|
Authored by: bugstomper on Saturday, August 04 2012 @ 07:29 PM EDT |
This was in the context of a rule 50(b) motion, which is about what a reasonable
jury could decide based on the evidence that was before it. Google could only
argue based on what witnesses actually said and did not say. They had Bloch's
testimony that rangeCheck did not increase performance of the running system.
They had Oracle's lack of testimony to contradict that. They had no evidence
that was before the jury that said that 2600 bounds checking calls that did not
signal an error would be any slower than skipping the calls, nor any other
testimony to indicate that Android was better off without rangeCheck. They did
include as close to that as they could from what was in evidence - That
Android's introduction and later success both happened without rangeCheck being
in the code.
[ Reply to This | Parent | # ]
|
|
Authored by: argee on Saturday, August 04 2012 @ 08:32 PM EDT |
Range Check is not called 2600 times at boot. It cannot be
called at all because it is not an API function. It cannot
be used outside of Timsort. If Timsort is called, it might
use RangeCheck a number of times.
---
--
argee[ Reply to This | Parent | # ]
|
|
Authored by: Anonymous on Sunday, August 05 2012 @ 12:06 PM EDT |
Agreed there are debates about if, and to what extent,
library methods should protect themselves.
It's also not uncommon to have multiple versions of the
library, often along the lines of "debug" and "production"
where the debug version does all kinds of self-protection
(and programmer protection) but suffers performance
penalties, and production code that does minimal, if any
protective tests in order to maximize performance.
Testing that the upper bound is more than the lower bound is
one of the programmer protections, while testing that a
particular index is within the upper and lower bounds
inclusive would be a reasonable production test.[ Reply to This | Parent | # ]
|
|
|
|
|