review request for 6717128

[Xiaobin Lu]

Webrev: http://javaweb.sfbay/~xl116366/webrev/6717128/

6717128: Using relative path for ALT_OUTPUTDIR does not work properly

Details:

The problem I saw when I use a relative path for ALT_OUTPUTDIR is that 
there is a redundant directory gets created. For example, if I set 
"ALT_OUTPUTDIR=../../vm_output" and launch the build from the top level 
make directory, I will see a vm_output directory got created 3 levels 
beyond the current directory (besides the actually output dir which is 2 
levels beyond the current dir) and that is unexpected and annoying.

And here is the root cause, on both Solaris & Linux, when building 
hotspot from the top level make directory, the Makefile there will 
launch a sub make process and use the OS specify Makefile to complete 
the build. For example, on Linux, the top level Makefile will launch 
"make -f make/linux/Makefile $(MAKE_ARGS)" at some point. The OS 
specific Makefile imports "make/defs.make" which contains a bunch of 
variable definitions for further references. The problem is in 
defs.make, it will re-evlaute ABS_OUTPUTDIR and since OUTPUTDIR is a 
relative path, it will make another directory which is 2 levels beyond 
the current directory which is the real output directory. That is 
exactly why I saw that extra directory got created. To fix this, we just 
need to export the already computed ABS_OUTPUTDIR as a absolute path in 
MAKE_ARGS variable so that even ABS_OUTPUTDIR is re-evaluated, the 
OUTPUTDIR is already an absolute path.

Along the above fix, I am also trying to fix the incremental build 
failure due to my recent put back. It relates the way we define GAMMADIR 
in defs.make which either gets from the make command line or gets 
figured out by traversing the directory until it can locate the src 
directory. In the incremental build case, unlike the normal build, we 
don't pass GAMMADIR=$(ABS_GAMMADIR) to the sub make process so any 
places which includes defs.make will expand the GAMMADIR definition and 
eventually get the wrong answer if it can't locate the src directory 
which will likely happen when building remotely. The fix is to instead 
of importing the whole defs.make in rules.make, just include the part as 
a separate file and the upside of doing that is it could avoid 
unnecessary redefinitions and re-evaluations of variables in defs.make.

Verified by:
Remote incremental build and normal build on Solaris & Linux.

Thanks,
-Xiaobin