curator-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sean Busbey <bus...@cloudera.com>
Subject Re: Is Curator's semver over binary or source compatibility?
Date Tue, 10 Mar 2015 05:43:37 GMT
I'm trying to figure out what sense of "use" is meant in that quote. Is it
safe to assume it means "all of the above"?

I ask because in anticipation of a dependency update, I ran 2.6.0 and 2.7.1
through the Java API Compliance Checker[1], and I'm trying to figure out
which (if any) of the following things I should file as bugs.

# org.apache.curator.utils.PathUtils.validatePath(String) in curator-client

In release 2.7.0, CURATOR-136 changed the return type of this method from
void to String. This is fine for a semver minor version under source
compatibility, but is a violation of semver under binary compatibility. A
downstream user will get NoSuchMethodError if their already compiled class
uses this method.

Note that changing it back will also be a breaking change, because the Java
language gives you no means to have two methods that only differ by return
type[3]. I think that means the fix for this problem is updating release
notes (or an upgrade guide or a FAQ depending on what else the community
already has).

#
org.apache.curator.framework.recipes.shared.SharedCountReader.getVersionedValue()
in curator-recipes
#
org.apache.curator.framework.recipes.shared.SharedValueReader.getVersionedValue()
in curator-recipes

In release 2.7.0, I think CURATOR-151 added these two methods to these
interfaces as a part of improving an API.

The changes are fine for binary compatibility provided nothing in the
framework ever calls them (doing so will result in NoSuchMethodError when
called on an instance compiled against the older interface).  AFAICT,
nothing in the framework accepts one of these interfaces and then calls
this method.

However, the addition of methods to Java interfaces breaks source
compatibility. The inter-process semaphore recipes work on arbitrary
SharedCountReader instances. That means that downstream folks who made
their own implementation of the SharedCountReader interface under 2.6.0
will get a compilation error when they attempt to update to 2.7.0+.

Removing either of these methods would be a breaking change under both
binary and source compatibility, so if source compatibility is desired the
fix in this case is likely the same as the PathUtils problem above.

[1]: http://ispras.linuxbase.org/index.php/Java_API_Compliance_Checker
[2]:
http://people.apache.org/~busbey/compat_reports/curator/2.6.0_to_2.7.1/compat_report.html
[3]: the JVM allows it, so some hijinks could get you a jar file that
didn't break binary compatibility with either 2.6 or 2.7, but it's a
terrible idea.

On Mon, Mar 9, 2015 at 7:06 PM, Jordan Zimmerman <jordan@jordanzimmerman.com
> wrote:

> The comment from Eric explains how we use it:
>
> Essentially, it tells you the API compatibility between multiple
> dependencies. For example,
>
> If you have version 1.0.0 and 1.0.1, you can use either one and you will
> not run into errors due to API conflicts (i.e. in Java, no ClassNotFound or
> NoSuchMethod type exceptions)
>
> If you have 1.0.0 and 1.1.0, you can replace 1.0.0 with 1.1.0 and you will
> not have errors due to API conflicts. If you try to swap 1.0.0 in place of
> something that depends on 1.1.0, however, you *might* have errors due to
> API conflicts. I.e. you can go forward without errors, but not backwards
>
> If you have 1.0.0 and 2.0.0, you will likely have errors due to API
> conflicts if you replace either one with the other.
>
>
>
> On March 9, 2015 at 6:18:16 PM, Sean Busbey (busbey@cloudera.com) wrote:
>
> Hi!
>
> I'm very happy to see that Curator has adopted semver[1].
>
> A couple of things that aren't immediately obvious:
>
> 1) Is the semver promise over binary compatibility, source compatibility,
> behavior compatibility or some combination thereof?
>
> 2) Is the semver promise over all public/protected things in all the
> published artifacts, or just some subset (e.g. just curator-client or
> curator-recipes, or some set of packages)?
>
> [1]:
>
> https://cwiki.apache.org/confluence/display/CURATOR/For+Curator+Committers#ForCuratorCommitters-versioning
>
> --
> Sean
>
>


-- 
Sean

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message