Test runs with jigsaw/jake have slowed significantly in recent months, it seems to be somewhere around end-September when we initially started to see this.
The problem can be demonstrated with jdk9/dev too, it's not specific to jigsaw/jake. Furthermore, the issue seems to be specific to areas that have TestNG tests, specifically TestNG.dirs.
An example, run the :jdk_time test group in the jdk repo with the jtreg 4.1 promoted build (b12). timeStats.txt reports:
Mean 1.92s
Standard deviation 3.76s
Total elapsed time 1m 43s
but with jtreg 4.1 nightly (code-tools/jtreg tip 8d338a3a3f6a) we get:
Mean 8.86s
Standard deviation 47.27s
Total elapsed time 6m 34s
Here's another example this time the javax/sql tests in the jdk repo. With jreg 4.1 promoted build (b12) we get:
Mean 1.13s
Standard deviation 3.26s
Total elapsed time 0m 23s
but with jtreg 4.1 nightly it takes much longer:
Mean 6.85s
Standard deviation 22.77s
Total elapsed time 1m 38s
The issue appears to be that javac is invoked more often or else is invoked with options that significantly increase the compilation time.
The problem can be demonstrated with jdk9/dev too, it's not specific to jigsaw/jake. Furthermore, the issue seems to be specific to areas that have TestNG tests, specifically TestNG.dirs.
An example, run the :jdk_time test group in the jdk repo with the jtreg 4.1 promoted build (b12). timeStats.txt reports:
Mean 1.92s
Standard deviation 3.76s
Total elapsed time 1m 43s
but with jtreg 4.1 nightly (code-tools/jtreg tip 8d338a3a3f6a) we get:
Mean 8.86s
Standard deviation 47.27s
Total elapsed time 6m 34s
Here's another example this time the javax/sql tests in the jdk repo. With jreg 4.1 promoted build (b12) we get:
Mean 1.13s
Standard deviation 3.26s
Total elapsed time 0m 23s
but with jtreg 4.1 nightly it takes much longer:
Mean 6.85s
Standard deviation 22.77s
Total elapsed time 1m 38s
The issue appears to be that javac is invoked more often or else is invoked with options that significantly increase the compilation time.