[sheepdog] [PATCH v2] dog: permit two or more depth of subcommands

[Liu Yuan]

On Wed, Nov 13, 2013 at 03:40:07PM +0900, Hitoshi Mitake wrote:
> At Wed, 13 Nov 2013 14:30:31 +0800,
> Liu Yuan wrote:
> > 
> > On Wed, Nov 13, 2013 at 03:17:48PM +0900, Hitoshi Mitake wrote:
> > > At Wed, 13 Nov 2013 14:15:19 +0800,
> > > Liu Yuan wrote:
> > > > 
> > > > On Wed, Nov 13, 2013 at 02:53:27PM +0900, Hitoshi Mitake wrote:
> > > > > At Wed, 13 Nov 2013 10:53:15 +0800,
> > > > > Liu Yuan wrote:
> > > > > > 
> > > > > > On Wed, Nov 13, 2013 at 12:22:31AM +0900, Hitoshi Mitake wrote:
> > > > > > > At Tue, 12 Nov 2013 15:38:17 +0800,
> > > > > > > Liu Yuan wrote:
> > > > > > > > 
> > > > > > > > On Tue, Nov 12, 2013 at 04:33:49PM +0900, Hitoshi Mitake wrote:
> > > > > > > > > At Mon, 11 Nov 2013 21:24:47 +0800,
> > > > > > > > > Liu Yuan wrote:
> > > > > > > > > > 
> > > > > > > > > > On Mon, Nov 11, 2013 at 10:16:57PM +0900, Hitoshi Mitake wrote:
> > > > > > > > > > > From: Hitoshi Mitake <mitake.hitoshi at lab.ntt.co.jp>
> > > > > > > > > > > 
> > > > > > > > > > > Current dog command dies when user invokes "dog node log level set"
> > > > > > > > > > > without any arguments because of its subcommand design. This patch
> > > > > > > > > > > removes this limitation.
> > > > > > > > > > > 
> > > > > > > > > > > Signed-off-by: Hitoshi Mitake <mitake.hitoshi at lab.ntt.co.jp>
> > > > > > > > > > > ---
> > > > > > > > > > > 
> > > > > > > > > > > v2: define do_generic_subcommand() as a macro for a common case (depth == 0)
> > > > > > > > > > > 
> > > > > > > > > > >  dog/common.c | 5 +++--
> > > > > > > > > > >  dog/dog.h    | 5 ++++-
> > > > > > > > > > >  dog/node.c   | 3 ++-
> > > > > > > > > > >  3 files changed, 9 insertions(+), 4 deletions(-)
> > > > > > > > > > > 
> > > > > > > > > > > diff --git a/dog/common.c b/dog/common.c
> > > > > > > > > > > index 4480b99..90b43ad 100644
> > > > > > > > > > > --- a/dog/common.c
> > > > > > > > > > > +++ b/dog/common.c
> > > > > > > > > > > @@ -224,7 +224,8 @@ int send_light_req(const struct node_id *nid, struct sd_req *hdr)
> > > > > > > > > > >  	return 0;
> > > > > > > > > > >  }
> > > > > > > > > > >  
> > > > > > > > > > > -int do_generic_subcommand(struct subcommand *sub, int argc, char **argv)
> > > > > > > > > > > +int do_generic_subcommand(struct subcommand *sub, int depth,
> > > > > > > > > > > +			  int argc, char **argv)
> > > > > > > > > > >  {
> > > > > > > > > > 
> > > > > > > > > > Why we need 'int depth'? I don't think we need depth since we don't know where
> > > > > > > > > > is the horizon. What I meant in my last reply is just make 5 as a bigger one, like
> > > > > > > > > > 
> > > > > > > > > > #define DOG_MAX_PARAMETERS 10
> > > > > > > > > 
> > > > > > > > > What does the word "horizon" mean in this context? I think we need the
> > > > > > > > > depth variable for detecting the beginning of arguments for
> > > > > > > > > subcommands.
> > > > > > > > >
> > > > > > > > 
> > > > > > > > I meant 'beginning of arguements for subcommands'. But boundary of subcommands
> > > > > > > > depends on context. E.g, 'dog vdi log level list', vdi is subcommand of dog,
> > > > > > > > log is subcommand of vdi and so on. So I think we only need a marco to define
> > > > > > > > the max allowed subcommdans. An extra parameter isn't necessary to me.
> > > > > > > 
> > > > > > > Do you mean we should replace the line:
> > > > > > > 	if (flags & CMD_NEED_ARG && argc < 5)
> > > > > > > with a line like this:
> > > > > > > 	if (flags & CMD_NEED_ARG && argc < DOG_MAX_PARAMETERS)
> > > > > > > ?
> > > > > > 
> > > > > > Yes
> > > > > > 
> > > > > > > If we do so, ordinal subcommands cannot work well.
> > > > > > 
> > > > > > So what is the problem?
> > > > > 
> > > > > If we changes do_generic_subcommand() in the above way, existing
> > > > > subcommands like "dog vdi cache flush <vdi>" cannot work well.
> > > > 
> > > > What exactly do you mean by 'not work well'?
> > > 
> > > Produces an error message like below even if the argument <vdiname> is passed:
> > > 
> > > Usage: dog vdi cache {flush|delete|info|purge} [-s snapshot] [-a address] [-p port] [-h] <vdiname>
> > > Available subcommands:
> > >   flush                   flush the cache of the vdi specified.
> > >   delete                  delete the cache of the vdi specified in all nodes.
> > >   info                    show usage of the cache
> > >   purge                   purge the cache of all vdi (no flush)
> > > Options:
> > >   -s, --snapshot          specify a snapshot id or tag name
> > >   -a, --address           specify the daemon address (default: localhost)
> > >   -p, --port              specify the daemon port
> > >   -h, --help              display this help and exit
> > > 
> > 
> > Okay, I take a look at your problem. I think you should fix node_log_level_set()
> > which cause the segfault instead of working around it by asking do_generic_subcommand
> > to check the parameters.
> > 
> > IMO, you check null parameter case inside node_log_level_set().
> 
> I don't think that node_log_level_set() should do the checking. If the
> function do so, new subcommands implemented in the future should do it
> and this policy produces lots of duplication.
> 
> In addition, if do_generic_subcommand() don't do the check, the
> architecture of nested subcommands shouldn't be allowed. We should
> remove the member *sub from struct subcommand.
> 

For some sense I agree with you that we need rework subcommand mechanism but
your patch is more a workaround to me than a complete fix.

I think we can do it more transparently and extra depth parameter is unnecessary:
 - since do_generic_subcommand is called iteratively, probably we can declare a
   static variable that indicate the depth and check against it?

Thanks
Yuan