Comment by David Jones (JIRA)

A flame graph from a Linux profile shows most of the time is spent in

djones6 (JIRA User) Was the snapshot from 6/21 the first one you could reproduce this with?

Based on the available snapshots it looks like the regression happened sometime between 05/09 and 05/21

Comment by David Jones (JIRA)

Is there anything I can do to help narrow this down further?

Comment by David Jones (JIRA)

I tried the master development snapshots and I don't see this slow compilation problem there. This only seems to be slow on 5.1 snapshots later than 2019-05-09-a. Here are some timings from a macbook pro, each time running
swift package clean ; swift build ; time swift build --build-tests

swift-DEVELOPMENT-SNAPSHOT (master):
2019-05-09-a: real 0m21.495s
2019-05-15-a: real 0m21.588s
2019-07-02-a: real 0m21.801s
... times seem very consistent.

swift-5.1-DEVELOPMENT-SNAPSHOT:
2019-05-09-a: real 0m19.756s
2019-05-21-a: real 0m49.115s
2019-05-29-a: real 0m52.031s
2019-06-12-a: real 0m51.183s
2019-06-16-a: real 1m25.675s
2019-06-20-a: real 1m25.253s
2019-07-02-a: real 1m22.068s
2019-07-10-a: real 1m20.807s
... looks like two distinct increases in compile time, the first somewhere between 05-09-a and 05-21-a (swift.org doesn't have any snapshots between these dates), then again between 06-12-a and 06-16-a.

However, all these times are significantly slower than 5.0.1: real 0m5.611s

Finally had time to track this down: #27150

…though admittedly that's only one change, and maybe there were two. I'm not seeing anything particularly slow in building the tests anymore, though.

Oh, I misread this most recent message, too. Maybe we have other speedups in master besides what I just tracked down…but this was definitely the biggest thing even with an Xcode 11 compiler. I am a little confused because the problem was introduced in #23867 which was in April.

5.1 branch (though probably not 5.1.0, not with Xcode 11's GM seed already up): #27164

Thank you, @belkadan!

All right, should be in both master and the 5.1 branches. David, would you mind checking to see if speeds are back to normal, or if you're still seeing a significant difference? (I know 5.1 may still be a little slower than 5.0 in some areas, unfortunately, but I didn't observe any other order-of-magnitude problems.)

Comment by David Jones (JIRA)

I tried with the latest (2019-09-16-a) 5.1 snapshot and compilation time has returned to roughly the same as 5.0.3. Repeating the experiment from my initial report:
5.0.3: real 0m1.178s, user 0m1.775s, sys 0m0.269s
5.1-DEVELOPMENT-SNAPSHOT-2019-07-25-a: real 0m18.886s, user 0m19.703s, sys 0m0.365s
5.1-DEVELOPMENT-SNAPSHOT-2019-09-16-a: real 0m1.204s, user 0m1.787s, sys 0m0.306s

I'd say that this is now resolved.