[#6004] fix: use fullName instead of names.get(0) when get role

[cool9850311]

What changes were proposed in this pull request?
use fullName instead of names.get(0) when get role Why are the changes needed?
Fix: #6004

Does this PR introduce any user-facing change?
NO

How was this patch tested?
existing ut

[tengqm]

move 54-56 before 53 so that this is easier to understand?

[jerqi]

Could you add a UT for this?

[cool9850311]

Could you add a UT for this?

This part is covered by testMetaLifeCycleFromCreationToDeletion in TestJDBCBackend, do i have to add other ut for this?

[jerqi]

Could you add a UT for this?

This part is covered by testMetaLifeCycleFromCreationToDeletion in TestJDBCBackend, do i have to add other ut for this?

Yes, you need to add a new test case. If the fullName contains ., legacy implement will produce an error result. The legacy test case doesn't cover this.

[jerqi]

role.test is a legal name.

[jerqi]

You should add the test case in TestJdbcBackend instead of this place.

[cool9850311]

But it does throw the IllegalArgumentException error?

[jerqi]

It won't throw IllegalArgumentException.

[jerqi]

You should create the role metadata object by

MetadataObjects.of(null, "role.test", ROLE);

[jerqi]

The name with dot should be legal name for role instead of illegal name. It shouldn't throw IllegalArgumentException.

[cool9850311]

Oh, I see, so I need to change the legacy implement code?

[jerqi]

Yes.

[cool9850311]

@jerqi
Is this correct?
just test the method with and without dot

[cool9850311]

public static MetadataObject of(List<String> names, MetadataObject.Type type) {
    Preconditions.checkArgument(names != null, "Cannot create a metadata object with null names");
    Preconditions.checkArgument(!names.isEmpty(), "Cannot create a metadata object with no names");
    Preconditions.checkArgument(
        names.size() <= 4,
        "Cannot create a metadata object with the name length which is greater than 4");
    Preconditions.checkArgument(type != null, "Cannot create a metadata object with no type");

    Preconditions.checkArgument(
        names.size() != 1
            || type == MetadataObject.Type.CATALOG
            || type == MetadataObject.Type.METALAKE
            || type == MetadataObject.Type.ROLE,
        "If the length of names is 1, it must be the CATALOG, METALAKE, or ROLE type");

    Preconditions.checkArgument(
        names.size() != 2 || type == MetadataObject.Type.SCHEMA,
        "If the length of names is 2, it must be the SCHEMA type");

    Preconditions.checkArgument(
        names.size() != 3
            || type == MetadataObject.Type.FILESET
            || type == MetadataObject.Type.TABLE
            || type == MetadataObject.Type.TOPIC
            || type == MetadataObject.Type.MODEL,
        "If the length of names is 3, it must be FILESET, TABLE, TOPIC or MODEL");

    Preconditions.checkArgument(
        names.size() != 4 || type == MetadataObject.Type.COLUMN,
        "If the length of names is 4, it must be COLUMN");

    for (String name : names) {
      checkName(name);
    }

    return new MetadataObjectImpl(getParentFullName(names), getLastName(names), type);
  }

I am kinda confused now
So what is the Rule of Role name?
Does it have length limit?
For current impl
if Role name contains dot
insertRelation will throw IllegalArgumentException, cause it will run through MetadataObjects.of(null, "role.test", ROLE); and above checks and conflict with "must be the * type" Preconditions
insert and other methods will not throw any exception

[jerqi]

You can modify the logic MetadataObjects.of(null, "role.test", ROLE).
You would better modify the logic of MetadataObjects.parse.

[cool9850311]

Ok

[cool9850311]

@jerqi
could you review the new commit, Thanks.