RFR: 8035063: Option handling in sjavac needs to be rewritten

Fredrik Öhrström oehrstroem at gmail.com
Wed Mar 19 12:58:21 UTC 2014

The code by itself looks good. However I think that changing from . to / is
a really bad idea.

If someone committed a package directory that does not map correctly to the
package name, then that is a major problem. You will not find that source
through -sourcepath, which means that you just broke sjavac ability to
compile using multiple threads. This is just as bad as when Nashorn tried
to insert $ into java source file names! The horror, the horror.

I.e. do not replace . with /. Fix the problem in the jdk.


2014-03-19 12:29 GMT+01:00 Erik Joelsson <erik.joelsson at oracle.com>:

> Ah, I see now, it was "-i some.package.*" in the original code. Build part
> looks good to me now.
> /Erik
> On 2014-03-19 11:17, Andreas Lundblad wrote:
>> On Wed, Mar 19, 2014 at 10:03:33AM +0100, Erik Joelsson wrote:
>>> Hello,
>>> The input from our makefiles changes from "-i some.package" to "-i
>>> some/package/*". IIRC the include flag works recursively for sub
>>> packages. Will the single * do the same?
>>> /Erik
>> That's strange. What used to be "-i some.package" should now be "-i
>> some/package".
>> The -i works recursively only if the pattern ends with a '*'.
>> I did however log all patterns that were given as arguments to -i during
>> a 'make clean images' and all those patterns ended with a '.*'. (Maby I
>> missed some patterns somewhere?)
>> Regardless, if something that used to be "some.package" and is now
>> "some/package/*", I'd say it's a bug.
>> -- Andreas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/compiler-dev/attachments/20140319/ab790504/attachment.html>

More information about the compiler-dev mailing list