On 1/7/2020 6:48 AM, Jeff King wrote: > The assertion itself is quite old, so I wondered if it was even still > relevant. Removing it does produce a reasonable-looking graph: As I'm digging into this case, and finding when the assertion is hit, I see that the issue is in the line further below your coloring issue: > | | | | * dd068b4 Merge commit '8f076d8' into HEAD > | |_|_|/| > |/| | |/ > | | |/| > | |/| | > | * | | 8f076d8 5 What is output is actually this, above. But the logic that includes the assert is checking where the underscores end, and the shown underscores actually pass the check. The issue is that it seems like it really wants to show this: > | | | | * dd068b4 Merge commit '8f076d8' into HEAD > | |_|_|/| > |/| |_|/ > | |/| | > | * | | 8f076d8 5 Note that I dropped a line and compressed a slash into an underscore. It's on that line that this condition is being hit. Now, is this really the intended behavior? Maybe. Looking at the previous tests in 54125-log-skewed-merges.sh, I don't see any where we have two skewed merges in the same merge. Thanks, -Stolee