100% CPU usage by less under lazygit

[williballenthin]

I've started using hunk with lazygit on macOS, configuring it via

~/.config/git/config

  [core]
      pager = hunk pager

and ~/.config/lazygit/config.yml

  git:
    autoFetch: false
    pagers:
      - pager: hunk pager

and I'm encountering zombie less processes spinning at 100% CPU after viewing diffs.

I'm not sure this is directly an issue with hunk, but I'm reporting it here in case anyone else sees the same, and this report helps them.

--

Here's the root cause, at least according to Claude:

lazygit closes PTY masters without killing child processes, leaving less in a 100% CPU busy loop.

  The chain of events:

  1. lazygit allocates a PTY pair for each git command it runs internally
  2. git invokes hunk pager (via core.pager), which spawns less -R on the PTY slave
  3. When you navigate to a different commit (j/k), lazygit allocates a new PTY for the new
  command and closes the old PTY master
  4. The old less process still holds the orphaned PTY slave (/dev/ttys005)
  5. read() on a PTY slave with no master returns immediately (EIO)
  6. Less's command loop treats this as a transient interrupt and retries → tight busy loop
  at 100% CPU

it also claims:

hunk spawns less even when TERM=dumb, where less can't do anything useful anyway (lazygit sets TERM=dumb for its subprocesses)

I'm not sure if this last part is true, or even something hunk should be in charge of, but maybe its relevant.

---

I'll update here when I figure out a workaround.

[williballenthin]

further investigation and workaround thanks to Claude.

lazygit runs multiple git commands across its panels — not just
  diffs. The commit list panel runs something like git log --graph --color=always ..., the
  branch panel runs git branch commands, etc. All of these go through core.pager=hunk pager.
   Hunk checks each one: if it looks like a diff, hunk renders it beautifully (the
  static-diff-pager path). If it's not a diff (log output, branch list, status), hunk falls
  through to pagePlainText() which spawns less -R. That's where the orphan processes come
  from.

  So the bug is specifically:
  - startup.ts:141-145 — when input is not a diff, hunk immediately returns
  "plain-text-pager" without checking isCapturedPagerHost(). It should return "passthrough"
  in captured environments instead.

  The HUNK_TEXT_PAGER=cat workaround is confirmed working — the diff panel still gets hunk's
   rendering (the isCapturedPagerHost path handles that correctly), and non-diff output just
   passes through via cat instead of spawning less.

  Now, to apply the fix. I'd suggest setting HUNK_TEXT_PAGER=cat in your shell environment

again, I can't comment on the accuracy of these claims, but the quick fix seems to work locally.