If you were to have two implementations you would want to annotate the old one
as deprecated immediately. I don't think you would want to commit to supporting
it for more than a couple of months. How many users of the Scala MXNet
interface are there in any case? And do they really view the
The namespace change is the first thing that's done for most projects that
come to apache incubation
How many production deployments of MXNet Scala API are out there --- 3 ? 2
? 1.7643 ?
I would think its barely a handful of them.
Agree with Christopher Barber that MXNEt jumped the gun with
I'm not taking a side here, but just please consider that if you have two
separate implementations for awhile, the newer one will start to diverge
and over time, it will become harder and harder for the user to port his
code. You may find yourself supporting the old code for much longer than
you
How many people do we estimate are currently using the Scala interface?
Probably the actual blast radius should be considered. If it is very
small, then we can probably have more "wiggle room", so to speak.
On Tue, Mar 13, 2018 at 9:00 AM, Chernov, Anton wrote:
> I see (2)
that might be the last thing we want to do, i.e. keeping some code just for
the namespace change,
I am open to have such a PR with the assumption that
1. changing namespace is scheduled to be finished at 1.x versions
(alternatives might be when we graduate as TLP)
2. breaking backward
I see (2) being the appropriate way to go. Scala code is copied to a new
namespace and all the old code gets a deprecation mark which means it's only
supported for backwards compatibility and will not be modified unless there
will be an urgent fix.
On 13.03.18, 16:52, "kellen sunderland"
Maintaining backwards compatibility never results in the prettiest code,
but it seems pretty desirable here. There are relatively few files here,
so I agree there's some risk but I don't think it would take too much
time. Feel free to suggest alternatives Christopher.
On Tue, Mar 13, 2018 at
That sounds like a lot of work and it would be easy to get wrong if it is even
feasible.
On 3/13/18, 11:51 AM, "kellen sunderland" wrote:
I don't know about aliasing a namespace in Scala, but I wonder how hard it
would be to either (1) provide a fascade
re Chris: I do not have any good idea about this.
On Tue, Mar 13, 2018 at 8:13 AM, Chris Olivier
wrote:
> is it possible to somehow alias a namespace in scala
> in order to maintain backwards compatibility?
>
> On Tue, Mar 13, 2018 at 7:21 AM Nan Zhu
is it possible to somehow alias a namespace in scala
in order to maintain backwards compatibility?
On Tue, Mar 13, 2018 at 7:21 AM Nan Zhu wrote:
> +1
>
> and additional suggestion is do it ASAP
>
> On Mon, Mar 12, 2018 at 11:21 PM, Chris Olivier
+1
and additional suggestion is do it ASAP
On Mon, Mar 12, 2018 at 11:21 PM, Chris Olivier
wrote:
> not sure I understand. How could changing a java namespace (effectively
> moving the files to a different location as well as changing the package
> names) be
-1, I cannot think about a significant benefit comparing to the headache of
the user
still take Spark as an example
if PySpark is in a separated repo, say now it's 1.0, can you tell me
whether this 1.0 support Arrow integration without going to their website
and carefully check which version of
I suggest the vote should call out if the change is breaking backward
compatibility or not.
I looked through the scala name changing thread and don't see justification
for a backward incompatible change.
I do agree it would be good to change the name space, but have not seen a
reason why the
13 matches
Mail list logo