Re: [PATCH] Completion: audit 'compset -P' calls to use shortest match where applicable, plus random drive-by tweaks.

[Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx>]

I'm going to leave it to others to say whether -P 1 is really correct
in all of these cases.  I have concerns about _lzop and _tail because
the pattern semantics have actually been changed but I haven't looked
closely to see if that matters.

On Sep 28,  5:46pm, Daniel Shahaf wrote:
} Subject: [PATCH] Completion: audit 'compset -P' calls to use shortest matc
}
} I'm not sure what to do about the handling of "_path_files -P foo": that -P
} option documents its argument as a string but interprets it as a pattern.
} Should we update the documentation to match the implementation, or vice-versa?

That's this hunk:

}  if (( $#pfx )); then
} +  # ### Is it correct to interpret -P as a (greedy) pattern here?
}    compset -P "$pfx[2]" || pfxsfx=( "$pfx[@]" "$pfxsfx[@]" )
}  fi

I think it would be OK / correct to use ${(b)pfx[2]} there.  Presumably if
$pfx[2] doesn't literally match the filename, we're falling back to the
pfxsfx assignment to resolve it later.

Here in _cpio:

} -  if compset -P '*:'; then
} -    # TODO: doesn't need to be rsh.

The TODO comment may have been meant to indicate there should be a zstyle
to choose what remote shell program to use?  Dumping the comment and
forcing ssh might not be what was intended, but maybe it's OK, as it's
consistent with what we do for remote shell pretty much everywhere else.

In _rdesktop:

}      redir="${PREFIX%%:*}"
} -    if compset -P '*='; then
} +    if compset -P 1 '*='; then
}        curcontext="${curcontext%:*}:$redir"
} -      compset -P '*='
}        case $redir in

The old code does look suspicious, but does anyone know if it was doing
two "compset"s intentionally?