@asn mentioned on IRC a bug in libuidwrapper that was caused by a particular/weird implementation of libpam. I have no clue how to debug this problem.

Thanks for your comments. I saw the rekey test fail in other pipelines too, I think we have a probabilistic bug there, unrelated to this issue.

hi @Jakuje, I have updated the branch and I think I fixed all your concerns. I opened a separate ticket T200 for the remaining problems. I haven't tried the new CI image, I don't understand yet how to run the gitlab CI tests locally.

Hi @Jakuje, thanks for your review. All your comments are directly actionable so I'll fix them asap.
the two pkd tests that failed are related to my changes and bug T191 that I discovered this way, so I think we should merge it and acknowledge that there are unfixed bugs on our tree.
A few CI targets have two more failing tests. I didn't manage to reproduce them, I'm not sure if they did not exist before.
I'll see what I should do to have ubuntu as part of the CI targets. It makes totally sense because ubuntu is breaking every time I want to catch up on libssh dev :)

Could you please review my patches from https://gitlab.com/arisada/libssh-mirror/tree/fix-tests-ubuntu-tomerge ? There are fixes to this issue but also a testcase for T191.

I found the cause of the problem. The snprintf used to build the configuration file was reaching the maximum size of the buffer and was truncated, effectively removing the pidfile from the sshd configuration.
I'll push the commit on my review branch tomorrow.

In T137#2665, @adelton wrote:

I like the proposed wording, @Jakuje.

If the semantics get changed later, the documentation will change with it. But I'd prefer not waiting for the refactoring and just update the description now, as the first step.

That code is awfully old (probably 2004-2005). The idea was probably separating the SSH part and the SFTP parts. Today it makes little sense. Also both calls are synchronous and blocking. Your proposed patch describes well what it does today. Given the time I'd improve the actual implementation and move everything that handles network into sftp_init() and leave sftp_new() to memory allocation only like it's standard in the rest of the library.

Hi Jakub,

Closed it by mistake

cc @Jakuje, git blame says you touched that code last :)

I found this problem too when trying to set up the test environment on Ubuntu 18.04. The root cause is that the test calls the ssh binary from the fake root user. It fails because ssh can't find uid 0 in /etc/passwd, then fails because of missing known hosts keys and authentication keys. I don't know how that test could work elsewhere.
I made a patch that assumes it's running as root and create the missing keys. A better way would be to force the test cases to run as bob but I haven't found the proper way of doing this.

Thanks for your contribution!
I'll review this patch asap.

dh-gex kex was implemented and pushed to master.