Struct MergeSession 

pub struct MergeSession {
    pub merge_id: MergeSessionId,
    pub src_head: Option<OpId>,
    pub dst_head: Option<OpId>,
    pub lca: Option<OpId>,
    pub auto_resolved: Vec<MergeOutcome>,
    /* private fields */
}

Stateful merge in flight. Hold one per active merge between start and commit. Sessions are not thread-safe; the HTTP wrapper is expected to wrap them in a Mutex keyed by MergeSessionId.

Fields

merge_id: MergeSessionId

src_head: Option<OpId>

dst_head: Option<OpId>

lca: Option<OpId>

auto_resolved: Vec<MergeOutcome>

Outcomes the engine resolved unilaterally — Both (both sides agreed) and one-sided (Src / Dst). The agent sees these for audit but doesn’t need to act on them.

Implementations

impl MergeSession

pub fn start( merge_id: impl Into<MergeSessionId>, op_log: &OpLog, src_head: Option<&OpId>, dst_head: Option<&OpId>, ) -> Result<Self>

Start a merge session. Runs the engine in crate::merge and partitions the outcomes into auto-resolved and conflicts-needing-attention.

pub fn attach_file_conflicts( &mut self, conflicts: Vec<FileConflict>, needs_setfiles: bool, )

Attach the files dimension of the merge (#1007 PR 7): the conflicts a 3-way manifest diff surfaced (paths edited differently on both sides — see FileConflict) and whether the merge needs a SetFiles op at all. Called once, right after Self::start, by the caller that holds the store (the same layering ResolutionChecker uses: this crate tracks the session’s state machine, the caller computes the domain-specific diff). A no-op call with an empty conflicts and needs_setfiles: false — the default — leaves a session with no files dimension, exactly as before this feature existed.

pub fn needs_setfiles(&self) -> bool

Whether the merge needs a SetFiles op appended on commit, per MergeCommitOutput::needs_setfiles.

pub fn remaining_file_conflicts(&self) -> Vec<&FileConflict>

Pending file conflicts (those without a non-defer resolution). Mirrors Self::remaining_conflicts.

pub fn resolve_files( &mut self, resolutions: Vec<(FilePath, FileResolution)>, ) -> Vec<FileResolveVerdict>

Submit file resolutions in batch. Mirrors Self::resolve: unlike sig resolutions there is no type-check to run (SetFiles carries no program semantics), so this is the only resolve path files need — no resolve_files_checked counterpart.

pub fn remaining_conflicts(&self) -> Vec<&ConflictRecord>

Pending conflicts (those without a non-defer resolution).

pub fn resolve( &mut self, resolutions: Vec<(ConflictId, Resolution)>, ) -> Vec<ResolveVerdict>

Submit resolutions in batch. Returns one verdict per input. Accepted resolutions are recorded; rejected ones leave the previous resolution (if any) in place so partial submissions don’t clobber earlier good work.

pub fn resolve_checked( &mut self, resolutions: Vec<(ConflictId, Resolution)>, checker: &dyn ResolutionChecker, ) -> Vec<ResolveVerdict>

Submit resolutions in batch, type-checking each against the composed program before accepting it (#834).

This is the loop the session was built for — “submit N resolutions, see which broke type-checking, fix them, retry” — made real. Structural validation (Self::validate_resolution) runs first; a structurally-valid resolution is then overlaid on dst’s head together with every resolution accepted so far, and the injected ResolutionChecker type-checks the result. A resolution whose composed program doesn’t type-check is rejected with ResolutionRejection::TypeError and not recorded, so the session’s accepted set stays type-correct at every step.

Resolutions are processed in order and accumulate: a later resolution is checked against the program the earlier accepted ones already produced. Interdependent picks (two conflicts that only compose together) should therefore be submitted in dependency order, or a rejected one resubmitted after its partner lands — the same way git needs both halves of an intertwined conflict resolved before the tree builds. Unresolved conflicts contribute nothing to the projection: they leave dst’s (always-valid) side standing, so a partial batch still composes.

pub fn validate_resolution( &self, conflict_id: &ConflictId, resolution: &Resolution, ) -> Result<(), ResolutionRejection>

Validate a single resolution against the session’s pending conflicts. Pure (no side effects); the caller decides whether to accept.

pub fn commit(self) -> Result<MergeCommitOutput, CommitError>

Finalize the merge. On success returns the resolved sig and file resolutions, in id order, plus whether the caller must append a SetFiles. The caller is responsible for synthesizing the final Operation::Merge (and, if needs_setfiles, the follow-up SetFiles) op against the store and persisting them; this function returns the engine’s view of “what to land,” not the persisted op ids.

Sig conflicts are checked before file conflicts: a session with both kinds pending reports CommitError::ConflictsRemaining first, exactly the precedence #977’s merge gate already used for dependency conflicts vs. type errors — one blocker surfaced at a time keeps the agent’s retry loop simple.

Trait Implementations

impl Debug for MergeSession

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for MergeSession

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for MergeSession

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.