I just noticed a hashCode/equals violation that we created a few revisions ago. We compute the hash of the validBits in
ICM, but we only compare validBits up to the number of colors in the colormap. One could construct two ICMs that have
different validBits that are identical in the first N bits (N = number of colors), but have different bits beyond that,
and those 2 ICMs would evaluate as equals(), but their hashcodes would be different.
Possible solutions:
- Truncate validBits when it is accepted in the constructor (validBits.and(...))
- Manually compute the hash of the first N bits of validBits
- Not use validBits in the hash code since it is allowed to omit significant
data from the hash
(Note that everything in hashcode must participate in equals(), but not vice
versa)
The same is true of nBits in ColorModel. (nBits can be copied and truncated
with Arrays.copyOf)
The same is *not* true of maskBits in PCM since the constructor creates an
array of the precise length it needs...
...jim
On 6/2/16 7:07 AM, Jayathirth D V wrote:
Hi Phil,
I have updated the code with all the changes I mentioned previously. I am
caching hashCode when first time hashCode() is called.
Please find the updated webrev for review:
http://cr.openjdk.java.net/~jdv/7107905/webrev.09/
Thanks,
Jay
-----Original Message-----
From: Philip Race
Sent: Wednesday, June 01, 2016 10:06 PM
To: Jayathirth D V
Cc: Jim Graham; 2d-dev@openjdk.java.net
Subject: Re: Review Request for JDK-7107905: ColorModel subclasses are missing
hashCode() or equals() or both methods
Please post the updated webrev.
-phil.
On 6/1/16, 12:02 AM, Jayathirth D V wrote:
Hi Phil& Jim,
Collating all the changes needed:
1) I have removed both equals()& hashCode() in CCM but forgot to add the file
while creating webrev. I will include changes in CCM in updated webrev.
2) Caching of hashCode value in IndexColorModel& PackedColorModel :
We can cache hashCode value when constructor is called or when
hashCode() is called for first time. What approach we have to follow?
3) Comment section of equals() method, I will update it to :
1449 * the target object must be the same class (and not a
subclass) as this
4) I will use .equals() to compare colorSpace values in CM.equals() so it will
be like we are not intending ColorSpace class to never override equals() method.
Please let me know your inputs.
Thanks,
Jay
-----Original Message-----
From: Jim Graham
Sent: Wednesday, June 01, 2016 4:14 AM
To: Phil Race
Cc: Jayathirth D V; 2d-dev@openjdk.java.net
Subject: Re: Review Request for JDK-7107905: ColorModel subclasses are
missing hashCode() or equals() or both methods
I think we should use .equals() here, otherwise it looks like a recommendation
that the intent is for those classes to never implement it...
...jim
On 05/31/2016 03:31 PM, Phil Race wrote:
I don't know of any design intent, so yes, I meant more as a
practical matter.
Unless they subclass then using equals() which I thought unlikely it
makes no difference here.
But it would be proof against that to use equals, unlikely to matter
as it might be ..
-phil.
On 05/31/2016 03:19 PM, Jim Graham wrote:
On 05/31/2016 02:50 PM, Phil Race wrote:
On 05/31/2016 12:20 PM, Jim Graham wrote:
Hi Jay,
You were going to remove hashCode/equals from CCM, but instead you
simply removed it from the patch. You still need to edit it to
remove the existing methods.
Oh yeah ! CCM is gone from the latest webrev. I expect that was not
intentional.
Also, I'm not sure == is a good way to compare ColorSpace
instances
- Phil?
Neither ColorSpace nor ICC_ColorSpace over-ride equals or hashCode
and all the predefined ones are constructed as singletons so it
seems unlikely to cause problems
Should it use .equals() on principle, though? Otherwise we are
baking in the assumption that it doesn't implement equals() into our
implementation of the CM.equals() method. Might it some day
implement equals (perhaps it isn't a practical issue today, but it
might be in the future when we forget that it was omitted in this
usage we create today).
I guess what I'm asking is if it is a design feature that different
objects are considered non-equal (such as an object that, for
instance, has only predetermined instances that are shared by
reference and reused). I think color spaces are sort of in-between
in that we define a few constants that simply get used by reference
in 99% of cases, but that isn't a design feature, more like a
practical issue...
...jim