Week 7: Merge Conflict

Two true, incompatible histories, and what actually gets decided when a system runs out of authority to decide it

5 April 2027

Open the slides

A merge conflict is not an error. It is a system reporting, accurately, that it has reached a decision it has no basis to make. Both versions are valid. Both histories are true. They cannot both be the file.

What a merge can do without asking#

Most of a merge is arithmetic, and it works. Given a common ancestor and two descendants, a three-way merge asks a narrow question of each region of text: did one side change it, both sides, or neither? If one side changed a region and the other left it alone, the change applies: the untouched side expressed no preference, so there is nothing to weigh. Merges succeed silently and often, and this is the reason.

The conflict arises in exactly one case: both sides changed the same region, differently. Now the system has two candidate texts, no ancestor to defer to and, as week 8 will put it more carefully, no ground on which to prefer either. So it stops, writes both into the file with markers around them, and hands the region back:

<<<<<<< HEAD
the committee reviewed the findings
=======
the committee noted the findings
>>>>>>> revision

Those markers are the honest part of the operation. The system is declining to fabricate a decision it cannot support, and saying so in the file where the problem is.

What actually gets decided#

The technical resolution is trivial: pick lines, delete markers, save. That is not the decision. The decision is which of two intents survives, and the two texts above are not stylistic variants: reviewed and noted make different claims about what the committee did, and one of them will be what the record says happened.

Three properties of this decision are worth naming, because the tooling conceals all three. It is lossy: the discarded version does not go anywhere the reader will find. It is unattributed: the resolved file records who committed the merge, not who was overruled. And it is unexplained: no format has a field for why, so the reason lives in someone’s memory or nowhere. A resolved merge looks exactly like text that was never contested.

That is what this week’s lab is for, and the reason the Merge Conflict, In Writing assessment asks for prose rather than a file: the artefact a merge produces has no room in it for the only part that was difficult.

Where this goes#

Someone resolved that conflict, and the file now reads as though there was never a question. Next week asks what entitled them to decide, and what anyone’s claim to hold the real version actually rests on.