[
https://issues.apache.org/jira/browse/ATLAS-1062?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
David Radley updated ATLAS-1062:
--------------------------------
Description:
Comments on the Type architecture as documented in the latest build:
1) it would have been helpful to see the new version of the UML to see how the
concepts formally relate to each other. I wonder why this has been removed and
not updated
2) I am not sure why we need the concept of a metatype and a supertype. On the
face of it these words mean the same thing. I think a uml diagram would help
here. It seems to me that type names are conceptually enumerations - I suggest
including this in the UML. In the example :
Name: hive_table
MetaType: Class
SuperTypes: DataSet
Is DataSet a Class ?
It might be worth thinking about documenting the meta model and the model. I
have some experience of doing this so could work with you on this.
3) I suggest in addition to comment - have notes - which would be a time
ordered list of dated, authored comments
4) I notice in your example you have
typeName: “hive_table”
values:
name: “customers”
...
tableType: “MANAGED_TABLE”
I wonder if you could fold table Type and typeName together. Either having 2
typeNames or having a typename of "managed_hive_table". Or have a namedObject
super class that everyone inherits from.
5) I see that typeName= "hive_table" looks like a REST reply - shouldn't we be
returning a href to the type object -so the REST caller could navigate to it.
6) It would be useful to explain what a reverseAtributename is.
7) I notice attributes have datatypeName but hive_table has typeName. I suggest
we use the same wording to refer to types.
was:
I look at the Type architecture and I see that many elements have a name - it
seems there should be a higher abstraction for named things.
I suggest the highest thing IDataType in the Types architecture diagram have :
- a manditory name
- a manditory global unique identifier
- an optional short and long description
- a curator / owner
- associated notes (notes are free form text that could be added we track,
author, date , version if we allow notes to be updated).
It would make sense to have a higher level object that contains these
attributes, where IDataType is a subtype and a Term interface are subtypes.
Summary: Type architecture feedback (was: suggested model improvement)
> Type architecture feedback
> --------------------------
>
> Key: ATLAS-1062
> URL: https://issues.apache.org/jira/browse/ATLAS-1062
> Project: Atlas
> Issue Type: Improvement
> Reporter: David Radley
>
> Comments on the Type architecture as documented in the latest build:
> 1) it would have been helpful to see the new version of the UML to see how
> the concepts formally relate to each other. I wonder why this has been
> removed and not updated
> 2) I am not sure why we need the concept of a metatype and a supertype. On
> the face of it these words mean the same thing. I think a uml diagram would
> help here. It seems to me that type names are conceptually enumerations - I
> suggest including this in the UML. In the example :
> Name: hive_table
> MetaType: Class
> SuperTypes: DataSet
> Is DataSet a Class ?
> It might be worth thinking about documenting the meta model and the model. I
> have some experience of doing this so could work with you on this.
> 3) I suggest in addition to comment - have notes - which would be a time
> ordered list of dated, authored comments
> 4) I notice in your example you have
> typeName: “hive_table”
> values:
> name: “customers”
> ...
> tableType: “MANAGED_TABLE”
> I wonder if you could fold table Type and typeName together. Either having 2
> typeNames or having a typename of "managed_hive_table". Or have a namedObject
> super class that everyone inherits from.
> 5) I see that typeName= "hive_table" looks like a REST reply - shouldn't we
> be returning a href to the type object -so the REST caller could navigate to
> it.
>
> 6) It would be useful to explain what a reverseAtributename is.
> 7) I notice attributes have datatypeName but hive_table has typeName. I
> suggest we use the same wording to refer to types.
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
