Provided by: libsql-statement-perl_1.412-1_all bug


       SQL::Statement::Roadmap - Planned Enhancements for SQL::Statement and SQL::Parser

       Jens Rehsack - June 2010


       This document gives a high level overview of the future of SQL::Statement, SQL::Parser and
       its impact.

       The planned enhancements cover testing, performance, reliability, extensibility and more.


   Enhancements in SQL::Statement 1.xx
       SQL::Statement 1.xx will not receive big changes, but a few enhancements may help us to
       design SQL::Statement 2.xx much better.


       SQL::Statement is missing some functions, types, operators etc. It's supported to add
       missing functionality - but the implementation wasn't picked up during the modernizing of
       column evaluation. See RT#52397 for some more information.

       This should be done before SQL::Statement 1.xx reaches the end of its road.

       Parser improvements

       The SQL::Parser is implemented based on a lot of regular expressions and some manually
       developed logic. This creates some issues like RT#53416 or RT#55190. Further, trailing ";"
       causes SQL::Parser to croak. We need to decide what can be fixed without internal design
       changes and what has to wait.


       There is no intention to work on performance improvements in SQL::Statement 1.xx. The
       performance is good as it is and improvement requires design changes.


       Bugs will be fixed - where possible. SQL::Statement 1.28 is much more reliable than
       SQL::Statement 1.15. Even if a bug cannot be fixed all issues are gratefully received as
       they will be considered in the design process for SQL::Statement 2.xx better.


       SQL::Statement 1.xx is highly extensible, even if a more object oriented design would
       improve that. The 1.xx branch will not be redesigned for greater extensibility on a coding

   Enhancements in SQL::Statement 2.xx
       Concerning the procedural design of SQL::Statement 1.xx a rewrite of the basic components
       is required.

       SQL::Parser rewrite

       The SQL::Parser needs to be modified to be able to use a Backus Naur Form
       <>.  This would allow users and developers to
       rely on many different SQL dialects.  This will allow better extensibility from a feature
       point of view without losing ANSI SQL compatibility.

       SQL::Statement rewrite

       SQL::Statement should be reduced to a simple coordinating engine. The executing tasks
       should be organized into separated commands. This will reduce side effects and will open
       the door for higher level optimizations, reliability improvements or sub-selects (or other
       calculated tables).


       There is a large list of missing features but not all table backends will be able to
       support each new feature. The most popular requested features need additional discussion
       and everyone is welcome to do it on the


       Locking table within SQL scripts to manually control table consistence over several
       operations. The current locking support is restricted to one statement.

       Transaction support

       Executing statements on a temporary copy of the table data.

       The easiest way to implement this would be to create a SQL::Statement::RAM on "BEGIN
       TRANSACTION" and write the entire table back on "COMMIT" or discard on "ROLLBACK".

       Better performance could be achieved in cases where the implementation is enabled to
       memorize pending modifications and apply them at "COMMIT".  On the other hand there are
       already capabilities to improve some operations, which might create confusion in case of

       This needs more discussion.


       Adding, removing or modifying columns is not supported for created tables. A generic
       "ALTER TABLE" seems to rely on the implementation of the transaction support - until
       better ideas are provided.


       Currently some table backends have implicit support to access specified rows quicker than
       fetching each row and evaluating the where clause against the row data.

       An interface would be required to configure fetching to return only rows matching a
       restricted where clause. Another (probably better) way to support indices would be to
       fetch index entries at first and have an interface to the table fetching lines based on an
       index key.


       In most cases queries can be re-expressed without using sub-selects. But in any case,
       there are circumstances where sub-selects are required.

       The first implementation will do the sub-select before the primary statement is executed
       without any further optimization. Hopefully a later version will provide better
       Performance with some optimization.

       Query based variables

       Currently the only variable I can imagine is "ROWNUM". More suggestions are very welcome.

       Better SQL Script support

       In SQL::Statement 1.xx the function "RUN ()" provides SQL script execution. This function
       may have limitations and side effects (at least when the executed SQL touched the same
       tables as the primary statement).

       I plan to improve the SQL script support to remove the side effects on the one hand and
       have a more flexible and easier way to execute them.

       Finally it should be possible to execute a script via:

           $dbh->do( join( ";", @script ) );

       Trigger support

       Most important when doing complicated things is having callback functions for several
       events. While real triggers will not be possible for SQL::Statement and underlying pseudo-
       databases, callbacks could be provided via triggers.


       There are several performance optimizations required for SQL::Statement 2.xx.

       The first one should be done on a very high level (query optimization) by implementing
       algebraic evaluation of queries and clean implementation of typical database algorithms.
       With respect to the basic optimization rule premature optimization is the root of all
       evil, it is primarily targeted to have an adequately fast, reliable implementation of many
       algorithms (e.g. early incomplete evaluation to reduce amount of rows, transpose where
       clause to evaluate constants first) and a clever controller choosing the right algorithm
       for a specific query.

       The second optimization goal means: implementing most expensive methods in XS. This
       requires a good performance test suite as well as some real world usage cases.


       This is one of the primary goals of SQL::Statement. I hope to reach it using test driven
       development and I hope I get some more todo's from the users for this.


       The currently high level of extensibility should be increased on a coding level. This will
       be done by redesigning the entire parser and execution engine using object oriented
       techniques and design patterns.


       Many tests in SQL::Statement are not well organized. The tests should be reorganized into
       several parts:

       Basic API
           This part should test the entire basic API of SQL::Statement, SQL::Parser and probably
           the entire engine command classes.

       DBI / Table API
           This part should test if the API to DBI drivers work (maybe an empty test driver will
           be needed for that).

           This part should test the functionality of the SQL::Parser and the SQL::Statement

           This part should be used to implement full usage cases (ideally from real world
           projects) to allow for testing optimizations.


       Our priorities are localized to our current issues and proof of concept fixes for upcoming
       SQL::Statement 2.xx.

       Any additional priorities (as missing features, the SQL::Statement rewrite) will come
       later and can be modified by (paying) users.


       See <> for how you can help.

       If your company has benefited from the DBI or SQL::Statement, please consider if it could
       make a donation to The Perl Foundation "DBI Development" or "SQL::Statement Development"
       fund at <> to secure future development.

       Alternatively, if your company would benefit from a specific new DBI or SQL::Statement
       feature, please consider sponsoring its development through the options listed in the
       section "Commercial Support from the Author" on <>.

       Using such targeted financing allows you to contribute to DBI development (including
       SQL::Statement and PurePerl DBI drivers) and rapidly get something specific and directly
       valuable to you in return.

       Thank you.