0.6.1 --- the same bounded-transfer defect, and the tests that had never run

[Sunrisepeak]

This is openkal-linux's defect, in this file, unchanged. mcpplibs/openkal-linux#19 is the counterpart and carries the full derivation.

The defect

kal_timeout_read and kal_timeout_write decoded their kal_stream argument with okm::unpack, which is the decoder for an owned handle. A stream handle is not one: these are this environment's own descriptors, and kal_stream_read/kal_stream_write take them as such (fs.cpp:156, net.cpp:143, process.cpp:137-138, stream.cpp:6-8).

The decode did not fail. handle.h packs an owned handle as (generation << 32) | (fd + 1), so a bare descriptor N has exactly the shape of a packed handle naming N-1, and unpack accepts it whenever generations[N-1] is still zero. The wait was performed upon descriptor N-1 and the transfer upon N. Of the three standard streams only kal_stdin, whose handle is zero, failed to decode; kal_stdout decoded to descriptor 0 and kal_stderr to 1.

kal_timeout_accept and kal_timeout_recv_from keep the decode and are correct: a listener and a datagram are owned. The four call sites divide exactly along the borrowed/owned line and the two that were wrong were the two holding a borrowed handle.

Why it was invisible here, which is the larger half of this change

This package's own tests had never run. tests/ has held five suites since the implementation was written and no workflow invoked them. What ran was the specification's suite, which is a different instrument: it examines what every implementation must do, and these examine what this one does with the system beneath it.

The evidence that they had never run is in the suites themselves:

• all five announced their success as openkal-linux: ...;
• conformance_env_time.cpp did not compile at all, naming kal::time::prop_wall_available, which openkal 0.9 spells kal::time::wall_available.

So this change adds the mcpp test step, taken from openkal-linux unchanged including its assertion that every suite ran — a suite that discovered nothing reports success. The five suites are corrected to name this implementation and the stale property name is fixed.

The criterion

tests/conformance_timeout.cpp is new. There was no test of openkal.timeout in this package at all, and the specification's suite examines it through observations that a wait upon the wrong descriptor satisfies: a bounded read of the standard input is permitted to expire, so an implementation that expires for the wrong reason is indistinguishable there from one that is right.

• Sixteen channels and not one, because under the defect the wait was upon N-1 and whether that expires depends on what occupies N-1. For a channel made after another, N-1 is the previous channel's writing end, upon which input is never reported.
• First in main, because the mistaken decode corrected itself for every index at which an owned handle had already been released.

Measured

This system has no runner on the machine this was written on, so the six suites were compiled and run against openkal-linux, which provides the same interface:

 test result ok. 6 passed; 0 failed

and the new suite was then run against openkal-linux with the defect put back:

  0 of 16 bounded reads transferred
FAIL: a bounded read of a stream that has bytes waiting transfers them

Continuous integration on this branch is the criterion for this system, and it is the first run in which tests/ has ever executed here.