Re: [PATCH v2 5/5] Reftable support for git-core
- From
Jeff King <peff@peff.net>
- Date
- Jan 30, 2020, 07:26 UTC
- Message-ID
- <20200130072612.GD2189233@coredump.intra.peff.net>
- In-Reply-To
- <xmqqh80e5cjk.fsf@gitster-ct.c.googlers.com>
On Wed, Jan 29, 2020 at 10:43:11AM -0800, Junio C Hamano wrote:
Show 21 quoted lines
> Jeff King <peff@peff.net> writes: > > > Making "refs" a file instead of a directory does work nicely, as any > > attempts to read or write would get ENOTDIR. And we can fool > > is_git_directory() as long as it's marked executable. That's OK on POSIX > > systems, but I'm not sure how it would work on Windows (or maybe it > > would work just fine, since we presumably just say "yep, everything is > > executable"). > > > > So perhaps that's enough, and what we put in HEAD won't matter (since > > nobody will be able to write into refs/ anyway). > > I wonder if it would help to take the "looser repository detection" > code alone and have it in a release, way before the rest of the > reftable topic is ready. Then by the time a repository created by a > reftable-enabled Git appears on people's disks, all the older > versions of Git that are still in people's hands would at least know > that it is a repository supported by future Git that they themselves > do not know how to handle, stop repository discovery correctly and > refrain from damaging the repository with an extension unknown to > them?
That helps, but it just slightly expands the window where we do the right thing. I.e., in a rollout like this:
1. Status quo: we only consider it a repo if "refs/" is executable
2. We introduce a version of Git that can read and write reftables.
3. We flip reftables support on by default.
Then any version of git after step 2 is fine. It's the ones from step 1 and older we care about. And there's probably some time between steps 2 and 3 to let the new versions percolate, shake out bugs, etc.
Adding a new step 1.5, "looser repository detection" expands the window of "good" versions by however long it takes to go from step 1.5 to step 2. But I expect that to be much smaller than the window between 2 and 3, so I'm not sure it makes a meaningful impact.
(Of course people who voluntarily turn on the feature as soon as they get hold of a step 2 version have no window at all, but presumably early adopters are OK with the risk).
-Peff