Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [jdt-dev] "clean up" again

I see that my last post had few raise eyebrows.

I do agree with internal API usages, even my self I have used internal APIs in a few of my plugins in Eclipse JDT. I'm not saying we should not do that, but we should have a process of promoting such most wanted internal APIs into external APIs to be more maintainable. And internal API users must have some CI jobs to make sure the upstream is breaking things for you. So we can raise those as soon as possible rather than waiting until an RC release. And for me, this is what Continuous Integration is about.

Maybe you in JDT have a different approach in seeing quality, module design, and CI than what I used to, given the fact that you have lot of projects depending on you.

Maybe no point of raising the same thing again and again, just wasting all of our time.y


On Sun, May 31, 2020 at 6:27 AM Ed Merks <ed.merks@xxxxxxxxx> wrote:

Speaking as an adopter, EMF has 32 uses of org.eclipse.jdt.internal in
the code base.  As such, sensitivity to adopters is very much
appreciated.  I know API rights activists will cry out about API
violations and argue their inviolate right to make arbitrary changes
because, well, they are within their rights of course.  It seems one
can't reasonably argue against such things because rights are always
right.  But reality is subtle and complex, with shades or gray ruling
the day.

I can well imagine there are good reason why Object Teams is
incestuously intertwined with JDT, so from a hundred paces away, to
critique its design based purely on superficial, ideological dogma, with
no actual deep understanding of that design, leaves a bad smell.   Let's
not do that.  Balance and respect and more important than rights and
rules and there are few more deserving of respect than Stephan.


On 30.05.2020 14:16, Stephan Herrmann wrote:
> Vetos, for sure, should be used sparingly and carefully.
> Yet, as part of their responsibility, committers need a way to reject
> changes, which they assess as having negative impact on the project.
> When I wrote "For compiler tests, however, I would veto any such
> change" I said so in full confidence that the active committers will
> agree.
> Regarding Object Teams, once more: this is a side issue (other than
> the fact that without Object Teams I would never have started
> contributing to JDT), and few people on this list have sufficient
> insights into that project to seriously discuss its architecture. We
> could just drop this, but I still believe there is an aspect that goes
> far beyond Object Teams:
> JDT is used by *many* adopters. Putting aside all design rules and
> ideology, it's a fact of life that some of them use internals of JDT,
> have perhaps been doing so for a very long time. We are not going to
> re-write other projects' history. On the contrary, as a courtesy to
> our adopters (who are an important part of the eco system!) we should
> minimize changes that are likely to cause grief and pain downstream.
> By "minimize" I mean: be a bit more conservative than usual, perform
> only those changes that are "necessary" in one way or other. Just
> raise the bar for approval.
> BTW: we have several deleted methods brought back after adopters have
> explained their problems with a change in JDT. Some of this looks
> quite odd and violates a number of rules, but if it helps adopters it
> helps the eco system, which is more important :)
> best,
> Stephan
> On 30.05.20 10:41, Gayan Perera wrote:
>> I think this was a good discussion. But i think we need to have mix
>> of people when doing veto. Other wise things will never move forward
>> if we are stuck with the same group who thinks in the same way.
>> I know stephan will not agree with me. But i think object teams have
>> great technical debt from what i read in this thread. A product
>> should no depend on tests of another product. Moving tests from one
>> junit 3 to junit4 should not impact heavily. Use test fixtures
>> instead of depending raw test classes. Look at how IntelliJ platform
>> has done it for plugin developers.
>> Rather than simply depending on a test plugin and writing your tests
>> on it. Its always good to have APIs which are not bounded to certain
>> implementation details on test frameworks.
>> By the way cleaning up whitespace issues and cleaning up code
>> formatting in a single commit as a bulk task, I don’t understand the
>> what the big issue that stephan tries to highlight.
>> Finally if you depend on non API classes then be prepared to get
>> breaking changes. It shouldn’t matter how big the product is that
>> depend on non API classes.
>> Br,
>> Gayan
>> On Fri, 29 May 2020 at 18:50, <stephan.herrmann@xxxxxxxxx
>> <mailto:stephan.herrmann@xxxxxxxxx>> wrote:
>>     Thanks for answers here and others.
>>     I feel we can agree that the JUnit3 -> 4 conversion is different
>> from other
>>     mass changes:
>>     * This change is done with close consideration of each individual
>> case
>>     rather than mechanically applying some scheme over unseen amounts
>> of code.
>>     * I see a tangible benefit (during development / for reporting
>> etc.), with
>>     just few comments:
>>         - In JDT I never saw anything nearly as bad as what Alex
>> linked (from
>>     team.cvs)
>>         - Once a new contributor realizes that he has issues with a
>> new test
>>     method not being picked up, this should be the kind of question
>> that should
>>     get a helpful hint from others in very short time. For simple
>> questions like
>>     this feel free to ping even in short intervals.
>>         - This JUnit migration causes significant follow-up work in
>> Object
>>     Teams, but I was happy to get helpful explanations from Carsten
>> and thus I
>>     did not complain about this.
>>     * Test code is just a tiny little bit less critical than main
>> code. This
>>     concerns cleanliness of the git history as well as bugs
>> introduced by mass
>>     changes (= only indirectly affecting users, while still affecting
>> adopters).
>>     I think in JDT/UI this particular activity is still ongoing, and
>> I don't
>>     object to its completion (I have no idea about the percentage
>> completed?). I
>>     hope other committers can agree, too?
>>     For compiler tests, however, I would veto any such change. I'm a
>> bit less
>>     decided about other tests in JDT/Core or JDT/Debug, but I feel
>> terminating
>>     this activity when JDT/UI is done is a fair compromise, OK?
>>     Regarding save actions, I second what Jonah said. Actually I
>> think it was a
>>     bug to enable any save actions that are not in sync with the
>> existing code.
>>     I hope with this we can tick off two items from the list as being
>> not quite
>>     as controversial as some may have felt.
>>     Stephan
>>     Am 2020-05-28 15:52, schrieb Pyves .:
>>>     I've contributed a few patches to JDT Core and UI in the past
>>> couple of
>>>     years, so I'm guessing I fit in the "new contributors" category
>>> and may be
>>>     able to provide some insight.
>>>     My first contribution was fixing bug 424214 and I faced two
>>> problems
>>>     related to this discussion:
>>>     * I struggled to write new unit tests. At the time, I had never
>>> used JUnit
>>>     3 (which is understandable given that JUnit 4 was released early
>>> 2006). I
>>>     was probably trying to write a test method with a different naming
>>>     convention and it wasn't being picked up by the framework - no
>>> longer sure
>>>     at this point. And as JUnit 3 was not a thing I had used, I
>>> didn't even
>>>     realise it was JUnit 3 (in my mind it was some bespoke Eclipse test
>>>     utility running) and consequently I couldn't easily look up any
>>>     documentation to solve my problems. In the end, I ended up
>>> putting the
>>>     tests in an existing file and copy-pasted as much possible, not
>>> really
>>>     understanding how things fitted together. For anyone who has
>>> started
>>>     writing Java in the past decade or so, these mass migrations to
>>> JUnit 4,
>>>     even though they touch a lot of files and introduce commit
>>> noise, are useful.
>>>     * I struggled to get the contribution under 1000 lines to avoid
>>> the CQ.
>>>     The files I changed had not been cleaned up nor touched in years,
>>>     therefore some of the automatic save actions had introduced
>>> additional
>>>     diffs, for example import ordering. With Till Brichy's help I
>>> then had to
>>>     revert some of these automatic changes, just for the sake of
>>> getting under
>>>     the 1000 line limit in time for the M3 deadline. Note that this
>>> was my
>>>     very first usage of Gerrit, so reverting lines and pushing new
>>> patch sets
>>>     was not as straightforward for me as it would be now. "Fighting"
>>> against
>>>     save actions would not have been needed had the files been
>>> cleaned up
>>>     prior to my contribution.
>>>     Admittedly, these are only two small inconveniences which some
>>> of you may
>>>     even consider as anecdotal, but hopefully they do illustrate
>>> cases where
>>>     mass cleanups can help newcomers. :)
>>>     Best regards,
>>>     Pierre-Yves
>>>     Le jeu. 28 mai 2020 à 15:08, Aleksandar Kurtakov
>>> <akurtako@xxxxxxxxxx
>>>     <mailto:akurtako@xxxxxxxxxx>> a écrit :
>>>         On Thu, May 28, 2020 at 3:07 PM Stephan Herrmann
>>>         <stephan.herrmann@xxxxxxxxx
>>> <mailto:stephan.herrmann@xxxxxxxxx>> wrote:
>>>             On 28.05.20 13:20, Aleksandar Kurtakov wrote:
>>>             > On Thu, May 28, 2020 at 1:55 PM S A
>>>             <simeon.danailov.andreev@xxxxxxxxx
>>>             <mailto:simeon.danailov.andreev@xxxxxxxxx>
>>>             > <mailto:simeon.danailov.andreev@xxxxxxxxx
>>> <mailto:simeon.danailov.andreev@xxxxxxxxx>>> wrote:
>>>              > [...]
>>>             >     I can't make a comment on attracting other
>>> contributors in JDT.
>>>             >
>>>             >
>>>             > That I can comment :).
>>>             Are you speaking from your own experience of working on
>>> JDT code
>>>             (as a new
>>>             contributor), or are these words you put into the mouths of
>>>             others? I'd
>>>             appreciate if they speak for themselves.
>>>         I speak as the tech lead for Jeff and Roland  and
>>> discussions on a
>>>         weekly basis what/how/when/why to do so we can share the
>>> burden in JDT
>>>         with others. Being the one that have pushed for people to
>>> work on JDT
>>>         and the one that has followed up most of the late additions
>>> to the
>>>         team and specifically organizing the team work so JDT team and
>>>         community can grow - yes I do speak from my own experience
>>> and would
>>>         dare to even say that have a broader view of the project not
>>> worse
>>>         than many committers.
>>>         I haven't worked on JDT code itself a lot (releng fixes after
>>>         incomplete fixes in JDT and -Werror addition) but I would
>>> dare to say
>>>         that non-trivial part of the work in the last few releases
>>> has been
>>>         requested by/approved by/checked by/etc. by me personally incl.
>>>         freeing time for people to work on JDT and further .
>>>             I'm willing to learn from our new contributors. It's
>>> among the
>>>             committers that
>>>             we have to find a mode of operation that facilitates
>>> collaboration
>>>             and avoids
>>>             stepping on each others' toes. It seems this mode has
>>> not yet been
>>>             found.
>>>              > P.S. Only whoever hasn't looked at unreadable JUnit3
>>> test
>>>             suites results [...]
>>>             I'm looking at such results [1] all the time and I see
>>> no problem.
>>>             Do you care
>>>             to be more specific?
>>>         - go even figure which test triggered the failing setup. You
>>> want see
>>>         it in later builds cause these specific tests have been
>>> disabled and
>>>         other such has been updated to JUnit4 - doing it regularly
>>> when my
>>>         daily look at test results spots such thing.
>>>             Stephan
>>>             [1]
>>>             _______________________________________________
>>>             jdt-dev mailing list
>>>             jdt-dev@xxxxxxxxxxx <mailto:jdt-dev@xxxxxxxxxxx>
>>>             To unsubscribe from this list, visit
>>>         --         Alexander Kurtakov
>>>         Red Hat Eclipse Team
>>>         _______________________________________________
>>>         jdt-dev mailing list
>>>         jdt-dev@xxxxxxxxxxx <mailto:jdt-dev@xxxxxxxxxxx>
>>>         To unsubscribe from this list, visit
>>>     _______________________________________________
>>>     jdt-dev mailing list
>>>     jdt-dev@xxxxxxxxxxx <mailto:jdt-dev@xxxxxxxxxxx>
>>>     To unsubscribe from this list, visit
>>     _______________________________________________
>>     jdt-dev mailing list
>>     jdt-dev@xxxxxxxxxxx <mailto:jdt-dev@xxxxxxxxxxx>
>>     To unsubscribe from this list, visit
>> _______________________________________________
>> jdt-dev mailing list
>> jdt-dev@xxxxxxxxxxx
>> To unsubscribe from this list, visit
> _______________________________________________
> jdt-dev mailing list
> jdt-dev@xxxxxxxxxxx
> To unsubscribe from this list, visit
jdt-dev mailing list
To unsubscribe from this list, visit

Back to the top