The issue SQL is seeing is caused by a default dependency of exclude(".*")
added in build_rules.gradle. This breaks the normal method of building
shadow jars as everything must be explicitly included. SQL explicitly added
calcite to the jar, but not calcite's dependencies. I've been told this is
the desired behavior as we want to ensure everything included is relocated.
I don't know much about gradle, but this seems fragile. Is it possible to
have all dependencies automatically relocated so we don't need the
exclude(".*") rule?
Andrew
On Thu, May 17, 2018 at 7:41 PM Andrew Pilloud <apill...@google.com> wrote:
> Yep, I added the issue as a blocker.
> https://issues.apache.org/jira/projects/BEAM/issues/BEAM-4357
>
> On Thu, May 17, 2018, 6:05 PM Kenneth Knowles <k...@google.com> wrote:
>
>> This sounds like a release blocker. Can you add it to the list? (Assign
>> fix version on jira)
>>
>> Kenn
>>
>> On Thu, May 17, 2018, 17:30 Lukasz Cwik <lc...@google.com> wrote:
>>
>>> Typically we have a test block which uses a configuration that has the
>>> shadow/shadowTest configurations on the classpath instead of the
>>> compile/testCompile configurations. The most common examples are validates
>>> runner/integration tests for example:
>>>
>>> https://github.com/apache/beam/blob/0c5ebc449554a02cae5e4fd01afb07ecdb0bbaea/runners/direct-java/build.gradle#L84
>>>
>>> On Thu, May 17, 2018 at 3:59 PM Andrew Pilloud <apill...@google.com>
>>> wrote:
>>>
>>>> I decided to try our new JDBC support with sqlline and discovered that
>>>> our SQL shaded jar is completely broken. As
>>>> in java.lang.NoClassDefFoundError all over the place. How are we testing
>>>> the output jars from other beam packages? Is there an example I can follow
>>>> to make our integration tests run against the release artifacts?
>>>>
>>>> Andrew
>>>>
>>>
