Catch/convert between comparisons in patients.categorised_as/satifying expression

[rebkwok]

In a patients.satifsying expression (and by extension, patients.categorised_as), this is invalid:

patients.satisfying("20 < age < 50")

However, the study definition checks don't catch it.

cohortextractor.expressions.is_allowed checks for allowed tokens, but doesn't catch this one because the individual tokens are allowed.

The last attempt to validate (cohortextractor.expressions.validate_expression) by turning it into an SQL expression by replacing names with empty value (in this case 0) and executing a SELECT on it doesn't fail. (SELECT (20 < 0 < 50) doesn't raise any errors).

However, when executed on actual data, it fails with a SQL syntax error, which means that errors aren't raised to the user until they try to run the study against a real backend (see this thread for a real example.

We should either:

• turn that expression into one that works (age > 20 AND age < 50)
  OR
• check for expressions of that format and raise a validation error