Struct rustc_query_system::dep_graph::graph::CurrentDepGraph

pub(super) struct CurrentDepGraph<K: DepKind> {
    encoder: Steal<GraphEncoder<K>>,
    new_node_to_index: Sharded<FxHashMap<DepNode<K>, DepNodeIndex>>,
    prev_index_to_index: Lock<IndexVec<SerializedDepNodeIndex, Option<DepNodeIndex>>>,
    fingerprints: Lock<FxHashMap<DepNode<K>, Fingerprint>>,
    forbidden_edge: Option<EdgeFilter<K>>,
    anon_id_seed: Fingerprint,
    total_read_count: AtomicU64,
    total_duplicate_read_count: AtomicU64,
    node_intern_event_id: Option<EventId>,
}

CurrentDepGraph stores the dependency graph for the current session. It will be populated as we run queries or tasks. We never remove nodes from the graph: they are only added.

The nodes in it are identified by a DepNodeIndex. We avoid keeping the nodes in memory. This is important, because these graph structures are some of the largest in the compiler.

For this reason, we avoid storing DepNodes more than once as map keys. The new_node_to_index map only contains nodes not in the previous graph, and we map nodes in the previous graph to indices via a two-step mapping. SerializedDepGraph maps from DepNode to SerializedDepNodeIndex, and the prev_index_to_index vector (which is more compact and faster than using a map) maps from SerializedDepNodeIndex to DepNodeIndex.

This struct uses three locks internally. The data, new_node_to_index, and prev_index_to_index fields are locked separately. Operations that take a DepNodeIndex typically just access the data field.

We only need to manipulate at most two locks simultaneously: new_node_to_index and data, or prev_index_to_index and data. When manipulating both, we acquire new_node_to_index or prev_index_to_index first, and data second.

Fields

encoder: Steal<GraphEncoder<K>>

new_node_to_index: Sharded<FxHashMap<DepNode<K>, DepNodeIndex>>

prev_index_to_index: Lock<IndexVec<SerializedDepNodeIndex, Option<DepNodeIndex>>>

fingerprints: Lock<FxHashMap<DepNode<K>, Fingerprint>>

This is used to verify that fingerprints do not change between the creation of a node and its recomputation.

forbidden_edge: Option<EdgeFilter<K>>

Used to trap when a specific edge is added to the graph. This is used for debug purposes and is only active with debug_assertions.

anon_id_seed: Fingerprint

Anonymous DepNodes are nodes whose IDs we compute from the list of their edges. This has the beneficial side-effect that multiple anonymous nodes can be coalesced into one without changing the semantics of the dependency graph. However, the merging of nodes can lead to a subtle problem during red-green marking: The color of an anonymous node from the current session might “shadow” the color of the node with the same ID from the previous session. In order to side-step this problem, we make sure that anonymous NodeIds allocated in different sessions don’t overlap. This is implemented by mixing a session-key into the ID fingerprint of each anon node. The session-key is just a random number generated when the DepGraph is created.

total_read_count: AtomicU64

These are simple counters that are for profiling and debugging and only active with debug_assertions.

total_duplicate_read_count: AtomicU64

node_intern_event_id: Option<EventId>

The cached event id for profiling node interning. This saves us from having to look up the event id every time we intern a node which may incur too much overhead. This will be None if self-profiling is disabled.

Implementations

impl<K: DepKind> CurrentDepGraph<K>

fn new(
    profiler: &SelfProfilerRef,
    prev_graph_node_count: usize,
    encoder: FileEncoder,
    record_graph: bool,
    record_stats: bool
) -> CurrentDepGraph<K>

fn record_edge(
    &self,
    dep_node_index: DepNodeIndex,
    key: DepNode<K>,
    fingerprint: Fingerprint
)

fn intern_new_node(
    &self,
    profiler: &SelfProfilerRef,
    key: DepNode<K>,
    edges: SmallVec<[DepNodeIndex; 8]>,
    current_fingerprint: Fingerprint
) -> DepNodeIndex

Writes the node to the current dep-graph and allocates a DepNodeIndex for it. Assumes that this is a node that has no equivalent in the previous dep-graph.

fn intern_node(
    &self,
    profiler: &SelfProfilerRef,
    prev_graph: &SerializedDepGraph<K>,
    key: DepNode<K>,
    edges: SmallVec<[DepNodeIndex; 8]>,
    fingerprint: Option<Fingerprint>,
    print_status: bool
) -> (DepNodeIndex, Option<(SerializedDepNodeIndex, DepNodeColor)>)

fn promote_node_and_deps_to_current(
    &self,
    profiler: &SelfProfilerRef,
    prev_graph: &SerializedDepGraph<K>,
    prev_index: SerializedDepNodeIndex
) -> DepNodeIndex

fn debug_assert_not_in_new_nodes(
    &self,
    prev_graph: &SerializedDepGraph<K>,
    prev_index: SerializedDepNodeIndex
)

Layout

Note: Most layout information is completely unstable and may even differ between compilations. The only exception is types with certain repr(...) attributes. Please see the Rust Reference’s “Type Layout” chapter for details on type layout guarantees.

Size: 536 bytes