"Rules for Pull: Symmetric to pull..." sounds like tautology. I think push is right here.
Add link to original discussion.
|Deletions are marked like this.||Additions are marked like this.|
|Line 4:||Line 4:|
Original proposal: [[http://markmail.org/message/vyyjfyshhnb5acpc|RFC: dealing with dead, anonymous feature branches]]
Original proposal: RFC: dealing with dead, anonymous feature branches
We will use the following definitions:
- poisoned changeset
a changeset with poisoned=1 in its extra dict.
- dead head
- a poisoned changeset that is also a head.
- dead changeset
- a changeset from which you can only reach dead heads.
- dead branch
- a set of dead changesets
2. Rules for Push
When pushing changesets, the following two rules are observed:
- First, send changesets that are not on a dead branch.
- Then, send changesets descending from what is now remote heads.
The discovery protocols gives us the heads of the common set of nodes. From these we can compute the changesets that are not on a dead branch (rule 1). We can then look at the remaining heads among the nodes in the common set and push descendants of these.
No new information is neede by this protocol, everything can be computed locally from information about the common set of nodes.
3. Rules for Pull
Symmetric to push, but the server needs to filter the changesets.
4. Use Cases
An experiment can be stopped, whether it has been pushed to another server or not:
A new commit is made:
A pull is made, bringing ind C, D, and E and it is decided that X is no longer is relevant: