On 05/02/2016 12:06 AM, Alan Bateman wrote:
On 01/05/2016 17:02, Chris Hegarty wrote:
On 1 May 2016, at 14:50, Andrej Golovnin <andrej.golov...@gmail.com>
wrote:
Hi Alan,
If you would like to improve the performance a little bit more,
you can make additional changes (see the patch) to JmodTask
and maybe Claes can rerun his tests to see if it helps.
These look good too, even as general clean ups, but it would
be good to get some numbers on them.
Andrej - aside from the micro benchmarks, the real test is `make
jmods` where the tool runs N times (N = number of modules) to create
the packaged modules. Short lived tools have always been the JDK's
Achilles heel. It may be that we will eventually need a "jmod server",
like the sjavac server in the build today. Alternative it could be
wrapped in another tool that capable of creating several packaged
modules. There is a balance here as we also don't want to over
complicated things.
-Alan
Rather than build "sjavac"-like server functionality into selected
tools, maybe we could generalize the server part of sjavac to be a "tool
server", capable of running any tool implementing a simple Tool interface.
-- Jon
