all standards should have a "modified" variant

Looks more like UTF-16 in little endian order if you ask me, with \0
being translated to null somewhere along the line.

Sent from my (new) iPhone

On 6 Sep 2009, at 04:54, Paul Phillips wrote:

> http://scala-webapps.epfl.ch/hudson/job/scala-nightly-auxjvm/jdk=ibm6/36...
>
> Missing dependency 'class
> jnullanullvnullanull.nulllnullanullnnullgnull.nullOnullbnulljnullenullcnullt'
>
> ... and other such fun things. This is clearly related to the UTF8
> patch I just committed which uses java's built-in encoder rather
> than a
> hand-rolled one, and based on the pattern of nulls it looks related to
> the unfortunate "modified UTF-8" which java uses internally some
> places
> to avoid using 0000 in strings. What I don't know (and please fill me
> in if you have a guess) is why "IBM jdk6" is the only nightly which
> failed unhappily.
>
> Ideally it's some kind of insufficient cleaning in that build which
> resulted in a pickler format mismatch. You might be tempted to point
> out right now that paulp should let sleeping UTF8 encoders lie, and
> that's probably true in general, but encodings are something I believe
> can be done right for 2.8 and seeing that through required a few light
> whacks on the beehive with this handy stick.
>