3.1r1: one answer for a read of a point type the DUT doesn't have

[jadamcrain]

The contradiction

3.1r1 contradicts itself about how a device answers a read of a point type it has no points of:

• Table 11-1, and TB2016-001, which 3.1r1 follows: the device answers with an error, IIN2.1 or IIN2.2.
• Some of its steps: they still expect a Null Response, which has no error bits. Table 11-1 marks the update of those other sections as still to be done.

As printed, the answer therefore depends on the point type. A device without double-bit inputs must answer a read of g3 with an error, but a read of g4 with a plain Null Response (#7).

The resolution

One rule for every point type:

• No points: a device without the points, or with them uninstalled or disabled for the test, answers with a Null Error Response: no objects, IIN2.1 or IIN2.2 set, and IIN2.0 clear.
• Points but no events: a device with the points but no events still answers a read of those events with a Null Response, as printed. TB2016-003b's subset parsing tables say the same.

Test	As printed, without the points	Now
8.3.2	Null Response, IIN2.1/2.2 optional (8.3.1 requires the error)	the error
8.13.2b, 8.21b	an error response with IIN2.1 or IIN2.2, which may carry data	the error, with no data and IIN2.0 clear
8.14.2b, 8.17.2b, 8.19.2b, 8.22b	Null Response	the error; with the points but no events, a Null Response, as printed
8.22.1	Null Response, IIN2.1/2.2 optional (8.21.1 requires the error)	the error
8.23.2.x.1	an error response with any error bit	the error, IIN2.0 clear; the level 1/2 exit for a device without range reads is unchanged
8.24.2a–g	a, c–e and g read the type; b and f skip it (dnp3c ran none of them)	all seven read the type once and require the error
8.25.2c–e	the error; 8.25.1 also allows a Null Response	the error; with the points but no events, unchanged

A level 1 test of a device that can't read a type (8.13.2a and the like) now runs only for a device that has the type. A device without the type gets the test above. Before, a device with no points and reads switched off could pass with IIN2.0.

Unchanged

2.7 is unchanged. Where 3.1r1 reused 2.7's code, it now has its own:

• 8.17.2b has its own definition
• 8.14.2b and 8.22b override NoEvents only for a device without the points
• 8.19.2b passes AnalogInputChange a parameter whose default is 2.7's isNull

Documentation

• README: a new section, "Where 3.1 rev 1 contradicts itself", with the table above. It notes that the resolution changes only which IIN bits a 3.1r1 run accepts, not interoperability.
• CLAUDE.md: where a revision contradicts itself (its steps against its own tables, or against a bulletin it adopts), the conflict is resolved the most consistent way, the same for every point type, and listed in that README section.

Checks

• Tests: mvn test passes. A new MissingPointsTest runs all 25 affected tests against a device without the points:
  • each passes with IIN2.1 or IIN2.2
  • each fails with a plain Null Response, and with IIN2.0 set
  • 8.14.2b, 8.17.2b, 8.19.2b and 8.22b still want a plain Null Response from a device with the points but no events
• -describe dump: I compared every procedure's -describe output and the disabled lists, before and after, at each subset level in four configurations:
  • 2.7's are byte-identical
  • 3.1r1's differ only in the tests above and the level 1 "can't read" tests