Improvements to Trac's data models
Trac's datamodels are usually implemented in the <subsystem>/model.py files.
In the current situation (1.0.x/1.1.1), we have different APIs and different conventions for the datamodels. This page describes a few issues where the API and conventions can be improved and made more consistent.
Representation of data
Standardize missing values
For tickets, when a field is unset the database value is set to None / NULL (#11018). When retrieving NULL values from the database, we get None in Python, and the value returned from the model depends on the field type:
- For text fields, the empty string is returned from the model using the special value empty. See for example what we do for ticket fields.
- For time fields,
Noneis returned.
The pattern could be extended to other model classes, and even the Ticket model should be reviewed to ensure that all fields follow the pattern.
NOT NULL columns
Validation of the ticket summary is done in the IRequestHandler, to prevent creating a ticket with an empty summary. The schema could be changed to use NOT NULL in the column specification, since the empty string is replaced with NULL. Alternatively, a TracError could be raised in the Ticket model when inserting or updateing a ticket with an empty summary (#12458).
Class methods
Class methods are used for table-wide queries. One such example is select.
The signature of the select method is not consistent across all classes.
The Milestone class has an include_completed parameter in the select method.
We could reconcile the inconsistency by having where, limit and order_by parameters on each method, with the parameters directly mapping to phrases in the SQL query.
This would make the methods more generally useful to plugin developers.
Not only is the interface inconsistent, but also the return value. In most cases a generator is returned by select, however some select methods return a list:
