It still fails like this for me:

Clearly nobody is running this in their CI. :(

In D9387#141293, @mharbison72 wrote:

If this is purely from --config instead of a file and you're collecting metrics, would you want this to be an InputError? (I'm happy to queue this as-is, I just wanted to double check)

In D9341#141016, @sebhtml wrote:

In D9341#140899, @martinvonz wrote:

Do test-lfs-serve.t and test-transaction-rollback-on-sigpipe.t consistently fail with this patch? If they do, you should fix them :) If they don't (i.e. it was some flakiness you were experiencing), just don't mention that in the commit message and instead ask on #mercurial or send an email to mercurial-devel to see if others have seen the flakiness and ask how to fix it.

I think that the problem is that the 2 failing tests of Mercurial are not portable on FreeBSD 12.2.

In D9341#140972, @sebhtml wrote:

The test failure has a diff. I still think that's just noise that will confuse much more than it helps. Okay if I drop the test plan from the commit message in flight?

Sure. In my opinion, the test plan is not part of my commit message.
I am currently running the tests in a VM-BHYVE running Debian Linux 10. I except all tests to pass.
I will keep you posted.

In D9341#140969, @sebhtml wrote:

That's what I meant: you need to indent diffs that you include in the commit message, otherwise tools can't figure out where the commit message ends and the diff starts. But I still think you can just remove that whole thing from the commit message; it's implied that the test plan is to run the test suite, so you only need to describe your test plan if you did some other testing and/or did not run the test suite.

I did not include any diff in my commit message, as far as I know.
Given what you recommended, I added the prefix OUTPUT: in my test results in my test plan.

In D9341#140944, @sebhtml wrote:

Also, including diff output (such as from failing tests) without indentation makes the patch fail to apply:
unable to find 'tank/home/sebhtml/projects/mercurial/sebhtml-hg-log-bookmark/hg-committed/tests/test-lfs-serve.t' for patching

I think that the problem is that "hg phabimport" is fetching what I wrote in the test plan. I did not commit anything to tests/test-lfs-serve.t
This is weird.

In D9341#140899, @martinvonz wrote:

Do test-lfs-serve.t and test-transaction-rollback-on-sigpipe.t consistently fail with this patch? If they do, you should fix them :) If they don't (i.e. it was some flakiness you were experiencing), just don't mention that in the commit message and instead ask on #mercurial or send an email to mercurial-devel to see if others have seen the flakiness and ask how to fix it.

Do test-lfs-serve.t and test-transaction-rollback-on-sigpipe.t consistently fail with this patch? If they do, you should fix them :) If they don't (i.e. it was some flakiness you were experiencing), just don't mention that in the commit message and instead ask on #mercurial or send an email to mercurial-devel to see if others have seen the flakiness and ask how to fix it.