Date: September 13, 2026
Status: an open question, written down so it is not mistaken for a small one.
It decides nothing — no rule here, no change to the emitter. Two records
already stand against an inserted clone, and this page says what they are and
what a proposal would have to do with them.
Related: ADR-037 D3 (Shared stays written by
hand, and the reason: sharing changes when a value is cleaned up) and D7 (the rule
that sorts written from inferred),
ADR-005 §3 (compiler-inserted clones, rejected:
“diagnostics suggest, the user decides”),
ADR-283 D3 (the same ban, argued again for the
view case), ADR-297 (cleanup, which is what the
reason is about), Part I 8.3 and Part II
11.2 (a task moves what it uses, so the
clone goes before the task is built)
spawn is @detached, so what the body uses moves into the task when the task
is built. Keeping a value in the parent therefore costs a hand-written
.clone(), placed before the spawn — and that is visibly a chore. The
obvious next proposal is that the compiler place it: see that the parent reads
the value afterwards, clone it quietly, and let both sides have one. It reads
like a comfort feature. It is not one.
A clone is one more owner, and one more owner means a later cleanup. That is not
a side effect of the convenience; it is the convenience. And “sharing changes
when a value is cleaned up” is the whole reason
ADR-037 D3 gives for Shared being written by
hand rather than guessed: inferring it “would move a destructor silently”, and a
program can observe that it moved (ADR-297). An
inserted clone produces the same observable change by a different route — the
compiler decides, without the program saying so, that a value now stops being
cleaned up where it used to be. D7 states the sorting rule the same way round:
what is observable is written; what is not is inferred, with safe as the floor.
A cleanup point is observable, so it is written.
So the question is not “may the compiler be helpful here”. It is D3’s question
over again, asked about a clone instead of about a Shared annotation, and the
prohibition has already been written twice on its own account:
ADR-005 §3 rejects implicit clone outright —
“Nikaia never inserts a clone the user did not write; diagnostics suggest, the
user decides” — and ADR-283 D3 re-argues it for
the view case rather than leaning on it. Whoever proposes an automatic clone owes
an answer to that reasoning: some account of why a moved cleanup point is
tolerable when inserted by this rule and not by the one D3 refused, and why the
spawn case is different from the two §3 already covers. Not an appeal to how
much typing it saves. Until such an answer exists, one clones by hand.
