Tangram::Sucks - what there is to be improved in Tangram
Tangram has taken a concept very familiar to programmers in Java
land to its logical completion.
This document is an attempt by the coders of Tangram to summarise
the major problems that are inherant in the design, describe cases for which
the Tangram metaphor does not work well, and list long standing TO-DO
items.
DESIGN CAVEATS
- query language does not
cover all SQL expressions
- Whilst there is no underlying fault with the query object metaphor per
se, there are currently lots of queries that cannot be expressed in
current versions of Tangram, and adding new parts to the language is not
easy.
- some loss of encapsulation
with queries
- It could be said this is not a problem. After all, adding properties to a
schema of an object is akin to declaring them as "public".
Some people banter on about data access patterns, which
the Tangram schema represents. But OO terms like that are usually
treated as buzzwords anyway.
HARD PROBLEMS
- partial column
select
- This optimisation has some serious dangers associated with it.
It could either be
- no support for SQL
UPDATE
- It may be possible to write a version of
"$storage->select()" that does this,
which would look something like:
$storage->update
( $r_object,
set => [ $r_object->{bar} == $r_object->{baz} + 2 ],
filter => ($r_object->{frop} != undef)
);
- no explicit support for
re-orgs
- The situation where you have a large amount of schema reshaping to do,
with a complex enough data structure can turn into a fairly difficult
problem.
It is possible to have two Tangram stores with different
schema and simply load objects from one and put them in the other -
however the on-demand autoloading combined with the automatic insertion
of unknown objects will result in the entire database being loaded into
core if it is sufficiently interlinked.
- replace SQL
expression core
- The whole SQL expression core needs to be replaced with a SQL abstraction
module that is a little better planned. For instance, there should be
placeholders used in a lot more places where the code just sticks in an
integer etc.
- support for `large'
collections
- Where it is impractical or undesirable to load all of a collection into
memory, when you are adding a member and then updating the container, it
should be possible to do this without loading the entire collection into
memory.
This could actually be achieved with a new Tangram::Type.
MISSING FEATURES
- concise query
expressions
- For simple selects, the query syntax is too long. Getting remote objects
should take less code.
- non-ID
joins
- We can't join on anything but "ID" values
- tables with no primary
key
- We can't map tables unless they have a primary key, and it is called
"id" (or, at least, the same name as the rest of the
schema).
- tables with
multi-column primary keys
- We can't map tables when they have multiple primary keys. Well, you can,
but only if you make a view with an ID column which is functionally
derived from the multi-part keys. But that sucks.
- tables with
auto_increment keys
- These suck, but Tangram could still support them without requiring schema
hacks.
- tables without a
`type' column
- The 'type' column is unneeded for base tables which do not have
sub-classes.
- tables with custom
`type' columns
- For mapping schemata where some clever person has invented their own
special way of representing types using discrete column values.
- tables with
implicit (presence) `type' columns
- It should be possible to infer the type value based on knowledge of the
schema, and the tables which have rows.
- fully symmetric
relationships
- back-refs are read-only.
- bulk inserts
- Inserting lots of similar objects should be more efficient. Right now it
generates a new DBI statement handler for each object.
- `empty subclass' schema support
- You should not need to explicitly add new classes to a schema if a
superclass of them is already in the schema.
- warn about column
redefinitions
- Defining a column twice should be an error. Reported by Mark
Lawrence.