I wrote:
> I'm not sure I see why checkpointing every test should be particularly
> slow.


Michele Bini writes:
> Because you have to wait after every test for the checkpoint to
> be written to disk and the disk to be flushed

I'm not convinced that has to all that slow.

> and disk's builtin cache to be flushed, too

I don't think we have to wait for that.  When a test hangs the user is
going to stare at the screen for a bit before hitting reset.  By that time
the disk should have flushed itself.  It doesn't care that the computer is
hung.

A compromise:

Checkpoint every N tests.  When you come back from a crash write out a
record noting that a crash occurred and start testing from the last
checkpoint, checkpointing every test and warning the user that another
crash will occur.  When you come back from the second crash you know which
test is bad.  Skip it and go back to checkpointing every N tests.

I *really* don't like the idea of subjecting the user to two crashes,
though.
-- 
John Hasler
[EMAIL PROTECTED] (John Hasler)
Dancing Horse Hill
Elmwood, WI

Reply via email to