On 06/06/07, Yann Dirson [off-list ref] wrote:
I am however not happy at all with the way we delete patches and
series, starting with an existence check and then deleting. If any
error occurs midway, then we are left with an inconsistent state that
the user has to cleanup by hand. IMHO, we should have those methods
be as robust as possible, maybe starting by removing the formatversion
item, and printing a "cleaning up zombie stack" if does not find it.
So at least after fixing a "delete" bug, we could rerun the same
command and get to a sane state again.
This sounds OK for a quick fix. Longer term, I think we should support
some kind of transactions. One idea is to put the StGIT metadata in a
single file (or maybe two if we include the config) that gets checked
in after every operation.
--
Catalin