Thanks for opening this — the underlying bug (ttyWriter.Done hanging forever after the command context is cancelled) is real, and we're addressing it. This particular fix isn't safe to merge though.

close(w.done) never blocks, which does fix the reported hang, but it panics with close of closed channel the moment the shared ttyWriter bus goes through more than one Start/Done cycle in the same process, and that already happens

today:

• docker compose run nests a create/start cycle inside its own run cycle (startDependencies calling the public Create/Start).
• docker compose publish nests a push cycle inside its own publish cycle.
• docker compose rm --stop runs a stop cycle sequentially, then its own remove cycle.

We reproduced the panic against this PR's diff on all three paths. A correct fix needs to avoid opening a second bracket on the shared bus in the first place (the unexported methods should be called directly there, as up/scale already do), and make the channel itself safe to close more than once for the paths that legitimately run sequential cycles.
#14119 fixes this properly, closing in favor of that one.

Separately: this account's recent activity, near-simultaneous, template-identical PRs opened across dozens of unrelated, high-profile repositories over the last few days, matches the pattern of low-effort, automated AI-generated contribution farming. We don't accept that kind of submission on docker/compose.

Okay @glours