Failure when receiving data from the peer

[melnele]

• I was not able to find an open or closed issue matching what I'm seeing

Setup

• Which version of Git for Windows are you using?
  64-bit

$ git --version --build-options
Git-2.39.0.2-64-bit.exe

• Which version of Windows are you running?
  Windows 11 64-bit

$ cmd.exe /c ver
Microsoft Windows [Version 10.0.22621.963]

• What options did you set as part of the installation? Or did you choose the
  defaults?

Editor Option: Notepad++
Custom Editor Path: 
Default Branch Option:  
Path Option: Cmd
SSH Option: OpenSSH
Tortoise Option: false
CURL Option: WinSSL
CRLF Option: CRLFAlways
Bash Terminal Option: MinTTY
Git Pull Behavior Option: Merge
Use Credential Manager: Enabled
Performance Tweaks FSCache: Enabled
Enable Symlinks: Disabled
Enable Pseudo Console Support: Disabled
Enable FSMonitor: Disabled

Details

• Which terminal/shell are you running Git from?
  Bash/CMD

• What commands did you run to trigger this issue? If you can provide a
  Minimal, Complete, and Verifiable example
  this will help us understand the issue.

git fetch
or
git push

• What actually happened instead?

  fatal: unable to access 'https://bitbucket.org/*/*.git/': Failure when receiving data from the peer

I then downgraded to git version 2.38.1.windows.1 and it worked

[dscho]

I then downgraded to git version 2.38.1.windows.1 and it worked

@melnele That's a large step, testing only v2.39.0(2) and v2.38.1.

Could you bisect through the snapshots to figure out what causes the regression? My guess is that it's the cURL update, and only v2.39.0(2) has the regression, but not the preceding snapshot (could you also test this snapshot which sadly failed to upload?).

[melnele]

@dscho you are correct I just instaled Git-2.39.0-64-bit.exe and it's working

[melnele]

(could you also test this snapshot which sadly failed to upload?).

@dscho I tested it and it's working

[dscho]

@melnele would you have any public URL with which I could try to reproduce the problem?

[melnele]

@dscho unfortunately the repos I'm working on are private on bitbucket
but I found a similar repo that is public

$ git clone https://bitbucket.org/anand980980/angular-9-with-angular-material.git

Cloning into 'angular-9-with-angular-material'...
error: RPC failed; curl 56 Failure when receiving data from the peer
Receiving objects: 100% (276/276), 351.28 KiB | 524.00 KiB/s, done.
Resolving deltas: 100% (142/142), done.

[dscho]

$ git clone https://bitbucket.org/anand980980/angular-9-with-angular-material.git

That reproduces here, but only with the Secure Channel backend. It does work with the OpenSSL backend. That suggests that this is a Secure Channel-specific regression in cURL.

I won't be able to investigate or drive this any further in the next few weeks, so @melnele please use this as a workaround if you want to upgrade: git config --global http.sslBackend openssl.

[melnele]

This indeed works thank you for your help.

[Nashrah-Fatima]

fatal: unable to connect to github.com:
github.com[0: 20.207.73.82]: errno=Unknown error
This error I got in gitbash

[dscho]

$ git clone https://bitbucket.org/anand980980/angular-9-with-angular-material.git

That reproduces here, but only with the Secure Channel backend. It does work with the OpenSSL backend. That suggests that this is a Secure Channel-specific regression in cURL.

I just tried the same command with a custom build of libcurl-4.dll off of https://github.com/curl/curl's tip revision and it did not fail (read: it seems that the regression might have been fixed). Next stop: I'm building libcurl-4.dll off of the curl-7_87_0 tag. If that shows the regression, I will bisect to figure out whether we can quickly do a backport (seeing as no hot-fix version of cURL was released).

[dscho]

It was a bit harder to bisect this than expected because somewhere in the middle the regression became a flake, and only later did it get fixed. After testing a while I am pretty confident that the regression is fixed by curl/curl@54120ef. I'll test with this a bit more, and pending that testing I will open a PR and get fixed Pacman packages deployed.

[dscho]

/add release note bug A regression introduced in Git for Windows v2.39.0(2) that prevented cloning from Bitbucket was fixed.

The workflow run was started

[dscho]

@melnele @jurko-gospodnetic please test the latest snapshot.

[melnele]

@melnele @jurko-gospodnetic please test the latest snapshot.

@dscho
I switched back to
git config --global http.sslBackend schannel
and it's working.

[dscho]

@melnele @jurko-gospodnetic please test the latest snapshot.

@dscho
I switched back to
git config --global http.sslBackend schannel
and it's working.

Thank you for confirming the fix!

[kandersondaxsonics]

Thank you for this fix. I apologize for not filing this earlier myself. I figured it was something local to my machine and/or network, not something more global, as I was the only one in my office having the issue. I was apparently the only one on the December release! I will also confirm that the latest release (2.39.1) fixed this for me.

The really frustrating part is that it would work about 1/10 of the time, so it would usually fail, but not always.

[dscho]

The really frustrating part is that it would work about 1/10 of the time, so it would usually fail, but not always.

Yep, I also found that very frustrating during my bisecting.

[jurko-gospodnetic]

On 12.1.2023. 21:51, Johannes Schindelin wrote: @melnele <https://github.com/melnele> @jurko-gospodnetic <https://github.com/jurko-gospodnetic> please test the latest snapshot <https://wingit.blob.core.windows.net/files/index.html>.

Only now got to this and yeah, the problem seems to have been resolved. Thanks! Best regards, Jurko Gospodnetić