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