Fix Jira issue #ROOT-6511 TSQLiteStatement: Missing / wrong check for column index, failing IsNull(), as reported by Oliver Freyermuth:
Finally, my colleagues managed to uncover a small issue in my TSQLiteStatement implementation:
"IsNull()" checked the column-count in the wrong way (it used the number of bound columns which is only valid in prepared inserts).
I have done the following changes to TSQLiteStatement in the attached patch:
1. In fNumPars, I now store either the "bind" columns for a prepared statement with bound parameters, or the number of returned columns for a simple select without binding. This also simplifies error handling and TSQLiteStatement::GetNumFields() now just returns the variable which is correct in both working modes.
2. All Get'ters now check the column-number consistently, as the SQLite-documentation states that "get"ting an invalid column number returns undefined results. Note that for bound parameters, SQLite behaves differently and does the error handling itself (which I already check in TSQLiteStatement).

rootcling will write out ROOT files; that does not work for libCore and RIO
which instead use rootcling_tmp.
All other libraries use rootcling -rootbuild.
Rename ROOTCLINGTMP to ROOTCLINGSTAGE1; rootcling -rootbuild is ROOTCLINGSTAGE2.

now the library names in the rootmaps are sane.
In addition, the creation of the tmva and genvector dicts was
simplified.

By Oliver Freyermuth.

Fix by Denis Bertini.

mechanism which does not involve rlibmap anymore but
rootcling instead.
Basically all the calls to the macro calling rlibmap have been
removed.

Signed-off-by: Danilo Piparo <danilo.piparo@cern.ch>
Signed-off-by: Fons Rademakers <Fons.Rademakers@cern.ch>

which happens through rootcling and not anymore through rlibmap.

Signed-off-by: Danilo Piparo <danilo.piparo@cern.ch>
Signed-off-by: Fons Rademakers <Fons.Rademakers@cern.ch>

including files not supported on iOS.

By Oliver Freyemuth.

Fix by Oliver Freyermuth.

Patch by Oliver Freyermuth.

when recently valgrinding our analysis-software again, I found two leaks stemming from lost pointers to prepared PgSQL-statements in TPgSQLStatement (we just started using the TSQLStatement more often some weeks ago).
The problem is that the TPgSQLServer already prepares a PGresult and adds a pointer to it to the stmt-structure which is passed to TPgSQLStatement.
The TPgSQLStatement creates a new prepared statement in its constructor and overrides this pointer with it.
In TPgSQLStatement:Process(), this pointer is overwritten again.

To not break any existing usecases of this, the attached patch fixes this problem by first PQclear'ing the prepared result if it is not NULL. This fixes both leaks (and at the moment, I do not observe more leaks in our usage of the TPgSQLStatement).

[From Oliver Freyermuth]

Solution attached as patch:
- SetTimestamp() writes timestamps of the format "2013-06-30 05:11:01.021"
  to the database and communicates the 3-digit precision (i.e., milliseconds)
  to the user in its comment. 3 digit precision is also what the strftime /
  datetime methods of sqlite use, all other digits are ignored by those methods.
- GetTimestamp() reads such timestamps, also limiting the fraction-precision
  to 3 digits and communicating that to the user from within the comment.

Patch by Oliver Freyermuth.

Solution attached as patch:
- SetTimestamp() writes timestamps of the format "2013-06-30 05:11:01.021"
  to the database and communicates the 3-digit precision (i.e., milliseconds)
  to the user in its comment. 3 digit precision is also what the strftime /
  datetime methods of sqlite use, all other digits are ignored by those methods.
- GetTimestamp() reads such timestamps, also limiting the fraction-precision
  to 3 digits and communicating that to the user from within the comment.

Patch by Oliver Freyermuth.

Patch provided by Oliver Freyermuth.

The attached patches add the following classes:
TSQLiteResult
TSQLiteRow
TSQLiteServer
TSQLiteStatement
and corresponding integration into both the ROOT-buildsystem and cmake.
Naturally, also the macro for the plugin-manager is included.
Patches supplied for ROOT 5.34.08 and current master (as SQL-part is
rarely changing, this statement should stay valid at least for some months).

Some usage notes:

* The handed path is evaluated as URI by sqlite3, which means it can be
  a relative or absolute path and can even contain additional settings
  as "URI" parameters, which are described in the SQLite-documentation.
* Functions "{Create,Select,Drop}
  DataBase(s)", "Reload", "Shutdown" are not implemented as SQLite
  operates on one DB per file only. The implemented stub methods return
  appropriate error results.
* "TSQLiteServer::GetColumns()" is not implemented (i.e. returns error
  and value 0): This is sadly not possible via a query in SQLite by
  their API design. This also means the TableInfo-objects cannot be queried.
* The result-row-count of a TSQLiteResult is always -1, as SQLite does
  not know the number of result rows before all have been retrieved
  (a valid optimization for a file-only DBMS).

For the person reviewing this patch, please note that the indices of
bind_parameters in SQLite start from 1, while those for the result-columns
start from 0.

As this is a rather large featureset to cover and the software of our
experiment currently only uses a small subset of it (only 'reading'
statements, and no prepared statements), I also wrote a very simple
testing-tool which creates a new database with a table with 10 columns
of very different type, inserts some example data, and rereads that
data both using a simple Query() and using a prepared Statement().
This tool and a very short makefile to build it are also attached if
needed for the review-process. The tool uses the commands described
in the documentation of the abstract baseclass TSQLStatement to
simulate real-life behaviour.
Please note that tool is rather hacked-together without real error-handling,
as it is not meant for long-term usage, but might be helpful in the
review-process / adapted to a unit-test.

Apart from the mentioned specifics, nothing is really SQLite-only,
thus I did not change the documentation of the baseclass in the patch
(this documentation also lists all SQL-implementations). At least this
list in the documentation might be extended to make users aware of the
new implementation.

SQlite plugin provided by Oliver Freyermuth (o.f@cern.ch).

Fix problematic error handling and always rollback transaction instead
of trying to commit.

Provided by by Oliver Freyermuth.

The attached patch adds the methods SetLargeObject() and GetLargeObject() with similar syntax to GetBinary() / SetBinary().

Provided by by Oliver Freyermuth.

Patch by Oliver Freyermuth.

Patch by Oliver Freyermuth.