proposal 185

This is a fix for Issue 185.

The current duration(string) function accepts a floating point value with a s suffix indicating a duration in seconds which is at parity with what the protobuf utilities support for parsing google.protobuf.Duration values from strings; however, this format is incredibly difficult to use in practice. A more human- readable format like the one supported by golang time..ParseDuration would vastly improve the usability of duration(string).

The golang time.ParseDuration method supports an easy to read duration string as input:

Examples:
    0
    1h2m3..4s
    +2.999999999us
    -24h

Grammer:
   duration: sign? (0 | (digit ('.' digit)? unit)+)
   sign: ('+' | '-')
   digit: 0..9+
   unit: ('ns' | 'us' | 'ms' | 's' | 'm' | 'h')

This functionality is also available in the abseil.io C++ libraries as time::ParseDuration.

Go and C++ support inf as an input, but as the input is not a valid google.protobuf.Duration value, infinite duration values should not be accepted in CEL. Likewise, the valid duration range will be limited to the duration values which can be supported by the google.protobuf.Duration

Feedback from jimlarson:

• "Grammar"
• Second example doesn't match grammar - should it be 1h2m3.4s?
• Not the usual numeric syntax: allows 007s, consider (0|[1-9][0-9]*)(.[0-9]+)? for numbers.
• Allows oddities, need to specify the semantics:
  • Arbitrary order: 15m4h3s
  • Repeats: 3s13s7s
  • Fractional things which aren't usually fractions: 0.8h
  • Fractional parts of one unit can overlap with other units: 2.0047h4.009s12.4853ms43.9247us.
  • Consider requiring order of descending scale, no repeats, ditch fraction except perhaps 3672.432s with no other components.