Files
nixjit/evaluator/nixjit_context/src/lib.rs
2025-09-14 17:39:57 +08:00

217 lines
8.4 KiB
Rust

use std::ptr::NonNull;
use bumpalo::{Bump, boxed::Box};
use hashbrown::HashMap;
use itertools::Itertools;
use petgraph::graphmap::DiGraphMap;
use nixjit_builtins::{
builtins::{GLOBAL_LEN, SCOPED_LEN}, BuiltinFn, Builtins, BuiltinsContext
};
use nixjit_error::{Error, Result};
use nixjit_eval::{Args, EvalContext, Value};
use nixjit_hir::{DowngradeContext, Hir};
use nixjit_ir::{AttrSet, ExprId, Param, PrimOpId};
use nixjit_lir::Lir;
use crate::downgrade::DowngradeCtx;
use crate::eval::EvalCtx;
use crate::resolve::ResolveCtx;
mod downgrade;
mod eval;
mod resolve;
/// The main evaluation context.
///
/// This struct orchestrates the entire Nix expression evaluation process,
/// from parsing and semantic analysis to interpretation and JIT compilation.
pub struct Context<'bump> {
ir_count: usize,
hirs: Vec<Hir>,
lirs: Vec<Box<'bump, Lir>>,
/// Maps a function's body `ExprId` to its parameter definition.
funcs: HashMap<ExprId, Param>,
repl_scope: NonNull<HashMap<String, ExprId>>,
global_scope: NonNull<HashMap<&'static str, ExprId>>,
/// A dependency graph between expressions.
graph: DiGraphMap<ExprId, ()>,
/// A table of primitive operation implementations.
primops: [(usize, BuiltinFn<Self>); GLOBAL_LEN + SCOPED_LEN],
bump: &'bump Bump,
}
impl Drop for Context<'_> {
fn drop(&mut self) {
// SAFETY: `repl_scope` and `global_scope` are `NonNull` pointers to `HashMap`s
// allocated within the `bump` arena. Because `NonNull` does not convey ownership,
// Rust's drop checker will not automatically drop the pointed-to `HashMap`s when
// the `Context` is dropped. We must manually call `drop_in_place` to ensure
// their destructors are run. This is safe because these pointers are guaranteed
// to be valid and non-null for the lifetime of the `Context`, as they are
// initialized in `new()` and never deallocated or changed.
unsafe {
self.repl_scope.drop_in_place();
self.global_scope.drop_in_place();
}
}
}
impl<'bump> Context<'bump> {
pub fn new(bump: &'bump Bump) -> Self {
let Builtins { global, scoped } = Builtins::new();
let global_scope = global
.iter()
.enumerate()
.map(|(idx, (k, _, _))| {
// SAFETY: The index `idx` comes from `enumerate()` on the `global` array,
// so it is guaranteed to be a valid, unique index for a primop LIR.
(*k, unsafe { ExprId::from_raw(idx) })
})
.chain(core::iter::once((
"builtins",
// SAFETY: This ID corresponds to the `builtins` attrset LIR, which is
// constructed and placed after all the global and scoped primop LIRs.
// The index is calculated to be exactly at that position.
unsafe { ExprId::from_raw(GLOBAL_LEN + SCOPED_LEN) },
)))
.collect();
let primops = global
.iter()
.map(|&(_, arity, f)| (arity, f))
.chain(scoped.iter().map(|&(_, arity, f)| (arity, f)))
.collect_array()
.unwrap();
let lirs = (0..global.len())
.map(|idx| {
// SAFETY: The index `idx` is guaranteed to be within the bounds of the
// `global` primops array, making it a valid raw ID for a `PrimOpId`.
Lir::PrimOp(unsafe { PrimOpId::from_raw(idx) })
})
.chain((0..scoped.len()).map(|idx| {
// SAFETY: The index `idx` is within the bounds of the `scoped` primops
// array. Adding `GLOBAL_LEN` correctly offsets it to its position in
// the combined `primops` table.
Lir::PrimOp(unsafe { PrimOpId::from_raw(idx + GLOBAL_LEN) })
}))
.chain(core::iter::once(Lir::AttrSet(AttrSet {
stcs: global
.into_iter()
.enumerate()
.map(|(idx, (name, ..))| {
// SAFETY: `idx` from `enumerate` is a valid index for the LIR
// corresponding to this global primop.
(name.to_string(), unsafe { ExprId::from_raw(idx) })
})
.chain(scoped.into_iter().enumerate().map(|(idx, (name, ..))| {
// SAFETY: `idx + GLOBAL_LEN` is a valid index for the LIR
// corresponding to this scoped primop.
(name.to_string(), unsafe {
ExprId::from_raw(idx + GLOBAL_LEN)
})
}))
.chain(core::iter::once((
"builtins".to_string(),
// SAFETY: This ID points to the `Thunk` that wraps this very
// `AttrSet`. The index is calculated to be one position after
// the `AttrSet` itself.
unsafe { ExprId::from_raw(GLOBAL_LEN + SCOPED_LEN + 1) },
)))
.collect(),
..AttrSet::default()
})))
.chain(core::iter::once(Lir::Thunk(
// SAFETY: This ID points to the `builtins` `AttrSet` defined just above.
// Its index is calculated to be at that exact position.
unsafe { ExprId::from_raw(GLOBAL_LEN + SCOPED_LEN) },
)))
.map(|lir| Box::new_in(lir, bump))
.collect_vec();
Self {
ir_count: lirs.len(),
hirs: Vec::new(),
lirs,
funcs: HashMap::new(),
global_scope: NonNull::from(bump.alloc(global_scope)),
repl_scope: NonNull::from(bump.alloc(HashMap::new())),
graph: DiGraphMap::new(),
primops,
bump,
}
}
pub fn downgrade_ctx<'a>(&'a mut self) -> DowngradeCtx<'a, 'bump> {
DowngradeCtx::new(self)
}
pub fn resolve_ctx<'a>(&'a mut self, root: ExprId) -> ResolveCtx<'a, 'bump> {
ResolveCtx::new(self, root)
}
pub fn eval_ctx<'a>(&'a mut self) -> EvalCtx<'a, 'bump> {
EvalCtx::new(self)
}
/// The main entry point for evaluating a Nix expression string.
///
/// This function performs the following steps:
/// 1. Parses the expression string into an `rnix` AST.
/// 2. Downgrades the AST to the High-Level IR (HIR).
/// 3. Resolves the HIR to the Low-Level IR (LIR).
/// 4. Evaluates the LIR to produce a final `Value`.
pub fn eval(&mut self, expr: &str) -> Result<nixjit_value::Value> {
let root = rnix::Root::parse(expr);
if !root.errors().is_empty() {
return Err(Error::parse_error(
root.errors().iter().map(|err| err.to_string()).join("; "),
));
}
let root = self
.downgrade_ctx()
.downgrade_root(root.tree().expr().unwrap())?;
let ctx = self.resolve_ctx(root);
ctx.resolve_root()?;
Ok(self.eval_ctx().eval(root)?.to_public())
}
pub fn add_binding(&mut self, ident: &str, expr: &str) -> Result<()> {
let root = rnix::Root::parse(expr);
let root_expr = root
.ok()
.map_err(|err| Error::parse_error(err.to_string()))?
.expr()
.unwrap();
let expr_id = self.downgrade_ctx().downgrade_root(root_expr)?;
self.resolve_ctx(expr_id).resolve_root()?;
// SAFETY: `repl_scope` is a `NonNull` pointer that is guaranteed to be valid
// for the lifetime of `Context`. It is initialized in `new()` and the memory
// it points to is managed by the `bump` arena. Therefore, it is safe to
// dereference it to a mutable reference here.
unsafe { self.repl_scope.as_mut() }.insert(ident.to_string(), expr_id);
Ok(())
}
}
impl Context<'_> {
fn alloc_id(&mut self) -> ExprId {
self.ir_count += 1;
// SAFETY: This function is the sole source of new `ExprId`s during the
// downgrade and resolve phases. By monotonically incrementing `ir_count`,
// we guarantee that each ID is unique and corresponds to a valid, soon-to-be-
// allocated slot in the IR vectors.
unsafe { ExprId::from_raw(self.ir_count - 1) }
}
fn add_dep(&mut self, from: ExprId, to: ExprId) {
self.graph.add_edge(from, to, ());
}
}
impl BuiltinsContext for Context<'_> {}