Closed melezhik closed 1 year ago
I have an interesting effect, when using zef install -to=home
Rakudo that is tested first has way more faster time , then the following one, irrespectively on version ...
Here, 2022.07 is tested first report:
21:27:59 :: O----------------O--------O-----------------------------------------------------------O------------O
21:27:59 :: | Rakudo Version | Status | Time | Linux Dist |
21:27:59 :: O================O========O===========================================================O============O
21:27:59 :: | 2022.07 | OK | 1:26.51 real,129.52 user,7.70 sys / CPU Percentage: 158% | debian |
21:27:59 :: | 2022.06 | OK | 3:55.38 real,353.37 user,23.51 sys / CPU Percentage: 160% | debian |
21:27:59 :: ----------------------------------------------------------------------------------------------------
Here, Rakudo 2022.06 is tested first report:
20:47:00 :: O----------------O--------O-----------------------------------------------------------O------------O
20:47:00 :: | Rakudo Version | Status | Time | Linux Dist |
20:47:00 :: O================O========O===========================================================O============O
20:47:00 :: | 2022.06 | OK | 1:30.74 real,133.11 user,8.00 sys / CPU Percentage: 155% | debian |
20:47:00 :: | 2022.07 | OK | 4:04.41 real,369.63 user,24.25 sys / CPU Percentage: 161% | debian |
20:47:00 :: ----------------------------------------------------------------------------------------------------
so 1 min VS 4 min is significant difference, a I don't know why this happens, maybe because when we run tests for the second whatever rakudo versions , cache (~/.raku
) from the first Rakudo installation somehow affects the second Rakudo tests?
also worth to mention that 4 minutes for tests is way too much ... 🤔
is it even legit way to install into home (-to=home) and reuse the same files (~/.raku
) across different rakudos ?
cc @ugexe
Worth noting that It's the precompilation overheads that are the bottleneck, not the tests themselves..
On a second run, zef test .
takes about 30sec on my modest debian x86 i7 PC. prove6 -I. -j4
about 12sec.
@dwarring the point I am trying to make here - that the tests that runs second time are always slower irrespectively on rakudo version, which is weird, as following the logic pre-compilation already happened (drying the first tests run) and the second tests run should be faster at this point, what I see is quite opposite ...
Also I use
-to=home
to address issue when tested against various rakudo versions ( the issue is not presented here, however this is a general problem ).