[
https://issues.apache.org/jira/browse/DERBY-3652?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Rick Hillegas updated DERBY-3652:
---------------------------------
Attachment: derby-3652-02-aa-dontWidenExceptForSmalllint.diff
Attaching derby-3652-02-aa-dontWidenExceptForSmallint.diff. This removes code
which was matching small numeric args to bigger Java types. For instance, SQL
INTEGER was able to match Java long. I removed this bad matching logic for
INTEGER, BIGINT, REAL, and DOUBLE.
When I removed the corresponding logic for SMALLINT, method resolution broke
when CleanDatabaseTestSetup tried to call the system procedure
SYSCS_INPLACE_COMPRESS_TABLE. According to the Reference Manual, the numeric
arguments of this procedure are typed as SMALLINT. According to the ANSI rules,
those args can match Java args of type short or java.lang.Integer. The Java
signature for that method, however, presents int as the type of its numeric
args. I think that the solution may be to redeclare the signatures of various
Java methods in SystemProcedures so that they will match, under the ANSI rules,
with the signatures described in the Reference Manual.
In addition, the logic in ClassInspector.classConvertableFromTo() looks like it
may match SQL character types to Java numeric types. This would violate the
ANSI rules too. This needs to be looked into later.
The signature test runs cleanly for me standalone. I will run the complete
regression suites and see what breaks. Touches the following files:
M java/engine/org/apache/derby/iapi/services/loader/ClassInspector.java
Changes ClassInspector.classConvertableFromTo() to not match smaller numeric
types to larger numeric types.
M
java/testing/org/apache/derbyTesting/functionTests/tests/lang/AnsiSignaturesTest.java
Uncommented most of the tests for signatures which are not supposed to find any
matches.
> Derby does not follow the SQL Standard when trying to map SQL routines to
> Java methods.
> ---------------------------------------------------------------------------------------
>
> Key: DERBY-3652
> URL: https://issues.apache.org/jira/browse/DERBY-3652
> Project: Derby
> Issue Type: Bug
> Components: SQL
> Affects Versions: 10.5.0.0
> Reporter: Rick Hillegas
> Attachments: badsignatures.sql,
> derby-3652-01-aa-mixTypesOnFirstPass.diff,
> derby-3652-01-ab-mixTypesOnFirstPass.diff,
> derby-3652-01-ac-mixTypesOnFirstPass.diff,
> derby-3652-01-ad-mixTypesOnFirstPass.diff,
> derby-3652-02-aa-dontWidenExceptForSmalllint.diff,
> derby-3652-badmatches.diff, SignatureMapping.html, SignatureMapping.html,
> SignatureProblems.java, signatureProblems.sql
>
>
> I have only tested this in the 10.5 trunk. However, I suspect that this
> affects all previous releases of Derby as well.
> In resolving method signatures for function/procedure invocations, the SQL
> standard makes the following definitions in part 13, section 4.5 (parameter
> mapping). These definitions, in turn, refer to tables B-1 and B-3 in JDBC 3.0
> Specification, Final Release, October 2001 ([JDBC]).
> * Simply mappable - This refers to the correspondence of SQL and Java
> types described in [JDBC] table B-1. This is the table which defines the
> mapping of SQL types to Java primitives.
> * Object mappable - This refers to the correspondence of SQL and Java
> types described in [JDBC] table B-3. This is the table which defines the
> mapping of SQL types to Java wrapper objects.
> * Output mappable - For OUT and INOUT parameters, this refers to a single
> element array whose cell is simply mappable or object mappable. E.g.
> Integer[] or float[].
> * Mappable - This means simply, object, or output mappable.
> * Result set mappable - This means a single element array whose cell is a
> type which implements either java.sql.ResultSet or
> sqlj.runtime.ResultSetIterator.
> Putting all of this together, section 4.5 continues:
> "A Java method with M parameters is mappable (to SQL) if and only if, for
> some N, 0 (zero) <= N <= M, the data types of the first N parameters are
> mappable, the last M - N parameters are result set mappable, and the result
> type is either simply mappable, object mappable, or void."
> Section 8.6 gives more detailed rules, but they are hard to follow. According
> to section 8.6, when resolving a routine invocation, Derby should expect to
> find one and only one static mappable method with the expected external name
> (Java class + method name).
> I believe that this is a fair description of the rules. This, at least, is
> what some other databases appear to do. See, for instance,
> http://infocenter.sybase.com/help/index.jsp?topic=/com.sybase.help.ase_15.0.java/html/java/java126.htm
> and
> http://www.service-architecture.com/database/articles/mapping_sql_and_java_data_types.html
> We do not have a regression test which verifies that Derby applies the SQL
> standard resolution rules. There may be several divergences from the
> standard. This JIRA is a place to track those discrepancies. Here is one that
> I have noticed:
> The following SQL signature
> ( a int ) returns int
> should be mappable to any of the following Java signatures
> public static int f( int a )
> public static int f( Integer a )
> public static Integer f( int a )
> public static Integer f( Integer a )
> However, I observe that Derby is only able to resolve the first and third
> signatures (the ones with primitive arguments). I will attach a test case
> showing this problem.
> I will also attach an html table summarizing the simply and object mappable
> rules.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
