Hi! I ran into a race condition in the parallel task scheduler that causes a hang when a task with sources/outputs configured gets skipped ("sources up-to-date"). Its downstream dependent tasks never get scheduled and mise hangs indefinitely.

mise version: 2026.4.4 (latest)

Reproduction

I put together a minimal demo repo that reproduces the issue reliably on ubuntu-latest: https://github.com/patrickhoefler/mise-sources-outputs-hang

A failing CI run: https://github.com/patrickhoefler/mise-sources-outputs-hang/actions/runs/24010015826

The entire mise.toml is just two tasks:

[tasks.a]
sources = ["in"]
outputs = ["out"]
run = "cp in out"

[tasks.b]
depends = ["a"]
run = "echo b"

Run mise run b once to prime out, then run it in a loop. Task a will be skipped on every subsequent run, and eventually b hangs indefinitely (it's a race condition, so the number of iterations before it hangs varies):

[a] sources up-to-date, skipping
# b never runs

Root cause (best guess — I'm not a Rust programmer and not familiar with the codebase)

After digging in with some AI assistance, the issue appears to be in src/task/deps.rs. The pump_deps function in task_scheduler.rs calls subscribe() twice. subscribe() replaces self.tx with a new channel sender each time, but does not clear self.sent. This means that if a task completes (or is skipped) in the window between the first receiver being dropped and the second subscribe() call, its downstream tasks get permanently marked as "already sent" in self.sent — even though the send failed silently into the dead channel. The second subscribe() then skips them because they're already in self.sent, and they never get scheduled.

Proposed fix (take with a grain of salt)

In emit_leaves(), if self.tx.send() fails, remove the key from self.sent so it can be re-emitted once a live subscriber connects:

if self.sent.insert(key.clone()) {
    trace!("Scheduling task {0}", task.name);
    if let Err(e) = self.tx.send(Some(task)) {
        trace!("Error sending task: {e:?}");
        self.sent.remove(&key);  // allow re-emission on next subscribe()
    }
}

Happy to help test a fix! Let me know if you need any more information.

The investigation, root cause analysis, and proposed fix were worked out with the help of Claude Code.

mise doctor output

version: 2026.4.4 linux-x64 (2026-04-05)
activated: no
shims_on_path: yes
self_update_available: yes

build_info:
  Target: x86_64-unknown-linux-gnu
  Features: OPENSSL, RUSTLS_NATIVE_ROOTS, SELF_UPDATE
  Built: Sun, 5 Apr 2026 01:27:31 +0000
  Rust Version: rustc 1.94.1 (e408947bf 2026-03-25)
  Profile: release

shell:
  /bin/bash
  GNU bash, version 5.2.21(1)-release (x86_64-pc-linux-gnu)
  Copyright (C) 2022 Free Software Foundation, Inc.
  License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>

  This is free software; you are free to change and redistribute it.
  There is NO WARRANTY, to the extent permitted by law.

aqua:
  baked in registry tools: 2279

dirs:
  cache: ~/.cache/mise
  config: ~/.config/mise
  data: ~/.local/share/mise
  shims: ~/.local/share/mise/shims
  state: ~/.local/state/mise

config_files:
  ~/work/mise-sources-outputs-hang/mise-sources-outputs-hang/mise.toml

ignored_config_files: (none)

backends:
  aqua
  asdf
  cargo
  conda
  core
  dotnet
  forgejo
  gem
  github
  gitlab
  go
  npm
  pipx
  spm
  http
  s3
  ubi
  vfox

plugins:

toolset:

path:
  ~/.local/share/mise/shims
  ~/.local/share/mise/bin
  /snap/bin
  ~/.local/bin
  /opt/pipx_bin
  ~/.cargo/bin
  ~/.config/composer/vendor/bin
  /usr/local/.ghcup/bin
  ~/.dotnet/tools
  /usr/local/sbin
  /usr/local/bin
  /usr/sbin
  /usr/bin
  /sbin
  /bin
  /usr/games
  /usr/local/games
  /snap/bin

env_vars:
  MISE_LOG_LEVEL=info
  MISE_YES=1
  MISE_TRUSTED_CONFIG_PATHS=~/work/mise-sources-outputs-hang/mise-sources-outputs-hang
  MISE_GITHUB_TOKEN=REDACTED

settings:

No problems found