> On 19 Oct 2015, at 19:09, Andrew Wang <[email protected]> wrote:
>
> Could we make it so we run RAT after building docs but before running
> tests? I think that's what's causing these issues, it looks like temp
> output from running tests. This would be a nice intermediate step toward
> that plug-in you mentioned, since test output is certainly not part of the
> release tarball.
>
> These tests should still be fixed up to use target/ instead of
> build/test/data though. Steve, I'm happy to review if you're chasing these
> down, or we could trade roles.
>
> On Mon, Oct 19, 2015 at 9:43 AM, Allen Wittenauer <[email protected]> wrote:
>
>>
>> On Oct 19, 2015, at 9:21 AM, Andrew Wang <[email protected]> wrote:
>>
>>> Question, are these errors from running apache-rat:check after running
>>> tests? build/test/data looks like a carry-over from the ant days, where
>> it
>>> was used instead of target/blah/blah. Point being, I thought we only
>> needed
>>> to run RAT on a clean source directory.
yeah, some of the code had clearly dates from then
delving into the tests -it was actually worse. A lot of the code had the
System.getProperties("build.test.data","build/test/data"), so there were lots
of references likring
and once you go there, you discover that at lot of the tests didn't have
test-name-specific subdirs under that, just something like "/dfs", which of
course is doomed to lead to spurious failures once parallel test runs are on
(which they now are).
I've ended up doing a more brutal change than just
s/"build/test/data"/r/"target/test/data"/, as there was a more fundamental
issue with temp directory setup. There are now some methods in GenericTestUtils
to set these things up, though they didn't get picked up in the run:
https://issues.apache.org/jira/browse/HDFS-9263
Allen, will an hdfs patch pick up changes to hadoop common in the patch?
