Linus rejects late SPARC pull, rebukes rebase date tricks

Linus Torvalds has rejected a SPARC architecture pull for the Linux 7.4 merge window, telling maintainer Andreas Larsson to send the work for 7.5 instead. Torvalds objected that the request landed in the second week of the window and that the commits looked only hours old.

"The merge window is for things that were ready before the merge window, not for hurried last-minute updates," Torvalds wrote on the linux-kernel list.

Debian developer John Paul Adrian Glaubitz said the appearance was accidental: most patches had sat in Larsson's tree for weeks and in linux-next, but a rebase just before the pull had reset the commit dates. Larsson confirmed he had rebased only to insert a late Tested-by on a kernel thread stack-size change queued since September, apologized, and offered an alternate pull of the earlier history already carried by linux-next.

Glaubitz then proposed a git rebase recipe that amends commits while forcing the original author and committer dates to stick. Torvalds reacted sharply.

"STOP THIS DAMMIT! You are now telling people to LIE about what they do," he wrote. The fix is not to rebase pointlessly, and adding a Tested-by is "simply not a good enough reason to rebase long after-the-fact." He listed lying about commit times, rebasing mindlessly, and sending pulls for material that has not been in linux-next as practices to stop, and said he did not want to pull anything from Glaubitz given what he called a "completely broken mentality where it's ok to lie about rebases to hide the fact that you do it."

The deferred series spanned SPARC32 and SPARC64 work: seccomp support, fault-handling and fault-address fixes, cache flush corrections, an RCU stall workaround tied to the timer comparator, larger kernel thread stacks, and early memory setup cleanups that drop legacy bank construction.