Messages in this thread | | | Date | Fri, 29 Feb 2008 18:37:05 +0100 | From | Ingo Molnar <> | Subject | Re: [PATCH] x86_64 ia32 syscall restart fix |
| |
* Ingo Molnar <mingo@elte.hu> wrote:
> and one area where commit messages are totally important IMO is bug > forensics. For every regression we find we try to put in the commit ID > that broke it. Information like that is vital to have a good (and > objective) picture about how bugs get into and get out of the kernel > and it also alerts us to change/improve infrastructure if certain > categories of bugs happen too often.
another "commit space" feature Thomas and me was thinking about was to put in "backport suggestions" for -stable the following way:
Backport-suggested-by: Ingo Molnar <mingo@elte.hu>
and the -stable tree could then notice it, and once it has been backported, they could put in their "done" notifiers via:
Backported-from: 67ca7bde2e9d3516b5
or:
Backport-rejected: 67ca7bde2e9d3516b5
This way the act of suggesting backports to the -stable tree (and their rejection) could be fully automated, and the answer to the rather difficult question:
"has -stable picked up all backport requests, and if not, why?"
could be scripted up.
A further (small) variation of this scheme: if a fix is noticed to be a backport candidate later on, or a user notices that a fix that has gone upstream fixes a -stable bug too, this information could be signalled in a separate, special, empty commit:
Backport-suggested-by: 67ca7bde2e9d35, Ingo Molnar <mingo@elte.hu>
this way subsystem maintainers could have a reliable protocol of getting fixes integrated into -stable - purely via the commit messages in your tree.
... but then we decided that handling x86 architecture maintainance is work enough already, without us complicating our own life any further ;-)
But the idea is solid nevertheless, and if everyone did it the -stable guys would have a much easier life as well :-) [ We could start doing it in x86.git if there's general agreement and if the -stable guys specifically asked for this. ]
Ingo
| |