Share stdout write/print effect sources (negative lines, same behavior)

[jkbennitt]

Summary

IO.write, IO.print, and (on JS) IO.print_err are the same stdout/stderr
wrapper with a file descriptor and optional newline. Today that logic lives in
near-copy files:

• effs/write.js / print.js / print_err.js (7 lines each; same io_out shape)
• effs/write.c / print.c (nearly the same io_cstr → io_out run)

This follows the same simplification Bend already merged for audio and file
effects: related effects share one source; Base keeps importing that one path;
eff_src still embeds a file only once.

print_err.c stays its own file: it correctly routes through io_errs, which
is not the same as io_out(stderr, …).

No feature removed. No comment stripping. Behavior of IO.write / IO.print /
IO.print_err unchanged.

Negative diffstat

 AGENTS.md               |  1 +
 bend2/base.bend         |  6 +++---
 bend2/effs/print.c      | 28 +++++++++++++++++++++-------
 bend2/effs/print.js     | 10 ++++++++++
 bend2/effs/print_err.js |  7 -------
 bend2/effs/write.c      | 18 ------------------
 bend2/effs/write.js     |  7 -------
 7 files changed, 35 insertions(+), 42 deletions(-)

Net −7 lines (honest: smaller than a rough pre-patch estimate — the
#ifdef wrappers use most of what the shared io_stdout_run helper saves).

Test plan

• Local full-matrix mirror of gates/test.ts on this branch: PASS: 1427 / 1427 (check / interp / JS / C; ~158s). Official mini-cluster gates/test.ts was not run from our side (no bastion/cluster SSH).
• Smoke: IO.write / IO.print / IO.print_err alone and together on interpreter, emitted JS, and native C (including empty strings / UTF-8 / newline vs no-newline); matched pre-patch HEAD.
• gates/repo.ts: PASS: 46 / 46 (with ttok on PATH).
• No bench intent to change; runtime pins should be unaffected.

Notes

Happy to reshape naming (stdout.c vs keeping print.c as the shared home) to
match whatever convention you prefer after #950/#939.

[nicolas-abril]

Thank you for this. The sharing is correct: we emitted every test on main and on this branch and the only lines that differ are the effect sources themselves, and the per-effect #ifdef gating checks out for print-only, write-only, print_err-only and all three together. We are going to keep the one-file-per-effect layout for stdout, though: these files are small, essentially finished and will not drift, and a reader looking for IO.write should land in write.c rather than on a flag inside print.c. That is a stylistic call, not a fault in the work, so closing this one; the work is appreciated.

Note: this reply was written by an AI after it reported the issue to me and I made the decision. If anything here is wrong, reply and I will review it myself.