When to mark certain operator as @Stable is not clearly defined.

Can we define some criteria for deciding when to consider operator as
@Stable?

For example one criteria could be, if operator is running for >1 year in
production environment for some user. Can we come with some strategy like
this?
[It would be difficult for an open source project to track which user is
using which operators. So, above strategy may not work. ]

~ Yogi

On 14 December 2015 at 05:42, Timothy Farkas <[email protected]> wrote:

> +1
> On Dec 13, 2015 4:08 PM, "Chandni Singh" <[email protected]> wrote:
>
> > Hi,
> >
> > In Malhar there are  relatively smaller number of  operators that we use
> in
> > our demo applications, customer applications, POCs etc that are mature.
> >
> > The library is cluttered with operators especially in lib/util, lib/algo,
> > lib/math packages which can be cleaned up by either removing them or
> > improving them but that breaks semantic versioning.
> >
> > When we add new operators/utilities it takes certain time for them to
> > mature. Japicmp doesn't help because it doesn't honor @Evolving @Unstable
> > annotations for now.
> >
> > I wanted to propose that we add an annotation, let's say, re-use hadoop's
> > @Stable and mark the operators which are stable with it and perform
> semver
> > check on just these operators.
> >
> > The 0.7.0 version of japi cmp has the support for inclusions (as well as
> > exclusions) based on annotations.
> >
> > Here is the info:
> > https://github.com/siom79/japicmp/issues/88
> >
> > The reason I am inclined to the inclusion approach is that there are
> > relatively smaller number of operators which IMO are stable. A lot of
> them
> > aren't.
> > So instead of going and marking so many as Evolving, we will mark
> > relatively few of them as stable.
> >
> > Also new development can be facilitated by this. We wouldn't have to add
> > @Evolving to everything which is new. Instead we will mark it @Stable
> when
> > it is.
> >
> > Please let me know what you think?
> >
> > Thanks,
> > Chandni
> >
>

Reply via email to