On 2022-03-16 20:53, Alan Bateman wrote:
On 16/03/2022 08:44, Magnus Ihse Bursie wrote:
:
If you have such strong opinions on the data files shared between
java.base and jdk.charsets/jdk.localedata, I propose we leave them in
make/data for the time being, clean up the associated mess, and then
work out where they actually should be. Does that sound okay to you?
The concern, as before, is that it puts data files into src/java.base
that are used by the build to generate classes/resources for the
service provider modules. We also have the complication that the
charsets to include in java.base varies by platform so the
module/package for each charset is decided at build time. It's always
been low-priority to re-visit that and not clear if we could even get
to an agreement easily because there are IBM platforms that want
EBCDIC and other double byte charsets whereas other platforms don't
want these in java.base. So yes, if you can drop the move of the
charset data and CLDR data from the patch then it will make it easier
to discuss.
Okay, fine. I'll revert the move of `charsetmapping` and `cldr`, and put
them back in `make/data`. We can deal with them in due time. I agree
that they are tricky and might need more consideration than the rest of
the module-specific data files.
The `make/data` directory will not be entirely gone even with my
original patch. Left in there were files that did not belong to a
specific module, like macOS and Windows metadata files. One could argue
that the charsetmapping and cldr does not belong to a module as well.
The difference here is that it does not even make sense to talk about
module ownership for the other files in make/data , but these files are
clearly related to two modules (java.base and
jdk.charsets/jdk.localedata) -- just not a *single* module.
/Magnus
