consensus is not to do it this way. closed.

On Jun 29, 2:53 pm, Daniel Clifford <[email protected]> wrote:
> Hi Mads,
>
> The idea was indeed improved testability.
>
> Some of the tests we have to verify that internal event happens in the
> runtime are fragile. I have run into this with my current work on unboxed
> double arrays, but there are plenty of others examples. I would like to be
> write specific unit tests (not performance tests) to verify edge cases of my
> logic. But my tests will always pass with the auto-unboxing doesn't kick in
> at all, making the tests useless.
>
> This actually happened yesterday... somebody checked in a change that
> accidentally disabled the conversion to double arrays, and my code no longer
> got called, but all of the tests still passed. I could add a forcing
> function to cause the unboxing to happen, but then I can't actually test the
> logic which is intended to trigger the transition in the first place. I
> could write separate performance tests for each edge case, but that smells
> way too heavyweight and doesn't seem right either. It seems like
> whitebox-style testing in that case would be more robust and allow me to
> write simple, clear tests for the edge cases. My suggestion to Yang was that
> counters seemed to be a good mechanism to provide the needed
> instrumentation, and I didn't really see a downside, since there seem to be
> a fair number of counters that have been around for a while.
>
> Danno
>
>
>
>
>
>
>
> On Wed, Jun 29, 2011 at 5:22 AM, <[email protected]> wrote:
> > I'm not sure I understand why we would want to expose counters to
> > JavaScript?
>
> > Counters is an internal programmer tool that we use to gather statistics
> > for a
> > while and then delete again. I don't think we should do testing at that
> > level. I
> > would test behavior in  tests and performance in benchmarks. If we are
> > making a
> > change that improves performance I think we should create a benchmark that
> > shows
> > it to stop us from regressing. If we want to make sure that a test hits a
> > certain state, we should add functionality to force it instead of relying
> > on
> > counters I think.
>
> > There might be compelling use cases that I haven't thought of? :-)
>
> >http://codereview.chromium.**org/7281007/<http://codereview.chromium.org/7281007/>

-- 
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev

Reply via email to