[cmj0121]

Closes #368.

What

A channel or a function value inside a composite is now refused by every rendering spelling, under E4076. That
covers print, str(…), an f-string hole, .display() and .debug(), at any depth. The refusal names the part and
the whole:

E4076 a chan[int] is an identity rather than a value, and the language gives it no rendering — a list[chan[int]]
renders by its parts and this one is a place to send to or to jump to, whose address is not a rendering of anything
the program wrote

Before this, print [ch], (1, ch), {"a": ch}, a struct field and an enum payload printed the channel's address.
print [f] reached cc (operand of type 'zg_fnptr').

How

c_display_check already refused a bare identity. It now also walks the parts the structural renderer would reach,
using c_identity_part over c_render_parts. That keeps a single E4076 report site. The walk stops where
c_render_runs stops: a type that declares its own display answers for what it holds, so a channel behind one still
renders.

Tests

• reject: five cases, covering a list, a tuple holding a list (fn, str(…)), a map value (f-string), a struct
  field (.display()) and an enum payload. All five were RED on the compiler without the fix (four accepted, one
  reaching cc).
• corpus: codegen/identity_behind_override checks that list[W], a tuple and a map holding a type with its own
  display still render. The seed cannot print a list[W], so the case is an oracle skip. leaks --atExit is clean.
• Gates run one at a time, all OK: reject, lint, refuse, error-codes-check, docs-mirror, docs-links,
  fixpoint, oracle, corpus, sanitize-corpus.

Not here

The stale "a list has no rendering" comments are #371.

🤖 Generated with Claude Code

https://claude.ai/code/session_01N8jd1oK2orM6diHxhvWhC7