Looks good to me.
/Magnus
On 2014-02-05 11:35, Erik Joelsson wrote:
Hello,
This change is intended for jdk8u (and jdk9).
Webrev: http://cr.openjdk.java.net/~erikj/8033210/webrev.root.01/
Bug: https://bugs.openjdk.java.net/browse/JDK-8033210
Background:
In make/common/JavaCompilation.gmk, in the macro SetupJavaCompilation,
we use the (new) -h feature in javac, which makes javac automatically
run javah on classes with native methods. A problem is that javac will
always print these headers, regardless of if they are there already or
if they changed. To only cause recompilation of native code when
something actually changed, there is logic in the macro to generate
the headers to a temporary directory and then compare the output to
any existing header files in the actual headers directory, and then
only copy it over if it has changed. The temporary directory is then
deleted.
What I suspect is happening here is that, since the name of the
temporary directory is only a function of the name of the actual
header files directory, each call to SetupJavaCompilation, which
points to a common header files directory, will use the same temporary
directory. There is potential for a race if these execute in parallel.
On windows, there are a couple of extra calls to SetupJavaCompilation
that will execute in parallel, which is why this happens there and not
on any other platforms.
The fix is pretty simple. Make sure the temporary directory name is
unique for each call to SetupJavaCompilation.
/Erik
