Orphans on Windows

Describe the bug

lazygit creates many orphans on Windows. The orphans contain many instances (triple digits) of conhost.exe, git.exe and difft.exe (my secondary pager). The difft processes are suspended, while the others are not but don't consume any CPU.

Possibly related to #5675

To Reproduce

I only know how to reliably create the difft and conhost orphans.

Configure difft as a pager (externalDiffCommand: difft --color=always)
Open lazygit and select difft as a pager.
Quickly select different commits. In my case, difft is too slow to render each diff.
A handful of suspended difft orphans appear. Also a few conhost processes are spawned. They are displayed as children of the lazygit process in process explorer but remain alive after I exit lazygit. The conhost orphans also appear if I use a different pager (e.g. delta).

Expected behavior

No orphans.

Screenshots

Initial state - lazygit

After rapidly selecting different commits

After exiting lazygit

Version info:

commit=aafe61082e7ed383d318fd40e48f85645e6afc7b, build date=2026-07-15T11:34:26Z, build source=binaryRelease, version=0.63.1, os=windows, arch=amd64, git version=2.55.0.windows.3

Terminal info:

I'm on windows 11 using pwsh 7.6.4 inside of Windows Terminal 1.24.11911.0.

Kill the pty child's process tree on Windows to avoid orphans (#5879)

@NickGaertner

@NickGaertner Here's a fix, can you test this please?

I'd like to cut a new release tomorrow, so if you could get around to testing it until then, this would be great.

Kill the pty child's process tree on Windows to avoid orphans (#5879)

@stefanhaller

@stefanhaller Thanks for addressing the issue so quickly! I just tested the fix (fa5de1b) and unfortunately the conhost.exe orphans still persist. That said, I wasn't able to produce any git.exe or difft.exe orphans.
Please let me know if I can provide you with any additional information.

I can't reproduce conhost.exe orphans on my machine (neither can Claude with a test harness).

Could you answer a few more questions please:

What's your Windows build number? (Type ver in a cmd window to find out.)

After reproducing, could you paste the output of

Get-CimInstance Win32_Process -Filter "name='conhost.exe'" |
  Select-Object ProcessId,ParentProcessId,CreationDate,CommandLine

Do you get roughly one leaked conhost per fast keypress, or only occasionally after a burst of fast navigation? Does slow navigation (pausing a second on each commit) leak too?
Persistence: do the leaked conhosts ever disappear on their own (check again after ~10 minutes)?

Microsoft Windows [Version 10.0.22631.6199]

   ProcessId ParentProcessId CreationDate        CommandLine
   --------- --------------- ------------        -----------
        7124            7136 03/08/2026 12:30:09
        6760            6276 03/08/2026 12:30:09
        7176            7116 03/08/2026 12:30:09
        7204            6948 03/08/2026 12:30:09
        7284            7188 03/08/2026 12:30:09
       15696           20096 03/08/2026 12:48:05 \\?\C:\Windows\system32\conhost.exe --headless --width 106 --height 54 --signal 0x63c --server 0x688
       14440           20096 03/08/2026 12:48:05 \\?\C:\Windows\system32\conhost.exe --headless --width 106 --height 54 --signal 0x690 --server 0x614
       11580           20096 03/08/2026 12:51:08 \\?\C:\Windows\system32\conhost.exe --headless --width 106 --height 54 --signal 0x6e0 --server 0x6e8

- Not every single keypress causes a leak. For example, after quickly selecting 10 commits in a row I had two leftover conhost processes.
- I couldn't observe any leaks during slow navigation.
The leaked processes don't disappear on their own.

Additional info
I'm currently on a different machine than I was when I originally wrote the issue:

PowerShell: 7.6.3
git: 2.44.0.windows.1
Windows Terminal: 1.24.11911.0
lazygit: running fa5de1b from a checkout via go run .\main.go

fixup! Kill the pty child's process tree on Windows to avoid orphans (#…

Thanks; I pushed a fixup to the branch that may or may not change things, could you please test again?

I re-tested with e7c42cf but it didn't make a difference.

Kill the pty child's process tree on Windows to avoid orphans (#5879)

Reap the pty's conhost on Windows builds where it fails to run down (#…

Kill the pty child's process tree on Windows to avoid orphans (#5879)

Reap the pty's conhost on Windows builds where it fails to run down (#…

Reap pty process trees synchronously when quitting on Windows (#5879)

Kill the pty child's process tree on Windows to avoid orphans (#5879)

Reap the pty's conhost on Windows builds where it fails to run down (#…

Reap pty process trees synchronously when quitting on Windows (#5879)

I re-tested with e7c42cf but it didn't make a difference.

Ok, that was to be expected after all.

I reworked the branch to include a fix for the conhost orphans too. That was a different bug; the first one was really ours, but this one is a Windows bug that only occurs on older Windows versions. I added a workaround for it, so the orphans should be gone.

While working on this I discovered another problem: the index.lock problems that we fixed long ago could have been reintroduced with the PTY support. I don't think we have any reports for this yet, but that's probably just a matter of time. I made a separate PR to fix these (#5888); the PR you are testing is stacked on it, so it includes those fixes.

Can you retest again please? In addition to the flick-through-commits-quickly test, it would be useful to test the behavior of quitting lazygit while a very long diff is showing.

That seemed to do the trick. I tried to quickly switch commits and quit during diff calculation and wasn't able to create any orphans :) Thank you for your work!

Kill the pty child's process tree on Windows to avoid orphans (#5879)

Reap the pty's conhost on Windows builds where it fails to run down (#…

Reap pty process trees synchronously when quitting on Windows (#5879)

Kill the pty child's process tree on Windows to avoid orphans (#5879)

Reap the pty's conhost on Windows builds where it fails to run down (#…

Reap pty process trees synchronously when quitting on Windows (#5879)

Fix orphaned processes on Windows when quickly navigating between com…

Navigation Menu

Repository navigation

Orphans on Windows #5879

Describe the bug

To Reproduce

Expected behavior

Screenshots

Initial state - lazygit

After rapidly selecting different commits

After exiting lazygit

Version info:

Terminal info:

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Orphans on Windows #5879

Description

Describe the bug

To Reproduce

Expected behavior

Screenshots

Initial state - lazygit

After rapidly selecting different commits

After exiting lazygit

Version info:

Terminal info:

Activity

stefanhaller commented on Aug 2, 2026

NickGaertner commented on Aug 3, 2026

stefanhaller commented on Aug 3, 2026

NickGaertner commented on Aug 3, 2026

stefanhaller commented on Aug 3, 2026

NickGaertner commented on Aug 3, 2026

stefanhaller commented on Aug 3, 2026

NickGaertner commented on Aug 3, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions