
On 11/08/2017 05:26 PM, Eric Blake wrote:
On 11/08/2017 10:06 AM, Michal Privoznik wrote:
$ virsh complete some-command --arg partial<TAB><TAB>
This is excatly what my patches are doing. With one tiny difference. In fact bash script calls:
virsh complete "some-command --arg partial"
so that I can have cmdComplete which takes exactly one string argument.
Ouch. That difference matters, and I don't think you want to do that.
We don't want to reimplement shell parsing; if the user does:
virsh snapshot-create-as 'domain with space' 'name with space' 'description with space', then my approach feeds those three arguments as is (the virsh parser doesn't have to undo any shell quoting), while your approach HAS to over-quote the original command line, then reimplement shell quoting to remove those quotes in virsh, in order to reconstruct the original line in spite of being temporarily squashed through one argument.
Parsing of the string is then done within cmdComplete. I don't think that we have variable args in virsh for commands, do we?
Yes, we do. See 'virsh echo' for an example of its use.
So while implementing this I realized it will not fly. At ARGV level it is impossible to tell apart "--domain" and "--domain ". But the difference is important, because in the first case we want to offer list of all --options, while in the second case we want to call the --domain completer. Okay, this might not be the best example because there's no other option that shares the prefix, but for instance 'migrate' command has: --timeout --timeout-suspend --timeout-postcopy So if user types in: "virsh migrate --timeout" we have to bring up list: "--timeout" "--timeout-suspend" "--timeout-postcopy" and only after they type: "virsh migrate --timeout " we can call --timeout completer. Since it's impossible to differentiate these cases with VSH_OT_ARGV, I'm sticking with what I have now. Sure, it might has its own problems but those can be solved later. Michal