Details
-
Bug
-
Status: Closed
-
Critical
-
Resolution: Fixed
-
6.5.0
-
Untriaged
-
Centos 64-bit
-
No
Description
According to the design doc, niql number functions are supported in adv filtering, but ASIN and ACOS functions are not working as expected.
https://docs.google.com/document/d/19PkXWG8ovTS-E7l2Fqg_kRwsPTh_wHhjrrFBFPOWzQs/edit?pli=1#heading=h.jue93z4nyvp5
However, ATAN() in adv filter matches niql results.
doc:
{
|
"string_short": "x",
|
"string_medium": ")nx.\n\tPfDEEptfV2o",
|
"int": 93,
|
"float": 16.770881608010846,
|
"string_long": "KfdVIYXB4Hk@uA9ZrChdOGP\r^lEd/l lQaAEoqGTfznioxXTsxlDy\fkBWmIiAe\nRPqm6.-\\RK%ZE&WzSn)rvt_IJi",
|
"string_vlong": "egof$\ryps<QmEaQhg",
|
"bool": false
|
}
|
working filter (the doc matches this filter and is replicated to target):
ATAN(int)<>3.05682983181e+307
not working (doc matches filter but is NOT replicated to target):
ASIN(int)<>3.05682983181e+307
ACOS(int)<>3.05682983181e+307
Following queries return 1 when the above doc is present in src:
SELECT COUNT(*) FROM default WHERE 'ACOS(int)<>3.05682983181e+307'
|
SELECT COUNT(*) FROM default WHERE 'ASIN(int)<>3.05682983181e+307'
|
Attachments
Issue Links
Activity
Golang output the imaginary number result as "NaN", which stands for "Not-A-Number".
I think it is fair for gojsonsm matcher to check for NaN.
per Oct 14 email from John L, this is on track to be fixed by Nov 11. Adding label "approved-for-Mad-Hatter"
Added due date field (preset to Nov 15). Please update the due date to the current ETA for a fix. Thanks.
https://github.com/couchbaselabs/gojsonsm/pull/119 is now in gojsonsm repository. Build automation should kick in soon with a new build
Build couchbase-server-6.5.0-4754 contains gojsonsm commit abdb374 with commit message:
MB-36088 - Refactored much of data type comparisons and added N1QL-like collate to fast matcher
Build couchbase-server-6.5.0-4754 contains gojsonsm commit 0e8f997 with commit message:
MB-36088 - implement NaN float comparisons
Build couchbase-server-7.0.0-1034 contains gojsonsm commit abdb374 with commit message:
MB-36088 - Refactored much of data type comparisons and added N1QL-like collate to fast matcher
Build couchbase-server-7.0.0-1034 contains gojsonsm commit 0e8f997 with commit message:
MB-36088 - implement NaN float comparisons
Verified this behavior on 6.5.0-4821:
Any NaN (not-a-number) float values will be considered less than any other real number, and two NaNs will not yield equality. The only exception is the operators “<=” and “>=” would break this paradigm and return “true”. This is different from golang’s standards where NaNs are not equal when compared regardless of function. Applications using gojsonsm such as XDCR can check for collation.
Just observed that when int = 0 or 1, docs are getting replicated to target. So the problem occurs only when int>1.
https://s3.amazonaws.com/cb-engineering/Pavithra/collectinfo-2019-09-20T165550-ns_1%40172.23.105.133.zip
https://s3.amazonaws.com/cb-engineering/Pavithra/collectinfo-2019-09-20T165550-ns_1%40172.23.106.121.zip