I send this review prematurely. Here is a new webrev, which has been
properly tested. We had some test code compilations using the boot
javac, which hadn't properly declared DISABLE_SJAVAC. The copy paste
error in configure has also been corrected.
On my workstation, running "make java" with hotspot already built, the
speedup goes from 2:32 to 1:35. Running just "make" with the default
target goes from about 4:05 to 3:10.
Webrev: http://cr.openjdk.java.net/~erikj/8140312/webrev.02/
/Erik
On 2015-10-23 17:21, Magnus Ihse Bursie wrote:
On 2015-10-22 12:38, Erik Joelsson wrote:
In JDK-8135131, sjavac added a thin server mode, in which only the
server part of sjavac is used for running plain javac. This patch
adds a configure option to enable this mode of running to the jdk
build. Later, if it seems to work well, we will consider making it
default.
On my workstation (linux) running the default make target went from
about 4m55s to 4m05s when activating this.
Did that test really use the new sjavac server? :-) The code looks
broken, in common/autoconf/build-performance.m4 you set ENABLE_SJAVAC
and not ENABLE_JAVAC_SERVER in the new flag.
Apart from that, it looks good. :)
/Magnus
