From a release point of view - I know we are currently sorting out some
regressions and issues that have been uncovered by the latest release.
So if it makes things easier from a stability point of view, we can wait
to merge this only after we have done another release which addresses
only the regressions that we have found in our latest releases.
-Jaikiran
On 15/02/18 7:21 PM, Jaikiran Pai wrote:
I'm now done with the initial goals that I had in mind for this task.
I've opened a PR[1] for review. I've included a manual for this task
and it can be currently found here[2]. I would suggest reading the
manual first, before reviewing the PR, since the manual will give an
overall idea of what's being attempted with this task.
Except for the "fork" mode which is available in our existing junit
task and which was was in the TODO list, in my previous mail, the rest
have been implemented in this version. I plan to focus on the "fork"
mode after this task is made available to users and any feedback
received.
I have tried to address some of the (internal) technicalities, based
on some discussions that I have seen about JUnit, XML reporter and
such, are around OOMs issues when it comes to sysout and syserr
handling. Result formatters, which can be configured as listeners, are
allowed to say "sendSysOut" and/or "sendSysErr". In such cases, the
task will (internally) redirect the sysout/syserr to a
PipedOutputStream (which is backed by a PipedInputStream). The task
also sets up 2 separate threads - one which reads the PipedInputStream
and "feeds" it to the second thread which "delivers" this content to
such result formatters. The necessity to use a separate thread to
deliver the content of sysout/syserr is so that we don't run into
deadlocks (as noted in the Javadoc of Piped(output/input)Stream class)
if/when the result formatters themselves write something out to
sysout/syserr when they are working on this incoming sysout/syserr
content.
We have 3 pre-defined formatters all of which are capable of receiving
this streamed sysout/syserr data. Each of these do _not_ hold on to
this sysout/syserr data in-memory during the execution and instead
stream it out a temporary file. Once the execution completes and it's
time to present the result, these formatters stream/read back the
content from the temp file and write it out in a formatted manner to
the target report files.
Furthermore, the XML formatter doesn't use DOM and instead is based on
Stax for writing out the report. However, even with all this, there's
one place where I haven't yet been able to avoid reading the whole
generated sysout/syserr data into memory (thus potentially triggering
an OOM) - it's the XMLStreamWriter API which for writing out a CDATA
section (which is what we use for sysout/syserr content) in XML,
expects the entire String object. I'll have to see if there are ways
to avoid it, but I think this still is an improvement since this
loaded content is only held on for a short while in memory, during the
report writing and will be immediately garbage collected once that's
done.
Please review the PR - mostly the Ant specific constructs and
implementation details.
[1] https://github.com/apache/ant/pull/60
[2]
https://builds.apache.org/job/Ant-Build-Jaikiran/ws/manual/Tasks/junitlauncher.html
[3]
https://github.com/apache/ant/pull/60/files#diff-f985f89ce4779eff2cdcf164d09d5396R308
-Jaikiran
On 27/01/18 8:40 AM, Jaikiran Pai wrote:
Here's an update on where this effort now stands. As of yesterday, I
have the basic minimal functionality that I had planned for this
(new) task ready. There are other enhancements that this task will
need as we go along but at this point, this should be usable.
While working on this new task, I kept thinking whether it would make
more sense to just have this task as a separate project under the Ant
umbrella and have its own release/versioning cycle. Plus maybe a bit
of ease in building/testing it given that it won't need some of the
conditional logic that we do in Ant build itself when it comes to
such tasks. It did look like a good idea to separate it out but
ultimately I decided _not_ to do it mainly for the reason that I
think it makes more sense to have a task in Ant which allows users to
write tests (using a modern test framework) and test them, right out
of the box of an Ant installation. This first class experience, IMO,
outweighs all the other "advantages" that seem to come with having
this as a separate project.
Coming to the task itself, the task is called "junitlauncher". In the
current state of this task[1], the following goals/features are
implemented and functional:
- Ability to launch the JUnit 5 platform
- Ability to specify a classpath to use for the launched tests
- Ability to specify one or more single test classes that need to be
run as part of this launch
- Ability to specify specific methods on these test classses that
need to be run as part of the launch
- Ability to specify "batch tests" which essentially is a way to use
Ant's resource collections to pattern match files that should be
passed on to the JUnit platform to be evaluated and run as test cases.
- Ability to specify "listeners" for tests. These listeners are
expected to implement (intentionally) JUnit platform's interface and
_not_ any of Ant's interfaces.
- Test result formatters are implemented as "listeners". This task
comes with 2 (for now) implementations out of the box, "plain" and
"brief". The idea behind these 2 is the same as that of what we
current have with the "junit" task formatters.
- Users can define custom formatters as "listeners" by specifying a
class which implements the JUnit's test listener plus (optionally)
one of Ant's own (new) custom interface. This custom interface will
let them have access to output stream to which they might want to
write out the results.
(I might be missing a few more details, but these are the major
functional features. The manual that I plan to write, will have the
whole details)
In its current state the task should be able to run both "vintage"
(JUnit 4.x) and "jupiter" (JUnit 5.x) based tests.
TODO/enhancements for later (a few days/weeks down the line):
- Ability to "fork" these tests in a separate JVM. I haven't fully
thought about this and might need some inputs on whether we need this
or not. If we do add this, I might do it slightly differently that
what we current do with "junit" where there are numerous attributes
to the task/test elements which are only applicable if fork mode is
enabled. I might perhaps just introduce a new element within the
task's element which specifically is meant to deal with any forked VM
characteristics. But that's something I will get to after the other
easier enhancements are done.
- Provide the XML formatter out of the box. I was planning to do this
in the first version itself, but I haven't had enough time to
understand the schema of this XML plus whether or not the details
that we put in here are available through the new JUnit launcher APIs.
- Ability to more specifically say which JUnit test engine needs to
be used for the tests. Right now, the classpath decides which engine
gets used and thus which classes are considered as tests. If the
classpath has both "vintage" and "jupiter" engines then such tests
will be run. However, I want users to be able to say "just use
jupiter engine for these tests" without having to worry about
checking if the classpath is polluted with some other engines.
- Make the current "junitreport" task be usable with the XML results
of this "junitlauncher" task. I haven't yet got to this but this
should be doable I think, of course once the XML formatter itself is
ready and functional. I don't plan to introduce a new task for this
and instead plan to reuse/enhance the existing junitreport task to
work seamlessly both with existing "junit" task and the
"junitlauncher" task.
(Few other minor enhancements here and there and any user feedback
reports)
For those of you curious to see what the task usage is going to look
like, here's an example[2] build file which shows its many usages.
That file will be cleaned up a bit, before I send a PR for review/merge.
[1] https://github.com/apache/ant/compare/master...jaikiran:junit5
[2]
https://github.com/apache/ant/compare/master...jaikiran:junit5#diff-3e0911ed925129e68c543aca0abc5d7aR52
-Jaikiran
On 14/12/17 5:14 PM, Stefan Bodewig wrote:
On 2017-12-14, Jaikiran Pai wrote:
With that context, I would like to explain what I have attempted so
far and I would like inputs on whether I should follow this path
and/or if there are other suggestions on how we should go ahead. As
you are probably aware, the JUnitTask (the Ant task backing the
<junit> element) has very complex logic which deals with launching the
test runners responsible for running JUnit tests (3.x or 4.x). IMO,
most this logic can now be handed off to the JUnit 5 platform
launcher, using their APIs.
In addition some of the complexity inside of JUnitTask stems from the
fact that it must not refer to any JUnit 4.x API in order to be usable
in a pure JUnit 3.x environment. If you try to add JUnit 5 on top of
that this sounds diffciult.
I too had toyed with the idea of adding JUnit5 support but it is
quite a
bit further down my TODO list and I've not looked into JUnit5 as deeply
as you did. What you've got so far looks promising, please please push
ahead.
My idea was to create a completely new task and mainly try to keep the
XML output the way it has been for some fifteen years so junitreport
would still work. I even though about placing it in a separate antlib
rather than Ant's core to allow it get released independently.
My vote would go to a separate task.
Stefan
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@ant.apache.org
For additional commands, e-mail: dev-h...@ant.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@ant.apache.org
For additional commands, e-mail: dev-h...@ant.apache.org
