For real numbers that cannot be represented as a binary number, we use a
double-precision representation internally. There will some noise in the
least significant bits of the double value if the original value cannot be
represented as a binary number. This shouldn't matter too much in reality,
however, it seems that the noise gets multiplied while parsing the
individual digits of the input. This is a bug and a fix is currently in
progress.
Best regards
Jan
Am Samstag, 2. Juli 2016 17:42:59 UTC+2 schrieb Frank Celler:
>
> The pattern is basically: can a number represented as binary? For example
> 0.25 = 1/4. Will try to find out, why it is not using a nicer
> representation as decimal.
>
> Am Samstag, 2. Juli 2016 16:14:19 UTC+2 schrieb Scott B.:
>>
>> Frank,
>>
>> I see the issue without the use of the Java driver at all (but also with
>> it). I see it when entering data directly into the ArangoDB web interface.
>> I tried another document for testing purposes, and there is some kind of
>> pattern, but I'm not sure what it is. I added another simple test document
>> to a collection in both Arango 3.0.1 and Arango 2.8.9 via their web
>> interfaces, and then retrieved the saved documents, with different results.
>>
>> Data entered:
>>
>> {
>> "first": 0.43,
>> "second": 0.32,
>> "third": 0.2,
>> "fourth": 0.25,
>> "fifth": 0.26,
>> "sixth": 0.47,
>> "seventh":0.3,
>> "eighth":0.95
>> }
>>
>> On ArangoDB 2.8.9, a subsequent retrieval results in the following data:
>>
>> {
>> "first": 0.43,
>> "second": 0.32,
>> "third": 0.2,
>> "fourth": 0.25,
>> "fifth": 0.26,
>> "sixth": 0.47,
>> "seventh": 0.3,
>> "eighth": 0.95
>> }
>>
>> But on Arango 3.0.1, I get extra digits added to some of the numbers:
>>
>> {
>> "eighth": 0.9500000000000001,
>> "fifth": 0.26,
>> "first": 0.43000000000000005,
>> "fourth": 0.25,
>> "second": 0.32000000000000006,
>> "seventh": 0.30000000000000004,
>> "sixth": 0.47000000000000003,
>> "third": 0.2
>> }
>>
>> It isn't just .33 or .66/.67. I'm not sure I could actually describe a
>> pattern. I thought is the number ended in "5" it was fine, but then look
>> at how 0.95 was saved.
>>
>> I did try saving a simple document using arangosh, and it seems to work
>> as expected on 3.0.1. However, as soon as I open that document, and a
>> property to it, and then re-save it through the web interface, it
>> immediately adds extra digits to the previously correct numbers in the
>> document.
>>
>> So in short, the web UI and the Java drivers both seem to add extra
>> digits to some floating point numbers on Arango 3.0.1. Arangosh seems to
>> not have that issue.
>>
>> All three seems to work fine on Arango 2.8.9.
>>
>> On Saturday, July 2, 2016 at 5:01:57 AM UTC-6, Frank Celler wrote:
>>>
>>> In general, there is an issue with 1/3, because it cannot to represented
>>> as binary number. ArangoDB uses some standard conversion that produces the
>>> "expected" representation as "0.33". For example, if you use the
>>> "arangosh", you should see the expected representation. Could you check
>>> this with your dataset? I. e. does
>>>
>>> arangosh> db.collectionname.toArray()
>>>
>>> show 0.33 and 0.66? If it is showing 0.33000000000000007, it is possible
>>> that there are changes in Java driver causing this. I have to check with my
>>> colleagues.
>>>
>>>
>>>
>>> Am Samstag, 2. Juli 2016 05:10:15 UTC+2 schrieb Scott B.:
>>>>
>>>> My apologies, the above makes it look like the value 0.33 in the array
>>>> was changed to 0.6600000000000001. I ran several tests and copied and
>>>> pasted the wrong results. It actually ends up as:
>>>>
>>>> {
>>>> "first": 0.3300000000000001,
>>>> "second": [
>>>> 0.6700000000000004,
>>>> 0.5,
>>>> 0.33000000000000007
>>>> ]
>>>> }
>>>>
>>>> On Friday, July 1, 2016 at 8:31:34 PM UTC-6, Scott B. wrote:
>>>>>
>>>>> Is anyone else seeing odd behavior with ArangoDB 3.0.1 regarding
>>>>> floating point (Double) numbers?
>>>>>
>>>>> I had something odd pop-up in my application after moving to ArangoDB
>>>>> 3.0.1. JSON objects with decimal numbers (like 0.67) were suddenly
>>>>> getting
>>>>> extra precision added to them. To eliminate as many variables as
>>>>> possible,
>>>>> I saved the following very simply JSON document via the web interface
>>>>> (add
>>>>> document to collection) on both ArangoDB 2.8.9 and ArangoDB 3.0.1:
>>>>>
>>>>> {
>>>>> "first": 0.33,
>>>>> "second": [
>>>>> 0.67,
>>>>> 0.33,
>>>>> 0.5
>>>>> ]
>>>>> }
>>>>>
>>>>> On ArangoDB 2.8.9, it is saved exactly as expected. However, on
>>>>> ArangoDB 3.0.1, it gets saved as:
>>>>>
>>>>> {
>>>>> "first": 0.33000000000000007,
>>>>> "second": [
>>>>> 0.6700000000000002,
>>>>> 0.5,
>>>>> 0.6600000000000001
>>>>> ]
>>>>> }
>>>>>
>>>>> My ArangoDB 3.0.1 server is running on a clean Ubuntu 16.04 LTS
>>>>> install.
>>>>>
>>>>> My ArangoDB 2.8.9 server is running on Ubuntu 14.04 LTS with a lot of
>>>>> other services/software installed.
>>>>>
>>>>> I also have the same problem when saving any documents with floating
>>>>> point numbers using the Java driver to 3.0.1 (but it worked fine under
>>>>> 2.8.9).
>>>>>
>>>>>
>>>>>
--
You received this message because you are subscribed to the Google Groups
"ArangoDB" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
