You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@datasketches.apache.org by Jon Malkin <jm...@apache.org> on 2020/12/01 19:04:26 UTC
[DISCUSS] incomplete API change in previous version
We dropped serialization of unions from (most) c++ sketches with the move
to 2.0, but we have a JIRA from a while ago that notes we forgot to remove
deserialize() form hll_union.
It's easy enough to remove the methods. But it's actually possible to pass
in a regular HLL sketch to deserialize(), since the union just serialized
its internal gadget (in this case I believe it's _always_ a proper HLL
sketch; that's not true for other union gadgets).
So would we need to move to 3.0 just for removing the deserialize()
methods? I think strict semantic versioning would say yes? We haven't had
it marked as officially deprecated, although the JIRA suggests it was
already kinda of obvious.
I removed the code locally and didn't even need to change unit tests or the
python wrapper. So I can do the PR, but I'm not sure that we want to jump
to 3.0 just for this change.
Thoughts?
jon
Re: [DISCUSS] incomplete API change in previous version
Posted by leerho <le...@gmail.com>.
This sounds more like a bug fix so moving to 3.0 shouldn't be necessary.
Lee.
On Tue, Dec 1, 2020 at 11:04 AM Jon Malkin <jm...@apache.org> wrote:
> We dropped serialization of unions from (most) c++ sketches with the move
> to 2.0, but we have a JIRA from a while ago that notes we forgot to remove
> deserialize() form hll_union.
>
> It's easy enough to remove the methods. But it's actually possible to pass
> in a regular HLL sketch to deserialize(), since the union just serialized
> its internal gadget (in this case I believe it's _always_ a proper HLL
> sketch; that's not true for other union gadgets).
>
> So would we need to move to 3.0 just for removing the deserialize()
> methods? I think strict semantic versioning would say yes? We haven't had
> it marked as officially deprecated, although the JIRA suggests it was
> already kinda of obvious.
>
> I removed the code locally and didn't even need to change unit tests or
> the python wrapper. So I can do the PR, but I'm not sure that we want to
> jump to 3.0 just for this change.
>
> Thoughts?
>
> jon
>