On the contrary to other DBMS, MySQL supports schemas, but as databases.
Which means that JDBCConnection gets from the JDBC driver that MySQL does
not support schemas and so it tries to access TAP_SCHEMA tables with a
'TAP_SCHEMA_' prefix instead of 'TAP_SCHEMA.' one. Since MySQL does
behave like it supports schemas, JDBCConnection.supportsSchema must be
set to 'true' if the DBMS is MySQL.

Besides, since no schemas are returned by the JDBC driver
(cf DatabaseMetadata.listSchema(...)), the function
JDBCConnection.isSchemaExisting(String, DatabaseMetaData) must always assume
that the specified schema exists if the DBMS is MySQL. This is particularly
important when the existence of 'TAP_UPLOAD' must be checked, because if the
function returns 'false' the library will attempt to create the database/schema
'TAP_UPLOAD' and will obviously fail due to a lack of permissions. Of course,
it means that the TAP implementor must create manually the schema/database
'TAP_UPLOAD' him-/her-self.

The second particularity of MySQL is the quote character for identifiers.
It is a back-quote (`) instead of a double quote ("). To reflect this
difference, JDBCTranslator.appendIdentifier(...) has been overwritten in a new
JDBCTranslator extension: MySQLTranslator.

The translation of all SQL types and mathematical functions have been adapted to
MySQL according to the online documentation. Few tests have been done locally
with a tiny database. This seems to work but further testing should be
performed in order to ensure the stability of this implementation.

This infinite loop occured only when the replacement object is just
a wrapping of the matching object ; after replacement, the new object was
inspected for matching objects.

Example: infinite loop if we want to wrap all foo(...) functions with
         the function ROUND in the following query:
    SELECT foo(foo(123)) FROM myTable
	     Expected result:
    SELECT ROUND(foo(ROUND(foo(123)))) FROM myTable
	     But generated result was:
    SELECT ROUND(ROUND(ROUND(......foo(foo(123))))) FROM myTable

See the test case TestDBChecker.testClauseADQLWithNameNull() for more details.

since the commit 8e2fa9ff.

(005fc622). The position of a TokenMgrError
was not there anymore.

(https://github.com/gmantele/taplib/commit/7a70c6038cef460ab169682bed391bb5ae1de1e9)

It was not possible to use a GROUP BY with a qualified column name.
So finally, now, a GROUP BY is a ClauseADQL<ADQLColumn> instead of
a ClauseADQL<ColumnReference>. Indeed, according to the ADQL's BNF,
GROUP BY items are only columns as they would appear in the SELECT
clause (i.e. qualified or not, delimited or not). On the other
hand an ORDER BY accepts ONLY column index or non-qualified column
name/alias.

The class ColumnReference is kept for backward compatibility (or in
case the next version of the ADQL grammar make items of GROUP BY and
ORDER BY of the same type: index or qualified column). Besides, this
class is still inherited for the ORDER BY clause items
(see adql.query.ADQLOrder).

a qualified column name should be allowed, but still no column index should be.

This error has been raised on the issue #32 by Zarquan.

Two classes have been modified after compilation:
    - ADQLParser - a simple cast for one of the automatically generated
                   constructor.
    - ParseException - the token position has been stored for better
                       syntax error messages.
A note has been added at the top comment for both files to highlight
the modified parts and how to restore them after re-generation.

This import can not be resolved in a standalone version of adql. It
also creates a circular dependency: ADQL then depends on TAP, which
depends on starjava, but starjava depends on ADQL.

the automatic datatype detection was missing for CENTROID functions.
--
Additionally, some JUnit test files of the `adql` package has been moved to
the correct location.

The "normal" JOIN:
    A JOIN B ON A.id = B.id JOIN C ON B.id = C.id
is correctly interpreted as:
    ( (A JOIN B ON A.id = B.id) JOIN C ON B.id = C.id )
But with a NATURAL JOIN, the tree is mirrored:
    A NATURAL JOIN B NATURAL JOIN C
gives:
	( A NATURAL JOIN (B NATURAL JOIN C) )
instead of:
    ( (A NATURAL JOIN B) NATURAL JOIN C )
This is not a problem when the SQL translation is identical to the ADQL
expression, but for some DBMS a conversion into a INNER JOIN ON is necessary
and in this case we got the following SQL:
    A JOIN B JOIN C ON A.id = B.id ON B.id = C.id
Which seems to work, but is syntactically strange.

This commit should fix the generated tree. A "normal" JOIN and a NATURAL JOIN
should now have the same form. A JUnit test has been added into TestADQLParser
to check that: testJoinTree().

This is particularly annoying for numeric functions like sqrt(...)
(example query: SELECT TOP 1 sqrt(2) as s2 FROM whatever_table).
This bug has been raised by M. Taylor.

These two columns let recommend an order, respectively, for tables in their
schema, and columns in their table.
In addition of these new columns, "arraysize" has been also added.
All these new columns are already supported when creating a TAP_SCHEMA from
an XML file.

sub-query. Without this information, it was impossible to resolve columns makingreference to sub-queries of the clause FROM. See the JUnit test case for a
concrete example.
(error raised by Hendrik Heinl - ARI/GAVO)

column in TAP_SCHEMA. Such names are normalized (without any prefix and without
double quotes) when getADQLName() is called, but are returned just trimmed
by getRawName(). This latter is just used by TAPMetadata when writing the XML
description of all TAP tables.

in DBType.DBDatatype (for UNKNOWN and UNKNOWN_NUMERIC). This reset is performed
after each JUnit setting a special custom value (otherwise it prevents other
JUnit to run correctly)

clause SELECT.

[ADQL] SqlServerTranslator: bug fix for TOP vs LIMITgrammar. Adding support for math functions, including translations for TRUNCATE and MOD non-standard implementation.

Before correction, if an ADQlObject (e.g. a function or a sub-query) contains
another ADQLObject and that both (i.e. parent and child) are matching in a
SimpleReplaceHandler and are asked to be replaced, only the parent
seemed to have been replaced. However, the child has been replaced, but
in the former instance of the parent ; and so its replacement is not
visible in the final query.

For instance:
if all mathematical functions must be replaced by a dumb UDF named 'foo' in
the ADQL query:
        "SELECT sqrt(abs(81)) FROM myTable"
,the result should be:
        "SELECT foo(foo(81)) FROM myTable"
,but before this correction it was:
        "SELECT foo(abs(81)) FROM myTable".

This function was originally static. Since no other classes benefits of this
static state, this function is no longer static. This might however raise a
small incompatibility for the users of the library which used it.

Before this commit, unsigned integers were also allowed (as for ORDER BY).

Because of a test of the first parameter (non existing for functions
like PI() and RAND()), a NullPointerException was thrown and so
was stopping brutally an ADQL query interpretation.

deals with NATURAL JOINs and JOINs using the keyword USING so that being
supported by SQL Server. Basically, they are translated as a list of ON
conditions.
Warning: This translator is just guaranteed to solve the NATURAL and USING
issue. Support for datatypes conversion and case sensitivity has to be
reviewed. Besides no geometrical function is translated for SQL Server.

in first position. This should be totally harmless and should be conform to
the behaviour of a NATURAL or USING keywords in DBMS.

This is easily possible for concatenations, string constants and User Defined
Functions having a FunctionDef. A new special datatype was needed for
numeric functions and operations: UNKNOWN_NUMERIC. This special type
can not be set with FunctionDef.parse(...) and it behaves exactly like the type
UNKNOWN, except that DBType.isNumeric() returns true (as .isUnknown()).
Thus, while writing the metadata of a result in TAP, nothing changes:
an UNKNOWN_NUMERIC type will be processed similarly as an UNKNOWN type:
to use the type returned from the database ResultSet or to set VARCHAR.
(no modification of TAP was needed for that)

Functions whose some parameters are another function were not correctly
identified: since the inner functions were not yet identified, their type
was UNKNOWN and so it makes the identification of the parent function
much easier since an UNKNOWN parameter is not checked. But, it was a
problem if the parameter occurs to be finally of the wrong type.

(indeed, a Concatenation object has no name).

functions must be casted into NUMERIC. Otherwise Postgres rejects the query.

Now, all occurences of ' are replaced by '' in the SQL translation.

types. The notion of "unknown type" is different in function of the target
object:
  - a DBType and a FunctionDef have an unknown type if their function
    isUnknown() returns true. In such case, the other functions such as
	isNumeric/String/Geometry() will return false.
  - an ADQLOperand (e.g. ADQLColumn) does NOT have a isUnknown() function.
    But if the type of the operand is unknown, its functions isNumeric(),
	isString() and isGeometry() must ALL return true. Otherwise, just one of
	these functions can return true.