harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Stepan Mishura" <stepan.mish...@gmail.com>
Subject Re: [buildtest] Are there obsolete fields in the test's description? (Was: Re: [bti][p-unit] finally, it is ready)
Date Thu, 01 Nov 2007 08:41:52 GMT
Hi Alexei,

On 10/30/07, Alexei Fedotov <alexei.fedotov@gmail.com> wrote:
> Stepan,
> You asked:
> > What made them [fields] "obsolete"?
> Actually the thing I'm suggesting by my patch is to make these fields
> "optional".

This is not "true" - they are optional. It is possible to omit them in
the test's description - the harness runs the test, only warning is
printed. For example, if you omit 'testID' then exception is thrown
and the test is not run.

> I believe I may want to make Harmony becoming more
> convenient for a test developer I am.
> The author field was made obsolete by Geir's decision to keep authors
> outside of source code (remember removing Ivan Volosyuk from
> interpreter sources?) This decision was discussed a bit, but there
> were no strong arguments against.

IMO, 'authors' names in code source and 'authors' field in test's
description are two different things. I wouldn't mix them. Yes, there
is the agreement (OK to remove) about the first one and the second one
hasn't been discussed before.

> I don't see a reason of keeping date-of-creation attribute as a
> mandatory field. The following argument is a bit stronger: this
> argument is duplicated with another one and should be completely
> removed. The first file modification is usually done at the Day of
> Creation, and the current format allows several modification dates.

Again, I believe that if there is a reason to remove some feature then
the removal should be entire and complete. We shouldn't remove only
warnings and leave other debris of functionality in the harness.

For example, currently the harness has the option to select tests for
running based on authors name. If I want to run all tests created by
you I should pass to the harness something like: "-execopt
Selector:exclude:Author Fedotov". Also in this case a warning about
tests that were not selected because of missing authors field is very
helpful. And your patch removes it. The same for "creation-date".

> Thanks.
> BTW, to make this discussion a bit more interesting for techies, let
> me add here a discussion of memory cleanup algorithm from our chat
> with Andrew:
> Andrew: Does it make sense to invoke System.gc() multiple times to
> release memory completely?
> me: I have the following assertion in my tests: allocated == finalized
> and there is no other way to check that all objects are finalized than
> to invoke gc and to check that amount of available memory no longer
> increase (I also check that max chunk size stabilizes)
> Andrew: so to release memory completely, which approach is better? 1.
> invoke gc multiple times; 2. gc, thread.sleep
> me: I believe I combine both:
>  sleep does actual finalization work :-)
>  gc() is for check

I'd like to add a couple of notes here: If you like to discuss stress
tests details it is better to start new thread with corresponding
topic. And before revealing private conversation details please make
sure to get consent from all parties involved.



View raw message