smart_add uses _get_inventory, _write_inventory, but should not

See also bug 146176

Also relates to calls to set_state_from_inventory

Confirmed that this is still the case in bzr.dev 5807. I'm guessing it shouldn't be marked In Progress. So I'm re-marking it as assigned to ~canonical-bazaar but not on Martin's active plate. Still considered a Medium priority and tech-debt sort of bug.

I might have a stab at this. If smart_add() used the WorkingTree API that would mean I didn't have to implement a custom smart_add() for bzr-svn, bzr-hg and bzr-git.

I would be very happy if you get it going; it will probably speed it
up, make it more readable, and also let it work with foreign branches.
 The add code at the moment does a lot of manipulations on the
inventory after reading it in, which makes it easy to keep track of
things like adding parents before children, but it does also mean it's
very tied to having mutable inventories.

Perhaps we can pair on this at the sprint if you don't solve it beforehand.

Martin

On Mon, 2011-05-02 at 08:29 +0000, Martin Pool wrote:
> I would be very happy if you get it going; it will probably speed it
> up, make it more readable, and also let it work with foreign branches.
> The add code at the moment does a lot of manipulations on the
> inventory after reading it in, which makes it easy to keep track of
> things like adding parents before children, but it does also mean it's
> very tied to having mutable inventories.
>
> Perhaps we can pair on this at the sprint if you don't solve it
> beforehand.
I've done some investigation into it; mostly trying to get it to use
MutableTree.add(), but it seems like apply_inventory_delta is actually
the right way to go as some existing inventory entries might actually
have to change kind (files being replaced by directories, etc).

It'd be nice to pair on this, it should be reasonably easy.

Cheers,

Jelmer