On Sat, 25 Aug 2018, Chris Cunningham wrote:

> Hi again.
> After the discussion a while back about 64 bit being slower than 32, I ran the full test suite for the latst 5.2a release against both on Windows.
> 
> 32bit - 22:52.193
> 64bit - 25:07.789
> 
> Yep, looks like the truth.

I tried running the tests myself on windows 10 and they really take 
extremely long to finish compared to linux. I suspect that I/O operations 
(files and network) are responsible for this.

> 
> The results - same number of failures (21) and expected failures (29) on both.
> 32bit had 2 errors
>   - WebClientServerTest>>#testGetFields (Test timed out)
>   - AllocationTest>>#testOutOfMemorySignal
> 
> 64bit has 2 errors:
>   - WebClientServerTest>>#testServerDestroy (twice) (Test timed out)
> No idea why it had it twice - weird.
> 
> Both also had 1 unexpected passes as well.  How do you find these in the TestRunner?

I think the passing test has been fixed. I don't remember which one it 
was, but it was a fairly basic one.

Levente

> 
> Hmm.  The timeouts - before writing this email, I tried it a few times and continued to get timeouts. Then, looked away, fiddled, and they ran.  After clicking on one of them about 30 times, I got the
> timeout error again, then once again they are clearly sub-second.  Very annoying heisen bugs.
> 
> The 32bit outOfMemorySignal is failing at allocationTest>>vmParameterAt:put:, where it is getting called to set parm 67 to 1712567184 (instead of 0, which it was previously).  I'm not sure how to
> debug this - anything else that would help?
> 
> Thanks,
> cbc
> 
>