bcachefs Development Process

In the ongoing saga of Kent Overstreet and Linus Torvalds' differing approaches to kernel development, the story has taken a surprisingly straight path. The issue has been that Kent insists on submitting new feature patches at moments in the development cycle that are reserved for bug fixes and stabilization. He justifies this by saying that his patches are very important and that the rules should be flexible enough to accommodate him, while Linus and various others justify the current development cycle by saying that the stabilization period makes for better testing and debugging overall.

In general, Linus has tended to accept Kent's patches, though the conversation in response to those patches has been heated. The last time I wrote about this, Linus had asked for Kent's patches to come in at least a couple days before his preferred release day (Sundays), so the analysis bots (and humans, too) could have a moment to give them a shake before putting them into an official release candidate. Kent had agreed to submit patches earlier than Saturdays. So there was some movement, and peace descended on the valley.

This time, on June 20 (a Friday), Kent submitted, as he said, "Entirely too many patches." Among the pile of fixes, he included a "new option: journal_rewind," which "lets the entire filesystem be reset to an earlier point in time." As he described it, "this is only a disaster recovery tool, and right now there are major caveats to using it [...]. I'll likely be making some changes to the discard path in the future to make this a reliable recovery tool."

[...]