Well almost.
But according to Bloch's testimony, the sequence of
events was roughly:
- Long before Android, the (slower) sort code
in Sun's
arrays.java used RangeCheck to generate the error
messages/exceptions
returned to apps/applets that called the
sort method with various kinds of bad
parameters.
- The initial (before TimSort) Android implementation of
the sort method for arrays contained non-infringing code
that emulated
RangeCheck error messages closely enough to
work in real life, and any
differences in behavior from the
Sun implementation would thus have become an
implicit part
of the Android API.
- Then Bloch wrote the Java port of
Python's TimSort and
provided a license to both Sun and Google. All that new
code is legally in both products with rights belonging to
each party. No one
has disputed that, and phase 1 is over,
so that is now a hard fact of the
case.
- In the copies of TimSort Bloch gave to Sun, he reused
the exact
RangeCheck code from Sun's existing arrays.java.
He did that to ensure that
the error messages would not be
changed by making arrays.java use TimSort in
place of its
old sort code, thus minimizing the risk of incompatibility
with
previous versions of Sun's arrays.java. This was legal
because the 9 lines
remained Sun's property available only
from Sun. Bloch has testified, and this
is undisputed, that
he assumed a Sun engineer would paste the TimSort code into
Sun's arrays.java, eliminating the extra copy of RangeCheck.
The copy of
RangeCheck in Sun TimSort was there only to
allow Sun test engineers to verify
that the TimSort code
would be a perfect fit for the merge, when not changing
the
RangeCheck code.
- In the copies of TimSort Bloch gave to Google,
he did
not insert a similar temporary copy of the exact
RangeCheck-like code
from the existing Android arrays.java,
but assumed
somebody else would do
it, and simply handed them the
same
file he had given to Sun, including 9
lines that were Sun's
private code. This infringes Suns copyright in those
9
lines (Jury so ruled, Google is guilty under the law for
this). Problem was
that nobody did that work, and the 9
wrong lines shipped in millions of phones.
This was no
benefit to Google, as any functional
difference between the sort
error messages in Sun
arrays.java and Android arrays.java would thus be
changed,
thus changing the Android API in a potentially incompatible
way
(precisely what Bloch was trying to avoid in his
contribution to Sun). So
copying the 9 lines RangeCheck was
either harmful to the reliability and
performance of Android
or had no effect at all.
- In 2010 or 2011,
Oracle/Sun spotted the 9 lines that
that should have been replaced before the
2009 release and
told Google (in the most impolite way possible). So Google
looked closely at the code and replaced it by something
belonging to Google,
possibly the code from the original
Android arrays.java, perhaps by new and
different code.
However the technical harm of putting the Sun
RangeCheck
behavior in the Android TimSort, where the Android 1.1 sort
error
logic should have been is strong proof of 0
benefit,
especially (I have
not checked the various file versions in
git, sorry) if it can be shown that
the replacement code now
in Android restores the original Android error
handling,
fixing the bug introduced by those 9 wrong lines.
Note: There
may have been some information that the
original
RangeCheck was written by
Bloch back in the day, but this
doesn't really matter.
[ Reply to This | Parent | # ]
|
