// Ontological tree preorder ASG traversal
//
// Copyright (C) 2014-2023 Ryan Specialty, LLC.
//
// This file is part of TAME.
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program. If not, see .
//! Preorder traversal of ontological tree.
//!
//! This traversal is instantiated by [`tree_reconstruction`].
//!
//! This traversal should be used to reconstruct a source representation of
//! the compilation unit from the current state of the [`Asg`].
//!
//! Traversal Properties
//! ====================
//! This is a [depth-first search][w-depth-first-search]
//! visiting all nodes that are _reachable_ from the graph root
//! (see [`Asg::root`]).
//! [`ObjectIndex`]es are emitted in pre-order during the traversal,
//! and may be emitted more than once if
//! (a) they are the destination of cross edges or
//! (b) they are shared between trees
//! (most likely due to compiler optimizations).
//!
//! The tree is defined by the graph ontology,
//! not an arbitrary graph traversal.
//! This traversal is initialized by pushing each target [`ObjectIndex`] of
//! the ASG root
//! (see [`Asg::root`])
//! onto the stack.
//! Each iteration pops a single node off of the stack and visits it,
//! until no more nodes remain on the stack,
//! after which the traversal completes and the iterator is exhausted.
//! If the node was reached via a tree edge,
//! its edge targets are pushed onto the stack.
//! If a node is a target of a cross edge,
//! its edges targets are _not_ added to the stack for later traversal.
//!
//! Targets of a cross edge
//! (see [`ObjectRel::is_cross_edge`])
//! will be emitted multiple times:
//!
//! 1. The target of a cross edge is emitted each time a cross edge is
//! followed; and
//! 2. When the node is encountered on a tree edge.
//!
//! The traversal relies on the ontology to enforce a tree-like structure
//! and to properly define cross edges via `ObjectRel::is_cross_edge`.
//! A _tree edge_ is an edge that is not a cross edge.
//! Consequently,
//! if a cross edge is replaced by a tree edge,
//! then this traversal interprets that edge as part of _multiple_ trees,
//! effectively inlining it as if the user had entered the exact same
//! code in both locations.
//! You should choose carefully where in the lowering pipeline you wish
//! for this traversal to take place so that the tree reconstruction has
//! the desired properties.
//!
//! Because the graph is expected to be a DAG
//! (directed acyclic graph),
//! this traversal _does not track visited nodes_;
//! this ensures that nodes shared by trees due to optimizations like
//! common subexpression elimination will have proper trees
//! reconstructed.
//! If there are exceptional subgraphs where cycles do appear,
//! this traversal's implementation must be modified to take them into
//! account,
//! otherwise it will iterate indefinitely.
//!
//! Edges are visited in the same order that they were added to the graph,
//! so the tree reconstruction should match closely the order of the
//! source file.
//! However,
//! note that compiler passes,
//! if present,
//! may modify the graph beyond recognition,
//! though they should retain ordering where it is important.
//!
//! _Objects that do not have a path from the root will not be visited by
//! this traversal._
//! These objects are expected to act as additional metadata,
//! and must be queried for explicitly.
//! Such querying can be done during the traversal since this visitor holds
//! only a shared immutable reference to the [`Asg`].
//!
//! For more information,
//! see [`ObjectRel::is_cross_edge`].
//!
//! [w-depth-first-search]: https://en.wikipedia.org/wiki/Depth-first_search
//!
//! Depth Tracking
//! ==============
//! Each [`ObjectIndex`] emitted by this traversal is accompanied by a
//! [`Depth`] representing the length of the current path relative to the
//! [`Asg`] root.
//! Since the ASG root is never emitted,
//! the [`Depth`] value will always be ≥1.
//! Because nodes are always visited when an edge is followed,
//! a lower [`Depth`] will always be emitted prior to switching tree
//! branches.
//!
//! Let _S_ be an undirected spanning tree formed from the ontological tree.
//! At each iteration,
//! one of the following will be true:
//!
//! 1. [`Depth`] will increase by 1,
//! representing a tree edge or a cross edge;
//! 2. [`Depth`] will remain unchanged from the previous iteration,
//! representing a sibling node in the tree; or
//! 3. [`Depth`] will decrease by ≥1,
//! representing a back edge to an ancestor node on _S_.
//!
//! This depth information is the only means by which to reconstruct the
//! structure of the tree from the emitted [`ObjectIndex`]es.
//! For example,
//! if you are producing output in a nested format like XML,
//! an unchanged depth means that the current element should be closed
//! and a new one opened,
//! and you will close _one or more_ elements on a back edge.
//!
//! Note that,
//! because the [`Depth`] represents the current _path_,
//! the same [`ObjectIndex`] may be emitted multiple times with different
//! [`Depth`]s.
use std::fmt::Display;
use super::super::{object::DynObjectRel, Asg, Object, ObjectIndex};
use crate::{
parse::{self, Token},
span::{Span, UNKNOWN_SPAN},
};
// Re-export so that users of this API can avoid an awkward import from a
// completely different module hierarchy.
pub use crate::xir::flat::Depth;
#[cfg(doc)]
use super::super::object::ObjectRel;
/// Produce an iterator suitable for reconstructing a source tree based on
/// the contents of the [`Asg`].
///
/// The implementation of this traversal is exceedingly simple because of
/// its reliance on important graph invariants,
/// but it embodies a number of important and subtle properties.
///
/// See the [module-level documentation](super) for important information
/// about this traversal.
pub fn tree_reconstruction(asg: &Asg) -> TreePreOrderDfs {
TreePreOrderDfs::new(asg)
}
/// Pre-order depth-first search (DFS) using the ontological tree.
///
/// This DFS has an interesting property:
/// _it does not track visited nodes_,
/// relying instead on the ontology and recognition of cross edges to
/// produce the intended spanning tree.
/// An [`ObjectIndex`] that is the target of a cross edge will be output
/// more than once.
///
/// See [`tree_reconstruction`] for more information.
pub struct TreePreOrderDfs<'a> {
/// Reference [`Asg`].
///
/// Holding a reference to the [`Asg`] allows us to serve conveniently
/// as an iterator.
asg: &'a Asg,
/// DFS stack.
///
/// As objects (nodes/vertices) are visited,
/// its relationships (edges) are pushed onto the stack.
/// Each iterator pops a relationship off the stack and visits it.
///
/// The traversal ends once the stack becomes empty.
stack: Vec<(DynObjectRel, Depth)>,
}
/// Initial size of the DFS stack for [`TreePreOrderDfs`].
///
/// TODO: Derive a heuristic from our systems.
const TREE_INITIAL_STACK_SIZE: usize = 8;
impl<'a> TreePreOrderDfs<'a> {
fn new(asg: &'a Asg) -> Self {
let span = UNKNOWN_SPAN;
let mut dfs = Self {
asg,
stack: Vec::with_capacity(TREE_INITIAL_STACK_SIZE),
};
let root = asg.root(span);
dfs.push_edges_of(root.widen(), Depth::root());
dfs
}
fn push_edges_of(&mut self, oi: ObjectIndex