Yesterday's test was a great success -- in that the test discovered a
huge flaw in the program. Or rather, it found a huge flaw in the
platform: the Timer object in Flash is horribly inaccurate.
I'm not a complete idiot -- I did some initial tests with flash before
spending ten hours writing that public test. Specifically, I ensured
that the clap-detection functionality was accurate enough. It wasn't
perfect (my physical metronome varied by 5 ms... although to be fair to
flash, it's possible that the metronome itself was varying), but it was
However, I didn't test that flash could make a light blink to within 5
ms. In retrospect, I should have. But it never occurred to me that it
would be a problem.
(and yes, it took almost all my willpower to avoid using "flash" as a
verb in the previous paragraph)
You can test it for yourself on your own computer: Metronome
On my system, a modest dual-2.0 Ghz macbook running OSX 10.4.11, half
the time I get a maximum delay of about 40ms, and half the time it
spikes up to over 200ms. That's totally unacceptable for a metronome --
especially when I'm going to be grading students on their clapping
ability. The whole point of this program, and my Master's thesis in
general, is that computers can provide objective, accurate, and above
all, trust-worth grading. If students can't trust that the computer
grades are accurate (at least to 5%), the program will be a pathetic toy
rather than an educational aid. And we have more than enough pathetic
toys masquerading as educational tools (or "edutainment").
Note that the above test is only timing the Timer class. I'm not
doing anything else here -- adding other stuff (like listening to claps
+ key presses, drawing flashing lights, etc) can only make it worse.
So what now? I can think of one last thing to try: display the metronome
as an animated gif on the web page. Haxe can pass information
program to display an animated gif (or svg/mng, if Internet Exporer has
finally caught up 1999/2001) on the HTML page.
Yeah, that's quite a bit of duct tape, even for me. But what's the
alternative? Abandon 90% of potential MEAWS users by ignoring Windows?
Spending a week setting up a development environment on windows so that
I can compile it? Neither of those options are particularly appealing.
Anyway, thank you to everybody who tried the past version. I'll
testing within 24 hours.