With the new run-test test running framework, we introduced
FindTests.gmk, which among other things, parses all the jtreg
TEST.groups files to generate make targets for each group. While this is
very nice, the current implementation in FindTests.gmk adds significant
time to a simple "do nothing" make execution.
I have experimented with a different implementation, based on caching
the expensive calculations in a makefile that automatically gets
re-generated if any of the inputs change.
On my Linux workstation, running the very early make target
"buildtools-langtools" with the old and new implementation gives the
following times (real time as reported by the time command):
Old: 460-470ms
New (warm cache): 280-285ms
New (cold cache): 370-380ms
So, even with the cache cold, the new implementation is faster, and with
the warm cache, 180ms is quite a lot. It can definitely be felt when
running interactively in the shell.
The new implementation also removes the need for the workaround in
InitSupport.gmk introduced in JDK-8213736, so I'm removing that.
Bug: https://bugs.openjdk.java.net/browse/JDK-8221762
Webrev: http://cr.openjdk.java.net/~erikj/8221762/webrev.01/index.html
/Erik
