advanced_search

AdvancedSearch

Builds database search queries with complex, nested logical expressions.

You provide an abstract syntax tree representing the search, and AdvancedSearch builds the search clause of the query (eg. the where clause, in SQL).

Good for	Not Good For
searches	reports
complex logic expressions	simple "conjunction of all parameters" forms
natural language	99% of business search forms

Input can be natural language,

'age less than 40 and ssn equal to 123-45-6789'

or a typical HTML form,

{
  age_lt: 40,
  ssn_eq: '123-45-6789'
}

You organize and name your parameters however you want. You convert those parameters into an AdvancedSearch::AST:

       and
    /       \
   lt        eq
  /  \      /  \
age  40   ssn   '123-45-6789'

A convenient S-expression syntax is provided to help you build your tree:

s(:and,
  s(:lt, s(:id, :age), s(:value, 40),
  s(:eq, s(:id, :ssn), s(:value, '123-45-6789')
)

AdvancedSearch converts your tree into a parameterized search clause (eg. the where clause, in SQL).

'where age < $1 and ssn = $2'
[40, '123-45-6789']

Finally, you do whatever you want with that search clause. Prepare a complete query and execute it using ActiveRecord's find_by_sql method, perhaps? Maybe execute it directly using a gem like pg or mysql2? What you do with your query, in the privacy of your own home, is your own business.

Complete Example

There is a complete example using ActiveRecord's find_by_sql method in the test suite. See:

spec/support/activerecord/models
spec/support/activerecord/searches/library_catalog.rb
spec/activerecord/searches/library_catalog_spec.rb

Adapters

adapter	gem	status
PG	pg	Proof of concept
Mysql2	mysql2	Not yet implemented
ActiveRecord	activerecord	If only ARel had docs

Design Goals

• Simple, not easy. You'll have to write substantial code to get started, but future changes will be simple.
• Requires basic understanding of graph theory, must know what a tree is.
• Agnostic: Support for specific databases and ORMs is provided via adapters.
• No dependencies
• Unit tests of your search objects do not need to touch database, so they are very fast.

Roadmap

• full-text search
• "raw" node?