Re: [Gnu-arch-users] Pipes?

[Tom Lord]

    > From: "Pierce T.Wetter III" <address@hidden>

    > > But xargs is designed to solve the common case.  The right thing here
    > > is to make sure that xargs works.

    >   I can easily think of cases where someone would want to pass more 
    > arguments
    > to tla then can fig in a command line...


Ok, well, fair enough.   Especially when you're talking about `add'
`delete' and all the other subcommands in that family?  You're right.

I confess that I'm speaking in more general terms than you -- yup --
those commands sure would be improved by tar-like --list options.

However, the "where does unescaping take place" question -- my general
principles analysis is still right.

    > > There's a choice about how to make xargs work.   One idea is that
    > > xargs could itself do unescaping.   It's going to pass arguments in an
    > > argv so there would be no ambiguity resulting from unescaped
    > > filenames.

    >   Thing is, xargs can do some unescaping already. It converts \\ to \ for
    > instance. It just seems to be doing different unescaping then the
    > pika escaping...

Right.  Pika escaping is new and nothing outside of hackerlab and tla
(where it isn't even merged yet) supports it.   But none of the extant
and more widely deployed alternatives really do as well.   So, it's a
good thing to start promoting. 


    > > The other idea is that all arch commands should unescape their command
    > > line arguments which are filenames.

    > > Subcommand expansion (backticks and $(...)) complicates things.


    >   Actually, I was thinking it might be a global option. Early on in 
    > main(), it would
    > look for a --pipe option, and if it found one, append stdin to argv 
    > before doing anything else.

Nah.  You want --list or --file for arguments of a particular type.

What about a program that needs to read _two_ lists of arguments, for
example?

-t