Bug

The double_as_int function is used in the AST code to convert floating-point literals into integers such that they can be compared against integer values.

Nominally this function is meant to do the following sorts of conversions:

1 < 1.1 -> 1 < 2
1 > 0.9 -> 1 > 0

Where depending on the kind of operations being performed we may need to compare the integer value against the floor or the ceiling of the integer literal.

Unfortunately, the double_as_int function has a bug where it will always return the floor of the floating-point literal for all operations besides == and !=.

This results in in correct behaviour in the following situations:

1. When filtering against a field like {"n": 1} the query n < 1.1 will erroneously not match the record, and the query n >= 1.1 erroneously will match the record.

2. When filtering against the timestamp range index when it contains a range like [1.1, 1.5] a query like timestamp < 1.4 will erroneously not match the archive's time range. (This is because the timestamp range index evaluation code has some behaviour where it will convert float literals to integers before converting them back to floats to compare against a double-encoded time range, which is nominally correct, but overly complicated).

CLP version

v0.5.1

Environment

ubuntu focal

Reproduction steps

Simplest reproduction is:

1. Ingest {"n": 1}
2. Observe that the query n < 1.1 returns no results