asterixdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mike Carey <dtab...@gmail.com>
Subject Re: Issue status for indexing open data? (@Ildar?)
Date Mon, 10 Aug 2015 19:21:22 GMT
Cool; thx!

On 8/10/15 12:12 PM, Ildar Absalyamov wrote:
> Submitted version has a regression in optimizer tests. I know what is the
> issue and will fix that tonight.
>
> 2015-08-10 10:44 GMT-07:00 Mike Carey <dtabass@gmail.com>:
>
>> Thx!
>> On Aug 9, 2015 11:15 PM, "Ildar Absalyamov" <ildar.absalyamov@gmail.com>
>> wrote:
>>
>>> It is used to deserialize the proper type, however this is needed only in
>>> open index case.
>>> I get around by avoiding storing any type-related information in the
>> index
>>> metadata, unless it is opened.  In closed case the type would be picked
>>> from the record metadata, thus getting nullability information and not
>>> replicating it anywhere.
>>> I created an uploaded a new patch set, feel free to check and prove it if
>>> everything else is OK.
>>>
>>>> On Aug 9, 2015, at 00:54, Mike Carey <dtabass@gmail.com> wrote:
>>>>
>>>> Q:  Is the nullability info drawn from the index entry really used?
>> Can
>>> you give an example of where/how? (Wondering if it's perhaps removable.)
>>>> On 8/7/15 2:39 PM, Ildar Absalyamov wrote:
>>>>> So it turned out I was wrong about about open indexes & nullable
>> types.
>>> Is
>>>>> is indeed not allowed that, is you look closely on parser's production
>>> rules
>>>>> The problem appears when we're trying to create an ordinary index on
>>>>> nullable field. The information about the field type is persisted in
>> the
>>>>> index metadata entry. In the open case the type is provided by the
>> user,
>>>>> whereas in the closed the type is extracted from the record's metadata
>>>>> entry (where we do have a nullability attribute). We can replicate the
>>>>> nullablility attribute in the index metadata entry as well, but I
>> think
>>>>> that is against the Till's original comment since this information is
>>>>> becoming spread around several places. Other solution would be to
>> carry
>>> "if
>>>>> (open) {} else if (closed) {}" logic throughout the code, where we do
>>> need
>>>>> nullability information, but that will be ugly.
>>>>>
>>>>> 2015-08-07 11:18 GMT-07:00 Michael Carey <mjcarey@ics.uci.edu>:
>>>>>
>>>>>> Q: (@Ildar) - What's the status of removing "?" from the open index
>> DDL
>>>>>> and reverting that (unnecessary) part of the metadata patch?
>>>>>>
>>>>>
>>> Best regards,
>>> Ildar
>>>
>>>
>
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message