Anatomy of a Query, Part 1
October 23, 2022
A FlipDB query is essentially a highly structured dfn or collection of dfns; it is an ordered set of names and expressions. It is thus much less sophisticated than an SQL query. This simplicity, however, brings certain benefits. It means that we can step through the query line by line (one name and expression at a time), and evaluate, inspect, or even change the code. In other words, the query may be traced just like a function in the Dyalog session, and in fact the FlipDB UI has its own tracer. This is not possible with SQL - an SQL query is not made up of individual, executable, traceable steps.
A FlipDB query usually starts with a specific table, for a default context, and is then composed of the following ordered sections or clauses:
- Precomputed expressions
- Where statements
- Where-not statements
- Group-by expressions
- Select expressions
- Order-By clause
- Having statements
Each section (except #6) is an ordered set of names and expressions, and may be thought of as a dfn.
Precomputed expressions are evaluated before any where statements are applied, and thus operate on the entire table. These may be used to specify constants, computed columns and aggregate values. (These latter two are especially valuable for expressions that may be row dependent, and thus change after the where clause is applied.) For example:
The FlipDB tracer steps through these lines just like the Dyalog APL tracer steps through a function. At each line variables may be inspected and played around with. Also like a dfn, a defined name may be referenced in the following lines of the section. The defined names may then be referenced in subsequent sections or clauses, which is of course the whole point of this clause.
Where and Where-Not Statements
Next comes where statements, a set of zero or more Boolean valued expressions:
These expression are not named by the user, but for tracing purposes the system assigns them names
Where3, etc. These statements could be packed into a one-liner, but there are significant advantages for the user in keeping the expressions as simple and short as possible, for tracing, debugging, and performance.
The tracer steps through these line by line, allowing the user to see exactly how many rows are restricted by each statement, and the net effect of each statement. At the end, the statements are and'ed together:
Where←∧/Where1 Where2 Where3 ...
Where not statements are similar to where statements, except they are or'ed together:
WhereNot←∨/WhereNot1 WhereNot2 WhereNot3...
The tracer steps through these as well. The final boolean restriction is then simply:
If the query is to be grouped, then one or more group expressions may be applied. For example:
These expressions are evaluated on the rows remaining after the Where and Where Not clauses are applied. These too are stepped through in the tracer, allowing detailed inspection.
After the optional grouping clause, the select expressions are evaluated. For ungrouped queries these will often be a simple list of column names. For grouped queries generally all of the expressions are aggregate expressions of one sort or another:
Note that the last expression is making use of a name defined two lines above it, as well as a name defined in the first section of computed values. Yet again, the tracer lets us step through these expression line by line.
The order-by clause is the only section that is not a set of names and expressions. Rather it is a list of names defined in the Select clause and a sort directive. For example:
The tracer does not step through the order-by clause line by line, as there is little value added.
Finally, one or more having expressions may be defined, which are analogous to where statements, except they are applied to the sorted result. For example:
These expressions may be traced just like where expressions.
A FlipDB query is an ordered set of names and expressions that both defines the query and also explicitly specifies how it is to be executed. Whereas an SQL RDBMS might have a query optimizer that selects a query execution path, FlipDB, for better or worse, just does what the user tells it, in the order the user tells it. While no doubt there are a number of costs to this simple model, there are also major benefits. The traceability of a FlipDB query is a direct consequence of the fact that the query is an ordered set of names and expressions. That is, we can read the query incrementally from top to bottom unlike an SQL query where we must mentally process and parse the query to the end before we can understand what is happening at the beginning. The computer and the user or author read and process the query in exactly the same way.
In a future post we will look at how FlipDB deals with joins, views, sub-queries, and all the complexities that arise when data is in a different table.