On Feb 29, 2012 8:05 AM, "Julian Foad" <[email protected]> wrote: > > Greg Stein wrote: > > I'm not sure this has anything to do with Ev2. The editor is for making the actual changes; ie. examine the merge sources, and if acceptable, then you call svn_editor_delete() to perform the deletion. The merge logic is in the *driver*, not the receiver. > > The way merge works is: client requests the source-side diff from the repo. Repo sends this diff (e.g. r10:20 branchA) via an 'editor'. Client (receiver of this edit drive[1]) knows about the target-branch (branchB) and how that differs from the base (r10 of branchA). Client figures out what changes to apply to the target branch -- which will be some of the incoming diff's changes, and some changes that aren't quite the same as in the incoming diff, and some other changes that the client decides it needs. Client applies these changes to the WC -- which it could theoretically do via another editor but it just does it directly. > > So the editor is being used to transmit a diff between two repo trees, not driven by merge code.
In your example you have two edits: 1) the holder of the state of branchA@10 is edited to make the state match branchA@20. (unrelated to Ev2, this series of edits is serialized and sent over the wire) 2) branchB is edited into some target state decided by the merge driver code. These edits are informed by the deserialized information from (1) plus "some other changes that the client decides it needs" The editor interface is not for merging. It is for *applying* the results of a merge, to a stateful representation of a tree to alter that held state. Cheers, -g
