Looks good to me. Thank you for fixing this. /Magnus
> 27 mar 2015 kl. 12:46 skrev Erik Joelsson <erik.joels...@oracle.com>: > > Hello, > > There are a number of issues caused by JDK-8076060: > > 1. A completely new configuration fails when running make the first time. > Second attempt succeeds. That's what failing in the bug description. > 2. Bootcycle-images is broken as the current working dir in Main.gmk has > changed from TOPDIR/make to TOPDIR. > 3. The sanity check that spec.gmk is actually a spec for the current source > tree fails if the source root has a symlink somewhere in the path, or if it's > in a special cygwin directory like /home/something/jdk9-dev. > > This patch addresses all of these. > > 1. was caused by the build output of module-deps.gmk ending up in the > ALL_MAIN_TARGETS. The problem was that the check for if module-deps.gmk > needed to be built only checked for the directory "make-support" instead of > "make-support/module-deps.gmk". The directory make-support currently exists > from just running configure. > > 2. Changed the make call for bootcycle-images to use an absolute path to > Main.gmk so that current working directory becomes irrelevant. > > 3. This sanity check is new. To get it to work better, I changed configure to > save two extra versions of the TOPDIR since the "topdir" calculated by make > doesn't always match what we found in configure. Then in the sanity check I > compare to all 3 different versions of TOPDIR. This seems to cover the cases > reported so far. Ideally a path comparison shouldn't be done just by > comparing strings, but we also need this to run fast at build time, so no > shell execution if it can be avoided. > > Bug: https://bugs.openjdk.java.net/browse/JDK-8076123 > Webrev: http://cr.openjdk.java.net/~erikj/8076123/webrev.top.01/ > > /Erik
