On 2020-04-30 15:07, Magnus Ihse Bursie wrote:
On 2020-04-30 14:55, Florian Weimer wrote:
* Magnus Ihse Bursie:
The code for setting up Java compilation has long been quite hard to
understand, and has a tricky API. Part of this is due to the support
for
the sjavac ("smart javac") system. We do not use sjavac anymore, and it
has not been tested for long. Part of the sjavac effort was extracted
into the "depend" javac plugin, and another part ended up as the java
server, both of which we still use.
This commit:
# HG changeset patch
# User ihse
# Date 1588142957 -7200
# Wed Apr 29 08:49:17 2020 +0200
# Node ID a8e4856be54f73225780a56b2b6ee9cc1e12ab8f
# Parent f53590a82709caf92213106954eab2bdc97cc498
8244036: Refresh SetupJavaCompilation, and remove support for sjavac
Reviewed-by: erikj
appears to have regressed build time (“time make jdk-image”) by
roughly one third, on an x86-64 machine with 8 cores/16 threads.
The configure arguments are “--with-zlib=system --with-lcms=system”,
so nothing sjavac-related as far as I can see.
Is this expected?
No, that is very much not expected. :-(
I made sure that no build performances were measured on my system,
"... no build performance *regressions* ..."
/Magnus
and since I saw no such indication, I did not make any more systematic
analysis.
What is the difference if you run with or without the javac server?
Both prior to and after this commit. Do you have any numbers to share?
/Magnus
