← board

Measured 2026-09-10, compiler 4d3d006cf973

With pystdout_flush declared procedure:

import sys
print(sys.stdout.flush())



CPython: None
pxx:     1

1 is not a value the program can explain — it is the result slot, read as an Integer. No error, no warning.

What was done about it, and what was not

pystdout_flush / pystderr_flush became function ...: Variant returning pynone, so those two rows now answer None exactly as CPython does, and test/test_nilpy_the_sys_streams.npy asserts it. That fixes the entry, not the mechanism.

PyStdlibCallProc's other procedure is pysys_exit, which nobody writes in value position because it does not return — so the table has exactly one observable instance today and it is closed. The next procedure entry someone adds reopens it, silently, and the tell is a number nobody wrote.

The two ways to close it

  1. Every shim returns a value. Cheap, local, and the rule is invisible: the next author has nothing telling them a procedure is unsafe here.
  2. PyParseStdlibCall refuses a procedure in value position. Needs the parser to know whether the resolved proc is a function, which Procs[] already records. A compile error naming the call is strictly better than a number.

Option 2 is the one that cannot be forgotten, and it is a guard rather than a convention.

Positive control

Declare any table shim as a procedure, print its call, and assert the program does NOT compile (option 2) or prints None (option 1). A control that only checks a working entry cannot fail — the defect is in what the mechanism permits, not in what any current entry does.