implement string context

This commit is contained in:
2026-05-17 17:02:49 +08:00
parent 9a17990d5e
commit d98e389606
11 changed files with 698 additions and 224 deletions
+447
View File
@@ -0,0 +1,447 @@
//! `builtins.hasContext`, `builtins.getContext`, `builtins.appendContext`,
//! `builtins.unsafeDiscardStringContext`,
//! `builtins.unsafeDiscardOutputDependency`.
//!
//! See `fix-abstract-vm/src/string_context.rs` for the
//! `StringContextElem` type.
use fix_abstract_vm::{
AttrSet, BytecodeReader, List as VmList, Machine, MachineExt, NixString, NixType, Step,
StrictValue, StringContext, StringContextElem, Value, VmRuntimeCtx, VmRuntimeCtxExt,
};
use fix_builtins::PrimOpPhase;
use fix_common::StringId;
use fix_error::Error;
use gc_arena::{Gc, Mutation};
use smallvec::SmallVec;
pub fn has_context<'gc, M: Machine<'gc>>(
m: &mut M,
ctx: &mut impl VmRuntimeCtx,
reader: &mut BytecodeReader<'_>,
mc: &Mutation<'gc>,
) -> Step {
let val = m.force_and_retry::<StrictValue>(reader, mc)?;
if !val.is::<StringId>() && val.as_gc::<NixString>().is_none() {
return m.finish_type_err(NixType::String, val.ty());
}
let has_ctx = !ctx.get_string_context(val).is_empty();
m.return_from_primop(Value::new_inline(has_ctx), reader)
}
pub fn unsafe_discard_string_context<'gc, M: Machine<'gc>>(
m: &mut M,
ctx: &mut impl VmRuntimeCtx,
reader: &mut BytecodeReader<'_>,
mc: &Mutation<'gc>,
) -> Step {
let val = m.force_and_retry::<StrictValue>(reader, mc)?;
if let Some(sid) = val.as_inline::<StringId>() {
return m.return_from_primop(Value::new_inline(sid), reader);
}
let Some(ns) = val.as_gc::<NixString>() else {
return m.finish_type_err(NixType::String, val.ty());
};
let sid = ctx.intern_string(ns.as_str());
m.return_from_primop(Value::new_inline(sid), reader)
}
pub fn unsafe_discard_output_dependency<'gc, M: Machine<'gc>>(
m: &mut M,
ctx: &mut impl VmRuntimeCtx,
reader: &mut BytecodeReader<'_>,
mc: &Mutation<'gc>,
) -> Step {
let val = m.force_and_retry::<StrictValue>(reader, mc)?;
if let Some(sid) = val.as_inline::<StringId>() {
return m.return_from_primop(Value::new_inline(sid), reader);
}
let Some(ns) = val.as_gc::<NixString>() else {
return m.finish_type_err(NixType::String, val.ty());
};
if ns.context().is_empty() {
let sid = ctx.intern_string(ns.as_str());
return m.return_from_primop(Value::new_inline(sid), reader);
}
let mut new_ctx = StringContext::new();
for elem in ns.context() {
let replacement = match elem {
StringContextElem::DrvDeep { drv_path } => StringContextElem::Opaque {
path: drv_path.clone(),
},
other => other.clone(),
};
new_ctx.insert(replacement);
}
let s: Box<str> = ns.as_str().into();
let new_ns = Gc::new(mc, NixString::with_context(s, new_ctx));
m.return_from_primop(Value::new_gc(new_ns), reader)
}
pub fn get_context<'gc, M: Machine<'gc>>(
m: &mut M,
ctx: &mut impl VmRuntimeCtx,
reader: &mut BytecodeReader<'_>,
mc: &Mutation<'gc>,
) -> Step {
let val = m.force_and_retry::<StrictValue>(reader, mc)?;
if !val.is::<StringId>() && val.as_gc::<NixString>().is_none() {
return m.finish_type_err(NixType::String, val.ty());
}
let elems = ctx.get_string_context(val);
struct Info {
path: bool,
all_outputs: bool,
outputs: SmallVec<[Box<str>; 2]>,
}
impl Info {
fn new() -> Self {
Self {
path: false,
all_outputs: false,
outputs: SmallVec::new(),
}
}
}
let mut by_path: std::collections::BTreeMap<Box<str>, Info> = std::collections::BTreeMap::new();
for elem in elems {
match elem {
StringContextElem::Opaque { path } => {
by_path.entry(path.clone()).or_insert_with(Info::new).path = true;
}
StringContextElem::DrvDeep { drv_path } => {
by_path
.entry(drv_path.clone())
.or_insert_with(Info::new)
.all_outputs = true;
}
StringContextElem::Built { drv_path, output } => {
by_path
.entry(drv_path.clone())
.or_insert_with(Info::new)
.outputs
.push(output.clone());
}
}
}
let mut outer_entries: SmallVec<[(StringId, Value<'gc>); 4]> = SmallVec::new();
for (path, mut info) in by_path {
info.outputs.sort();
info.outputs.dedup();
let mut sub: SmallVec<[(StringId, Value<'gc>); 4]> = SmallVec::new();
if info.all_outputs {
sub.push((ctx.intern_string("allOutputs"), Value::new_inline(true)));
}
if !info.outputs.is_empty() {
let items: smallvec::SmallVec<[Value<'gc>; 4]> = info
.outputs
.iter()
.map(|o| Value::new_inline(ctx.intern_string(o)))
.collect();
let list = VmList::new(mc, items);
sub.push((ctx.intern_string("outputs"), Value::new_gc(list)));
}
if info.path {
sub.push((ctx.intern_string("path"), Value::new_inline(true)));
}
sub.sort_by_key(|(k, _)| *k);
let sub_attrs = Gc::new(mc, AttrSet::from_sorted_unchecked(sub));
outer_entries.push((ctx.intern_string(&path), Value::new_gc(sub_attrs)));
}
outer_entries.sort_by_key(|(k, _)| *k);
let outer = Gc::new(mc, AttrSet::from_sorted_unchecked(outer_entries));
m.return_from_primop(Value::new_gc(outer), reader)
}
/// appendContext :: String -> AttrSet -> String
/// The context AttrSet maps store-path strings to `{ path?: Bool, allOutputs?:
/// Bool, outputs?: [String] }`. Each present field contributes one
/// StringContextElem to the result.
///
/// Requires forcing nested attrset values and list elements lazily, so it's
/// structured as a state machine with the following stack layout:
///
/// [strVal, attrs, idx, acc] - outer loop
/// [strVal, attrs, idx, acc, entryAttrs] - after entry forced
/// [strVal, attrs, idx, acc, list] - after `outputs` forced
/// [strVal, attrs, idx, acc, list, oidx] - output-element loop
/// [strVal, attrs, idx, acc, list, oidx, outElem] - after element forced
///
/// `acc` is a sentinel `NixString` whose `data` is empty and whose `context`
/// is the accumulator. The string value itself is preserved in `strVal` and
/// retrieved at finalization.
///
// TODO: handle thunk-valued `path` and `allOutputs` sub-attrs; currently they
// must be already-evaluated booleans.
pub fn append_context<'gc, M: Machine<'gc>>(
m: &mut M,
ctx: &mut impl VmRuntimeCtx,
reader: &mut BytecodeReader<'_>,
mc: &Mutation<'gc>,
) -> Step {
let (str_val, attrs) = m.force_and_retry::<(StrictValue, Gc<AttrSet>)>(reader, mc)?;
let initial_ctx: StringContext = ctx.get_string_context(str_val).clone();
let acc = Gc::new(mc, NixString::with_context("", initial_ctx));
m.push(str_val.relax());
m.push(Value::new_gc(attrs));
m.push(Value::new_inline(0i32));
m.push(Value::new_gc(acc));
reader.set_pc(PrimOpPhase::AppendContextLoop.ip() as usize);
Step::Continue(())
}
pub fn append_context_loop<'gc, M: Machine<'gc>>(
m: &mut M,
ctx: &mut impl VmRuntimeCtx,
reader: &mut BytecodeReader<'_>,
mc: &Mutation<'gc>,
) -> Step {
#[allow(clippy::unwrap_used)]
let idx = m.peek(1).as_inline::<i32>().unwrap();
#[allow(clippy::unwrap_used)]
let attrs = m.peek_forced(2).as_gc::<AttrSet>().unwrap();
if idx as usize >= attrs.entries.len() {
return append_context_finalize(m, ctx, reader, mc);
}
let entry_val = attrs.entries[idx as usize].1;
m.push(entry_val);
m.force_slot_to_pc(
0,
reader,
mc,
PrimOpPhase::AppendContextEntryForced.ip() as usize,
)?;
reader.set_pc(PrimOpPhase::AppendContextEntryForced.ip() as usize);
Step::Continue(())
}
pub fn append_context_entry_forced<'gc, M: Machine<'gc>>(
m: &mut M,
ctx: &mut impl VmRuntimeCtx,
reader: &mut BytecodeReader<'_>,
mc: &Mutation<'gc>,
) -> Step {
// Stack: [strVal, attrs, idx, acc, entryAttrs(thunk)]
// The slot still holds the Thunk pointer; re-force to extract the now-
// Evaluated value into the slot.
m.force_slot(0, reader, mc)?;
let entry_val = m.peek_forced(0);
let Some(entry_attrs) = entry_val.as_gc::<AttrSet>() else {
return m.finish_type_err(NixType::AttrSet, entry_val.ty());
};
#[allow(clippy::unwrap_used)]
let idx = m.peek(2).as_inline::<i32>().unwrap();
#[allow(clippy::unwrap_used)]
let outer = m.peek_forced(3).as_gc::<AttrSet>().unwrap();
let path_key = outer.entries[idx as usize].0;
let path_str_owned: Box<str> = ctx.resolve_string(path_key).into();
if !path_str_owned.starts_with("/nix/store/") {
return m.finish_err(Error::eval_error(format!(
"context key '{path_str_owned}' is not a store path"
)));
}
// Eagerly handle `path` and `allOutputs` (assumed already-forced
// booleans - most callers either set them to literal `true` or omit
// them entirely).
// TODO: force these two attributes correctly
let path_id = ctx.intern_string("path");
let all_outputs_id = ctx.intern_string("allOutputs");
let outputs_id = ctx.intern_string("outputs");
#[allow(clippy::unwrap_used)]
let acc_gc = m.peek(1).as_gc::<NixString>().unwrap();
let mut new_acc: StringContext = acc_gc.context().iter().cloned().collect();
if let Some(v) = entry_attrs.lookup(path_id)
&& v.as_inline::<bool>() == Some(true)
{
new_acc.insert(StringContextElem::Opaque {
path: path_str_owned.clone(),
});
}
if let Some(v) = entry_attrs.lookup(all_outputs_id)
&& v.as_inline::<bool>() == Some(true)
{
if !path_str_owned.ends_with(".drv") {
return m.finish_err(Error::eval_error(format!(
"tried to add all-outputs context of {path_str_owned}, which is not a derivation, to a string"
)));
}
new_acc.insert(StringContextElem::DrvDeep {
drv_path: path_str_owned.clone(),
});
}
let new_acc_gc = Gc::new(mc, NixString::with_context("", new_acc));
m.replace(1, Value::new_gc(new_acc_gc));
if let Some(outputs_val) = entry_attrs.lookup(outputs_id) {
m.replace(0, outputs_val);
m.force_slot_to_pc(
0,
reader,
mc,
PrimOpPhase::AppendContextOutputsForced.ip() as usize,
)?;
reader.set_pc(PrimOpPhase::AppendContextOutputsForced.ip() as usize);
return Step::Continue(());
}
let _ = m.pop();
#[allow(clippy::unwrap_used)]
let idx_back = m.peek(1).as_inline::<i32>().unwrap();
m.replace(1, Value::new_inline(idx_back + 1));
reader.set_pc(PrimOpPhase::AppendContextLoop.ip() as usize);
Step::Continue(())
}
pub fn append_context_outputs_forced<'gc, M: Machine<'gc>>(
m: &mut M,
_ctx: &mut impl VmRuntimeCtx,
reader: &mut BytecodeReader<'_>,
mc: &Mutation<'gc>,
) -> Step {
m.force_slot(0, reader, mc)?;
let list_val = m.peek_forced(0);
let Some(list) = list_val.as_gc::<VmList>() else {
return m.finish_type_err(NixType::List, list_val.ty());
};
if list.inner.borrow().is_empty() {
// Stack: [strVal, attrs, idx, acc, list] -> drop list, bump idx.
let _ = m.pop();
#[allow(clippy::unwrap_used)]
let idx_back = m.peek(1).as_inline::<i32>().unwrap();
m.replace(1, Value::new_inline(idx_back + 1));
reader.set_pc(PrimOpPhase::AppendContextLoop.ip() as usize);
return Step::Continue(());
}
m.push(Value::new_inline(0i32));
reader.set_pc(PrimOpPhase::AppendContextOutputElementLoop.ip() as usize);
Step::Continue(())
}
pub fn append_context_output_element_loop<'gc, M: Machine<'gc>>(
m: &mut M,
_ctx: &mut impl VmRuntimeCtx,
reader: &mut BytecodeReader<'_>,
mc: &Mutation<'gc>,
) -> Step {
#[allow(clippy::unwrap_used)]
let oidx = m.peek(0).as_inline::<i32>().unwrap();
#[allow(clippy::unwrap_used)]
let list = m.peek_forced(1).as_gc::<VmList>().unwrap();
let len = list.inner.borrow().len();
if oidx as usize >= len {
// Stack: [strVal, attrs, idx, acc, list, oidx] -> drop oidx & list,
// bump idx in place.
let _ = m.pop();
let _ = m.pop();
#[allow(clippy::unwrap_used)]
let idx_back = m.peek(1).as_inline::<i32>().unwrap();
m.replace(1, Value::new_inline(idx_back + 1));
reader.set_pc(PrimOpPhase::AppendContextLoop.ip() as usize);
return Step::Continue(());
}
let elem = list.inner.borrow()[oidx as usize];
m.push(elem);
m.force_slot_to_pc(
0,
reader,
mc,
PrimOpPhase::AppendContextOutputElementForced.ip() as usize,
)?;
reader.set_pc(PrimOpPhase::AppendContextOutputElementForced.ip() as usize);
Step::Continue(())
}
pub fn append_context_output_element_forced<'gc, M: Machine<'gc>>(
m: &mut M,
ctx: &mut impl VmRuntimeCtx,
reader: &mut BytecodeReader<'_>,
mc: &Mutation<'gc>,
) -> Step {
m.force_slot(0, reader, mc)?;
let elem = m.peek_forced(0);
let Some(output_name) = ctx.get_string(elem) else {
return m.finish_type_err(NixType::String, elem.ty());
};
let output_name: Box<str> = output_name.into();
#[allow(clippy::unwrap_used)]
let idx = m.peek(4).as_inline::<i32>().unwrap();
#[allow(clippy::unwrap_used)]
let outer = m.peek_forced(5).as_gc::<AttrSet>().unwrap();
let path_key = outer.entries[idx as usize].0;
let path_str: Box<str> = ctx.resolve_string(path_key).into();
if !path_str.ends_with(".drv") {
return m.finish_err(Error::eval_error(format!(
"tried to add derivation output context of {path_str}, which is not a derivation, to a string"
)));
}
#[allow(clippy::unwrap_used)]
let acc_gc = m.peek(3).as_gc::<NixString>().unwrap();
let mut new_acc: StringContext = acc_gc.context().iter().cloned().collect();
new_acc.insert(StringContextElem::Built {
drv_path: path_str,
output: output_name,
});
let new_acc_gc = Gc::new(mc, NixString::with_context("", new_acc));
m.replace(3, Value::new_gc(new_acc_gc));
// Stack: [strVal, attrs, idx, acc, list, oidx, outElem] -> drop outElem,
// bump oidx in place.
let _ = m.pop();
#[allow(clippy::unwrap_used)]
let oidx = m.peek(0).as_inline::<i32>().unwrap();
m.replace(0, Value::new_inline(oidx + 1));
reader.set_pc(PrimOpPhase::AppendContextOutputElementLoop.ip() as usize);
Step::Continue(())
}
fn append_context_finalize<'gc, M: Machine<'gc>>(
m: &mut M,
ctx: &mut impl VmRuntimeCtx,
reader: &mut BytecodeReader<'_>,
mc: &Mutation<'gc>,
) -> Step {
// Stack: [strVal, attrs, idx, acc]
#[allow(clippy::unwrap_used)]
let acc_gc = m.pop().as_gc::<NixString>().unwrap();
let _ = m.pop(); // idx
let _ = m.pop(); // attrs
let str_val_raw = m.pop();
// The strVal was already forced at entry; restrict() is infallible here.
let str_val = str_val_raw
.restrict()
.unwrap_or_else(|_| panic!("appendContext: strVal unexpectedly a thunk"));
let s_str = ctx.get_string(str_val).unwrap_or("").to_owned();
let context: StringContext = acc_gc.context().iter().cloned().collect();
let result = if context.is_empty() {
let sid = ctx.intern_string(s_str);
Value::new_inline(sid)
} else {
let ns = Gc::new(mc, NixString::with_context(s_str, context));
Value::new_gc(ns)
};
m.return_from_primop(result, reader)
}
+13
View File
@@ -1,3 +1,4 @@
mod context;
mod control;
mod conv;
mod eq;
@@ -5,6 +6,7 @@ mod io;
mod list;
mod path;
pub use context::*;
pub use control::*;
pub use conv::*;
pub use eq::*;
@@ -69,6 +71,17 @@ pub fn dispatch_primop<'gc, M: Machine<'gc>>(
ToString => to_string(m, ctx, reader, mc),
TypeOf => type_of(m, ctx, reader, mc),
HasContext => has_context(m, ctx, reader, mc),
GetContext => get_context(m, ctx, reader, mc),
AppendContext => append_context(m, ctx, reader, mc),
AppendContextLoop => append_context_loop(m, ctx, reader, mc),
AppendContextEntryForced => append_context_entry_forced(m, ctx, reader, mc),
AppendContextOutputsForced => append_context_outputs_forced(m, ctx, reader, mc),
AppendContextOutputElementLoop => append_context_output_element_loop(m, ctx, reader, mc),
AppendContextOutputElementForced => append_context_output_element_forced(m, ctx, reader, mc),
UnsafeDiscardStringContext => unsafe_discard_string_context(m, ctx, reader, mc),
UnsafeDiscardOutputDependency => unsafe_discard_output_dependency(m, ctx, reader, mc),
phase => todo!("primop phase {phase:?}"),
}
}