[mremond]

Why

io:format and io_lib:format printed every atom bare, so a trapped exit came out as {EXIT,bye} and atoms such as 'Foo' or 'hello world' did not read back as the same term, unlike on BEAM.

What changed

• ~w, ~p, ~W and ~P quote atoms with OTP's rules: bare only when an atom starts with a lowercase letter, holds only letters, digits, _ and @, and is not a reserved word (maybe and else included).
• Quotes, backslashes and control characters are escaped inside the quotes; characters above 255 print as \x{...} unless the t modifier is given.
• A new io_format_atom_quoting fixture runs natively and on BEAM.
• The Cowlib demo's two 'HTTP/1.1' lines now match OTP, so they leave its known differences; FEATURES, CHANGELOG, KNOWN_LIMITS and the compatibility records say so.

Before and after

Before: Programs printed {EXIT,bye}, Foo and hello world; 43 of 44 lines of an atom probe differed from OTP 29.

After: Programs print {'EXIT',bye}, 'Foo' and 'hello world'; all 44 probe lines are byte-identical with OTP 29.

The change is in text a program prints, so the before and after are shown as program output rather than pictures.

Checked

• Atom probe (44 lines: uppercase, empty, punctuated, reserved, escaped, Latin-1 and non-Latin-1 atoms through ~w, ~p, ~tw, ~tp, ~W, ~P and field widths) compared with OTP 29.1.2: Before: 43 of 44 lines differed. After: identical, including after bringing the branch up to date with the native timer change.
• io_format_atom_quoting regression: Fails before the change (prints {EXIT,bye}); passes after, 10 of 10 runs natively and 10 of 10 on BEAM.
• Cowlib 2.22.0 demo compared with OTP 29: 77 of 85 calls byte-identical (was 75); sorted comparison leaves only the 2 known byte-order lines.
• Full end-to-end suite: 468 passed, 0 failed, 2 existing skips.

How it was checked

cargo build --release --locked -p nera-runtime; cargo build --release --locked -p nera

cargo clippy --workspace --all-targets --locked -- -D warnings: clean

cargo test --workspace --locked --no-fail-fast: 2,139 passed, 0 failed, 1 ignored

./target/release/nera test -j 6 --timeout 30: 468 passed, 0 failed, 2 skipped (onecached fixtures needing external modules)

nera test io_format_atom_quoting ten times in a row: 10/10; erlc + erl running the same module ten times: 10/10

python3 scripts/test-cowlib-demo.py with the updated known differences: 4/4 tests pass, no known line newly identical

Probe module compiled with the base and patched compilers and with erlc, outputs diffed against OTP 29.1.2 on macOS arm64

After bringing the branch up to date with the native timer change: rebuilt, then io_format_atom_quoting, the six io_format/io_lib fixtures and the three timer fixtures pass (10/10), and the probe is still identical with OTP

Risk

Risk above low: Blast radius, medium — The shared term writer is used by every ~w/~p/~W/~P call and by crash messages; Behavior, medium — Printed output changes by default for atoms that need quotes, now matching BEAM

Full risk assessment

Dimension	Level	Why
Blast radius	medium	The shared term writer is used by every ~w/~p/~W/~P call and by crash messages
Behavior	medium	Printed output changes by default for atoms that need quotes, now matching BEAM
Undo	low	Revert and done; no saved data or settings
Evidence	low	Byte-identical with OTP 29 on a 44-line probe, regression on both engines, full suite
Sensitivity	low	Formatting only
Novelty	low	One helper beside the existing term writer, following OTP's io_lib
Dependencies	low	No packages added or updated

Not done

Nera's own crash message (Error in process ...) now quotes atoms too, since it shares the term writer.

The separate error_logger renderer keeps its own simpler quoting rule and is unchanged.