On 2017-03-30 15:42, Erik Joelsson wrote:
Hello,
Javadoc.gmk:159 looks like too much indentation
The variable JAVADOC_SOURCE_DIRS is starting to seem a bit redundant.
It does contain two directories that would not be included if you
replaced $(call PathList, $(JAVADOC_SOURCE_DIRS)) with $(call
GetModuleSrcPath), but that could be rectified by adding
$(SUPPORT_OUTPUTDIR)/rmic to GENERATED_SRC_DIRS and share/doc/stub to
SRC_SUBDIRS in Javadoc.gmk. Doing so would make the dependency
calculation and the actual source dir input be based on the same
definitions.
This is a really nice idea! Unfortunately, it was not so easy. :-(
Adding rmic to GENERATED_SRC_DIRS causes java.base compilation to fail with:
Compiling 2898 files for java.base
warning: [path] bad path element
"/localhome/hg/jdk9-sandbox/build/linux-x64/support/rmic": no such directory
error: warnings found and -Werror specified
1 error
I could of course add an mkdir somewhere before java.base-java is
executed, but it doesn't really feel good. I think it's better to
explicitly add the rmic gensrc dir to the javadoc generation.
The same goes for the share/doc/stub. Adding it caused a whole bunch of
errors when compiling java.management.rmi:
/localhome/hg/jdk9-sandbox/jdk/src/java.rmi/share/doc/stub/java/rmi/activation/ActivationGroup_Stub.java:72:
warning: [rawtypes] found raw type: MarshalledObject
public java.rmi.MarshalledObject newInstance(
^
missing type arguments for generic class MarshalledObject<T>
where T is a type-variable:
...
I assume that the sole reason for putting ActivationGroup_Stub.java into
the "share/doc/stub" directory was to prohibit it from being part of
normal compilation.
So, once more, I'll rather add that directory to just the javadoc
generation.
Nevertheless, using $(call GetModuleSrcPath) instead is very nice, even
if I have to add a couple of paths.
New webrev coming up soon...
/Magnus
/Erik
On 2017-03-30 15:05, Magnus Ihse Bursie wrote:
As a part of JEP 299, we should build the Javadoc as a single
combined output, instead of a dozen or so individual javadoc bundles.
This bug fixes this. The selection on what to include is now based on
modules instead of packages.
The fix in MakeBase.gmk is to keep CacheFind quiet if the src dir(s)
does not exist, otherwise find can emit an error message. (This was
provoked by the new call to SetupZipArchive).
The module selection has been contributed by Mandy Chang.
I intend to push this to JDK9. Since this is a noreg-doc bug, no
special RDP2 process is required.
Bug: https://bugs.openjdk.java.net/browse/JDK-8172312
WebRev:
http://cr.openjdk.java.net/~ihse/JDK-8172312-combined-javadocs/webrev.01
/Magnus
