculous with sed. > > e.g. A=/usr/bin B=/bin sed 's/$A/$B/' <in >out # won't work > > Changing delimiters doesn't help for the general case. I would have to > escape any input string before sed can handle it, which is more difficult > than writing a better substitute process. > > sed is over-complex and so it's stupid even at its simplest core business. > > I would like rather: > > sub "$A" "$B" <in >out > > etc. More binaries perhaps, but at least they are sane.
Funny to see this proposal. Never heard of gres(1), huh? [0][1] It was a ripped-out-of-ed tool like grep(1). The name stands for ``g/re/s///'' as grep stands for ``g/re/p''. It did exactly what you want sub(1) to do. Still at Bell Labs times gres died out when sed(1) became available. ;-) Seemingly, those that shaped Unix and its philosophy had no objections to its death. I can't recall from mind where I've read this story, possible in ``Software Tools'' or in ``The UNIX Programming Environment''. meillo [0] http://www.mkssoftware.com/docs/man1/gres.1.asp [1] http://docstore.mik.ua/orelly/unix/sedawk/ch03_02.htm#AUTOID-2409
