On 2017-09-04 14:34, Thomas Stüfe wrote:
Hi Magnus,
On Mon, Sep 4, 2017 at 12:09 PM, Magnus Ihse Bursie
<[email protected] <mailto:[email protected]>>
wrote:
On 2017-08-11 13:00, Thomas Stüfe wrote:
Hi all,
when building OpenJDK 10/hs on Windows, I get sporadic
configure errors.
Usually one of two things, either:
configure: The tested number of bits in the target (0) differs
from the
number of bits expected to be found in the target (32).
or the endianness test failing.
Most of the time, just retrying the configure run works.
I have the feeling this happens mostly when running several
builds in
parallel. I am currently using 32bit cygwin.
You cannot/should not run multiple "configure" in parallel in the
same directory. This is due to a limitation of the autoconf
framework that we are using -- they test features (like word size)
by creating a file in the current directory (conf.c I believe it's
called), compiling and running it. So if you run multiple
configure calls in parallel, these tests can interfere with each
other.
Thanks for looking into this.
Not sure I understand you: I call configure script from different
build output directories ("output-slowdebug", "output-release" etc),
one directory per build. But all builds reference the same source
tree. Would that be the problem you describe?
If you do
cd jdk10
mkdir -P build/output-slowdebug
cd build/output-slowdebug
bash ../../configure
(and similarly for output-release) then it should work, afaik. If you do
cd jdk10
bash configure --with-conf-name=output-slowdebug
(and similarly for output-release) then the configure scripts run in the
same directory, and it can fail.
So if you do the first option, and it still fails intermittently, then
there is something else at play.
/Magnus
..Thomas
Once the configure step is done, you can build multiple
configurations using "make" at the same time.
/Magnus
Does anyone see similar errors or maybe have a solution?
Best Regards, Thomas
