Re: [E-devel] my bad idea

[<Val...@vt...>]

On Tue, 22 Oct 2002 18:01:00 EDT, Najati Imam <nr...@uk...>  said:

> Yeah, its the seperation between the shell and the term that make this 
> most difficult, I think. That seperation is also one of the main 
> benefits of the current system. Hacking all shells with the right hooks 
> to provide all the info we need and then making all the changes to any 
> term would be one sizable undertaking. Oh well :-(

Hmm.. see "Cycle of Reincarnation" in the Jargon File. ;)

Bit of history - under Multics, the "shell" would only read commands, and did
NOT do globbing - that was done *in the program*.  In fact, *reading* the
command string was done by the program.  This allowed coolness like being
able to type "ftp foo*", and foo* would be read/globbed by the FTP command,
which could then provide "smart" completion and globbing (so 'foo<tab>' could
be expanded as a hostname rather than a filename, etc.  The "copy" command
could apply the knowledge that a given parameter *had* to be a directory rather
than a flat file, etc etc.

The major reasons this wasn't implemented in Unix shells are because you end
up with a lot of globbing/prompting code in every binary (and this was before
shared library support happened), and the fact it makes command pipelining
incredibly difficult to do - is that | a inter-command pipe char, or is it
part of a  tr '|' ';' sequence?  It's hard to figure out when the tr command
is reading/parsing the command string already, and may not want to hand the |
and ; back to the shell.. ;)
-- 
				Valdis Kletnieks
				Computer Systems Senior Engineer
				Virginia Tech