2.7.3 and serializable frustrations again

On Wed, 2009-01-14 at 10:36 +0100, Jan Kriesten wrote:
> It's frustrating reporting bugs (#1143), getting it fixed at last and then just
> to see it occur again just one release later! :-(

Shows the importance of including user-supplied test cases to avoid regressions.

Having said that, with the current test coverage it's also important
for users to test RCs to help catch problems early.

Ismael

Hi Ismael,

> Shows the importance of including user-supplied test cases to avoid regressions.

I've really no idea how to write a test to check this - that's too deep in the
compiler. I can help to discover bugs and provide the base to reproduce it, but
that's it in this case.

> Having said that, with the current test coverage it's also important
> for users to test RCs to help catch problems early.

There was no indication for me from the changelog that the new release would
have such an effect on my projects. Since this release should just be bugfixes
even more so.

It sounds a bit like "You didn't test it - it's your fault you didn't make sure
earlier". That can't really be the point.

I do my best to promote Scala - but introducing these bugs again (in the context
of my presentation in London on the 4th of Feb) is especially naughty to make
people confident to use the language in their own projects.

Regards, --- Jan.

On Wed, 2009-01-14 at 10:58 +0100, Jan Kriesten wrote:
> Hi Ismael,
>
> > Shows the importance of including user-supplied test cases to avoid regressions.
>
> I've really no idea how to write a test to check this - that's too deep in the
> compiler. I can help to discover bugs and provide the base to reproduce it, but
> that's it in this case.

I think you misunderstood what I said above. You included some code that
showed the problem in the bug, right? My point is that a regression test
should have been added to the Scala distribution to prevent it from
happening again.

> > Having said that, with the current test coverage it's also important
> > for users to test RCs to help catch problems early.
>
> There was no indication for me from the changelog that the new release would
> have such an effect on my projects. Since this release should just be bugfixes
> even more so.
>
> It sounds a bit like "You didn't test it - it's your fault you didn't make sure
> earlier". That can't really be the point.

Not at all. The point is that testing RCs is very important to help
catch regressions early given the current test coverage that Scala has.
That is just the reality of the situation. I think everyone hopes that
the situation will improve, but that will take time and will probably
require help from the community.

Besides, even projects that have a huge amount of tests and have been
out for years introduce critical bugs that are only uncovered by
third-party testing. See this Lucene corruption issue caused by a
Hotspot bug introduced in JDK6u4 that was only fixed in JDK6u10 as a
random example:

https://issues.apache.org/jira/browse/LUCENE-1282

Ismael