From 68b17986c7c272d9be8df9a7abb9b162329b9d65 Mon Sep 17 00:00:00 2001 From: Jonas Schievink Date: Wed, 21 Oct 2020 19:54:04 +0200 Subject: Implement DNF-based `#[cfg]` introspection --- crates/cfg/Cargo.toml | 1 + crates/cfg/src/cfg_expr.rs | 35 +++- crates/cfg/src/dnf.rs | 477 +++++++++++++++++++++++++++++++++++++++++++++ crates/cfg/src/lib.rs | 110 +++++++++++ 4 files changed, 621 insertions(+), 2 deletions(-) create mode 100644 crates/cfg/src/dnf.rs (limited to 'crates/cfg') diff --git a/crates/cfg/Cargo.toml b/crates/cfg/Cargo.toml index a6785ee8e..c68e391c1 100644 --- a/crates/cfg/Cargo.toml +++ b/crates/cfg/Cargo.toml @@ -17,3 +17,4 @@ tt = { path = "../tt", version = "0.0.0" } [dev-dependencies] mbe = { path = "../mbe" } syntax = { path = "../syntax" } +expect-test = "1.0" diff --git a/crates/cfg/src/cfg_expr.rs b/crates/cfg/src/cfg_expr.rs index db3655b74..3f12a3fa8 100644 --- a/crates/cfg/src/cfg_expr.rs +++ b/crates/cfg/src/cfg_expr.rs @@ -2,12 +2,12 @@ //! //! See: https://doc.rust-lang.org/reference/conditional-compilation.html#conditional-compilation -use std::slice::Iter as SliceIter; +use std::{fmt, slice::Iter as SliceIter}; use tt::SmolStr; /// A simple configuration value passed in from the outside. -#[derive(Debug, Clone, PartialEq, Eq, Hash)] +#[derive(Debug, Clone, PartialEq, Eq, Hash, Ord, PartialOrd)] pub enum CfgAtom { /// eg. `#[cfg(test)]` Flag(SmolStr), @@ -18,6 +18,37 @@ pub enum CfgAtom { KeyValue { key: SmolStr, value: SmolStr }, } +impl CfgAtom { + /// Returns `true` when the atom comes from the target specification. + /// + /// If this returns `true`, then changing this atom requires changing the compilation target. If + /// it returns `false`, the atom might come from a build script or the build system. + pub fn is_target_defined(&self) -> bool { + match self { + CfgAtom::Flag(flag) => matches!(&**flag, "unix" | "windows"), + CfgAtom::KeyValue { key, value: _ } => matches!( + &**key, + "target_arch" + | "target_os" + | "target_env" + | "target_family" + | "target_endian" + | "target_pointer_width" + | "target_vendor" // NOTE: `target_feature` is left out since it can be configured via `-Ctarget-feature` + ), + } + } +} + +impl fmt::Display for CfgAtom { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + match self { + CfgAtom::Flag(name) => write!(f, "{}", name), + CfgAtom::KeyValue { key, value } => write!(f, "{} = \"{}\"", key, value), + } + } +} + #[derive(Debug, Clone, PartialEq, Eq)] pub enum CfgExpr { Invalid, diff --git a/crates/cfg/src/dnf.rs b/crates/cfg/src/dnf.rs new file mode 100644 index 000000000..35f946e6f --- /dev/null +++ b/crates/cfg/src/dnf.rs @@ -0,0 +1,477 @@ +//! Disjunctive Normal Form construction. +//! +//! Algorithm from , +//! which would have been much easier to read if it used pattern matching. It's also missing the +//! entire "distribute ANDs over ORs" part, which is not trivial. Oh well. +//! +//! This is currently both messy and inefficient. Feel free to improve, there are unit tests. + +use std::fmt; + +use rustc_hash::FxHashSet; + +use crate::{CfgAtom, CfgDiff, CfgExpr, CfgOptions, InactiveReason}; + +/// A `#[cfg]` directive in Disjunctive Normal Form (DNF). +pub struct DnfExpr { + conjunctions: Vec, +} + +impl DnfExpr { + pub fn new(expr: CfgExpr) -> Self { + let builder = Builder { expr: DnfExpr { conjunctions: Vec::new() } }; + + builder.lower(expr.clone()) + } + + /// Computes a list of present or absent atoms in `opts` that cause this expression to evaluate + /// to `false`. + /// + /// Note that flipping a subset of these atoms might be sufficient to make the whole expression + /// evaluate to `true`. For that, see `compute_enable_hints`. + /// + /// Returns `None` when `self` is already true, or contains errors. + pub fn why_inactive(&self, opts: &CfgOptions) -> Option { + let mut res = InactiveReason { enabled: Vec::new(), disabled: Vec::new() }; + + for conj in &self.conjunctions { + let mut conj_is_true = true; + for lit in &conj.literals { + let atom = lit.var.as_ref()?; + let enabled = opts.enabled.contains(atom); + if lit.negate == enabled { + // Literal is false, but needs to be true for this conjunction. + conj_is_true = false; + + if enabled { + res.enabled.push(atom.clone()); + } else { + res.disabled.push(atom.clone()); + } + } + } + + if conj_is_true { + // This expression is not actually inactive. + return None; + } + } + + res.enabled.sort_unstable(); + res.enabled.dedup(); + res.disabled.sort_unstable(); + res.disabled.dedup(); + Some(res) + } + + /// Returns `CfgDiff` objects that would enable this directive if applied to `opts`. + pub fn compute_enable_hints<'a>( + &'a self, + opts: &'a CfgOptions, + ) -> impl Iterator + 'a { + // A cfg is enabled if any of `self.conjunctions` evaluate to `true`. + + self.conjunctions.iter().filter_map(move |conj| { + let mut enable = FxHashSet::default(); + let mut disable = FxHashSet::default(); + for lit in &conj.literals { + let atom = lit.var.as_ref()?; + let enabled = opts.enabled.contains(atom); + if lit.negate && enabled { + disable.insert(atom.clone()); + } + if !lit.negate && !enabled { + enable.insert(atom.clone()); + } + } + + // Check that this actually makes `conj` true. + for lit in &conj.literals { + let atom = lit.var.as_ref()?; + let enabled = enable.contains(atom) + || (opts.enabled.contains(atom) && !disable.contains(atom)); + if enabled == lit.negate { + return None; + } + } + + if enable.is_empty() && disable.is_empty() { + return None; + } + + let mut diff = CfgDiff { + enable: enable.into_iter().collect(), + disable: disable.into_iter().collect(), + }; + + // Undo the FxHashMap randomization for consistent output. + diff.enable.sort_unstable(); + diff.disable.sort_unstable(); + + Some(diff) + }) + } +} + +impl fmt::Display for DnfExpr { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + if self.conjunctions.len() != 1 { + write!(f, "any(")?; + } + for (i, conj) in self.conjunctions.iter().enumerate() { + if i != 0 { + f.write_str(", ")?; + } + + write!(f, "{}", conj)?; + } + if self.conjunctions.len() != 1 { + write!(f, ")")?; + } + + Ok(()) + } +} + +struct Conjunction { + literals: Vec, +} + +impl Conjunction { + fn new(parts: Vec) -> Self { + let mut literals = Vec::new(); + for part in parts { + match part { + CfgExpr::Invalid | CfgExpr::Atom(_) | CfgExpr::Not(_) => { + literals.push(Literal::new(part)); + } + CfgExpr::All(conj) => { + // Flatten. + literals.extend(Conjunction::new(conj).literals); + } + CfgExpr::Any(_) => unreachable!("disjunction in conjunction"), + } + } + + Self { literals } + } +} + +impl fmt::Display for Conjunction { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + if self.literals.len() != 1 { + write!(f, "all(")?; + } + for (i, lit) in self.literals.iter().enumerate() { + if i != 0 { + f.write_str(", ")?; + } + + write!(f, "{}", lit)?; + } + if self.literals.len() != 1 { + write!(f, ")")?; + } + + Ok(()) + } +} + +struct Literal { + negate: bool, + var: Option, // None = Invalid +} + +impl Literal { + fn new(expr: CfgExpr) -> Self { + match expr { + CfgExpr::Invalid => Self { negate: false, var: None }, + CfgExpr::Atom(atom) => Self { negate: false, var: Some(atom) }, + CfgExpr::Not(expr) => match *expr { + CfgExpr::Invalid => Self { negate: true, var: None }, + CfgExpr::Atom(atom) => Self { negate: true, var: Some(atom) }, + _ => unreachable!("non-atom {:?}", expr), + }, + CfgExpr::Any(_) | CfgExpr::All(_) => unreachable!("non-literal {:?}", expr), + } + } +} + +impl fmt::Display for Literal { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + if self.negate { + write!(f, "not(")?; + } + + match &self.var { + Some(var) => write!(f, "{}", var)?, + None => f.write_str("")?, + } + + if self.negate { + write!(f, ")")?; + } + + Ok(()) + } +} + +struct Builder { + expr: DnfExpr, +} + +impl Builder { + fn lower(mut self, expr: CfgExpr) -> DnfExpr { + let expr = make_nnf(expr); + let expr = make_dnf(expr); + + match expr { + CfgExpr::Invalid | CfgExpr::Atom(_) | CfgExpr::Not(_) => { + self.expr.conjunctions.push(Conjunction::new(vec![expr])); + } + CfgExpr::All(conj) => { + self.expr.conjunctions.push(Conjunction::new(conj)); + } + CfgExpr::Any(mut disj) => { + disj.reverse(); + while let Some(conj) = disj.pop() { + match conj { + CfgExpr::Invalid | CfgExpr::Atom(_) | CfgExpr::All(_) | CfgExpr::Not(_) => { + self.expr.conjunctions.push(Conjunction::new(vec![conj])); + } + CfgExpr::Any(inner_disj) => { + // Flatten. + disj.extend(inner_disj.into_iter().rev()); + } + } + } + } + } + + self.expr + } +} + +fn make_dnf(expr: CfgExpr) -> CfgExpr { + match expr { + CfgExpr::Invalid | CfgExpr::Atom(_) | CfgExpr::Not(_) => expr, + CfgExpr::Any(e) => CfgExpr::Any(e.into_iter().map(|expr| make_dnf(expr)).collect()), + CfgExpr::All(e) => { + let e = e.into_iter().map(|expr| make_nnf(expr)).collect::>(); + + CfgExpr::Any(distribute_conj(&e)) + } + } +} + +/// Turns a conjunction of expressions into a disjunction of expressions. +fn distribute_conj(conj: &[CfgExpr]) -> Vec { + fn go(out: &mut Vec, with: &mut Vec, rest: &[CfgExpr]) { + match rest { + [head, tail @ ..] => match head { + CfgExpr::Any(disj) => { + for part in disj { + with.push(part.clone()); + go(out, with, tail); + with.pop(); + } + } + _ => { + with.push(head.clone()); + go(out, with, tail); + with.pop(); + } + }, + _ => { + // Turn accumulated parts into a new conjunction. + out.push(CfgExpr::All(with.clone())); + } + } + } + + let mut out = Vec::new(); + let mut with = Vec::new(); + + go(&mut out, &mut with, conj); + + out +} + +fn make_nnf(expr: CfgExpr) -> CfgExpr { + match expr { + CfgExpr::Invalid | CfgExpr::Atom(_) => expr, + CfgExpr::Any(expr) => CfgExpr::Any(expr.into_iter().map(|expr| make_nnf(expr)).collect()), + CfgExpr::All(expr) => CfgExpr::All(expr.into_iter().map(|expr| make_nnf(expr)).collect()), + CfgExpr::Not(operand) => match *operand { + CfgExpr::Invalid | CfgExpr::Atom(_) => CfgExpr::Not(operand.clone()), // Original negated expr + CfgExpr::Not(expr) => { + // Remove double negation. + make_nnf(*expr) + } + // Convert negated conjunction/disjunction using DeMorgan's Law. + CfgExpr::Any(inner) => CfgExpr::All( + inner.into_iter().map(|expr| make_nnf(CfgExpr::Not(Box::new(expr)))).collect(), + ), + CfgExpr::All(inner) => CfgExpr::Any( + inner.into_iter().map(|expr| make_nnf(CfgExpr::Not(Box::new(expr)))).collect(), + ), + }, + } +} + +#[cfg(test)] +mod test { + use expect_test::{expect, Expect}; + use mbe::ast_to_token_tree; + use syntax::{ast, AstNode}; + + use super::*; + + fn check_dnf(input: &str, expect: Expect) { + let (tt, _) = { + let source_file = ast::SourceFile::parse(input).ok().unwrap(); + let tt = source_file.syntax().descendants().find_map(ast::TokenTree::cast).unwrap(); + ast_to_token_tree(&tt).unwrap() + }; + let cfg = CfgExpr::parse(&tt); + let actual = format!("#![cfg({})]", DnfExpr::new(cfg)); + expect.assert_eq(&actual); + } + + fn check_why_inactive(input: &str, opts: &CfgOptions, expect: Expect) { + let (tt, _) = { + let source_file = ast::SourceFile::parse(input).ok().unwrap(); + let tt = source_file.syntax().descendants().find_map(ast::TokenTree::cast).unwrap(); + ast_to_token_tree(&tt).unwrap() + }; + let cfg = CfgExpr::parse(&tt); + let dnf = DnfExpr::new(cfg); + let why_inactive = dnf.why_inactive(opts).unwrap().to_string(); + expect.assert_eq(&why_inactive); + } + + #[track_caller] + fn check_enable_hints(input: &str, opts: &CfgOptions, expected_hints: &[&str]) { + let (tt, _) = { + let source_file = ast::SourceFile::parse(input).ok().unwrap(); + let tt = source_file.syntax().descendants().find_map(ast::TokenTree::cast).unwrap(); + ast_to_token_tree(&tt).unwrap() + }; + let cfg = CfgExpr::parse(&tt); + let dnf = DnfExpr::new(cfg); + let hints = dnf.compute_enable_hints(opts).map(|diff| diff.to_string()).collect::>(); + assert_eq!(hints, expected_hints); + } + + #[test] + fn smoke() { + check_dnf("#![cfg(test)]", expect![[r#"#![cfg(test)]"#]]); + check_dnf("#![cfg(not(test))]", expect![[r#"#![cfg(not(test))]"#]]); + check_dnf("#![cfg(not(not(test)))]", expect![[r#"#![cfg(test)]"#]]); + + check_dnf("#![cfg(all(a, b))]", expect![[r#"#![cfg(all(a, b))]"#]]); + check_dnf("#![cfg(any(a, b))]", expect![[r#"#![cfg(any(a, b))]"#]]); + + check_dnf("#![cfg(not(a))]", expect![[r#"#![cfg(not(a))]"#]]); + } + + #[test] + fn distribute() { + check_dnf("#![cfg(all(any(a, b), c))]", expect![[r#"#![cfg(any(all(a, c), all(b, c)))]"#]]); + check_dnf("#![cfg(all(c, any(a, b)))]", expect![[r#"#![cfg(any(all(c, a), all(c, b)))]"#]]); + check_dnf( + "#![cfg(all(any(a, b), any(c, d)))]", + expect![[r#"#![cfg(any(all(a, c), all(a, d), all(b, c), all(b, d)))]"#]], + ); + + check_dnf( + "#![cfg(all(any(a, b, c), any(d, e, f), g))]", + expect![[ + r#"#![cfg(any(all(a, d, g), all(a, e, g), all(a, f, g), all(b, d, g), all(b, e, g), all(b, f, g), all(c, d, g), all(c, e, g), all(c, f, g)))]"# + ]], + ); + } + + #[test] + fn demorgan() { + check_dnf("#![cfg(not(all(a, b)))]", expect![[r#"#![cfg(any(not(a), not(b)))]"#]]); + check_dnf("#![cfg(not(any(a, b)))]", expect![[r#"#![cfg(all(not(a), not(b)))]"#]]); + + check_dnf("#![cfg(not(all(not(a), b)))]", expect![[r#"#![cfg(any(a, not(b)))]"#]]); + check_dnf("#![cfg(not(any(a, not(b))))]", expect![[r#"#![cfg(all(not(a), b))]"#]]); + } + + #[test] + fn nested() { + check_dnf( + "#![cfg(all(any(a), not(all(any(b)))))]", + expect![[r#"#![cfg(all(a, not(b)))]"#]], + ); + + check_dnf("#![cfg(any(any(a, b)))]", expect![[r#"#![cfg(any(a, b))]"#]]); + check_dnf("#![cfg(not(any(any(a, b))))]", expect![[r#"#![cfg(all(not(a), not(b)))]"#]]); + check_dnf("#![cfg(all(all(a, b)))]", expect![[r#"#![cfg(all(a, b))]"#]]); + check_dnf("#![cfg(not(all(all(a, b))))]", expect![[r#"#![cfg(any(not(a), not(b)))]"#]]); + } + + #[test] + fn hints() { + let mut opts = CfgOptions::default(); + + check_enable_hints("#![cfg(test)]", &opts, &["enable test"]); + check_enable_hints("#![cfg(not(test))]", &opts, &[]); + + check_enable_hints("#![cfg(any(a, b))]", &opts, &["enable a", "enable b"]); + check_enable_hints("#![cfg(any(b, a))]", &opts, &["enable b", "enable a"]); + + check_enable_hints("#![cfg(all(a, b))]", &opts, &["enable a and b"]); + + opts.insert_atom("test".into()); + + check_enable_hints("#![cfg(test)]", &opts, &[]); + check_enable_hints("#![cfg(not(test))]", &opts, &["disable test"]); + } + + /// Tests that we don't suggest hints for cfgs that express an inconsistent formula. + #[test] + fn hints_impossible() { + let mut opts = CfgOptions::default(); + + check_enable_hints("#![cfg(all(test, not(test)))]", &opts, &[]); + + opts.insert_atom("test".into()); + + check_enable_hints("#![cfg(all(test, not(test)))]", &opts, &[]); + } + + #[test] + fn why_inactive() { + let mut opts = CfgOptions::default(); + opts.insert_atom("test".into()); + opts.insert_atom("test2".into()); + + check_why_inactive("#![cfg(a)]", &opts, expect![["a is disabled"]]); + check_why_inactive("#![cfg(not(test))]", &opts, expect![["test is enabled"]]); + + check_why_inactive( + "#![cfg(all(not(test), not(test2)))]", + &opts, + expect![["test and test2 are enabled"]], + ); + check_why_inactive( + "#![cfg(all(not(test), a))]", + &opts, + expect![["test is enabled and a is disabled"]], + ); + check_why_inactive( + "#![cfg(all(not(test), test2, a))]", + &opts, + expect![["test is enabled and a is disabled"]], + ); + check_why_inactive( + "#![cfg(all(not(test), not(test2), a))]", + &opts, + expect![["test and test2 are enabled and a is disabled"]], + ); + } +} diff --git a/crates/cfg/src/lib.rs b/crates/cfg/src/lib.rs index 35f540ac3..0b0734213 100644 --- a/crates/cfg/src/lib.rs +++ b/crates/cfg/src/lib.rs @@ -1,11 +1,15 @@ //! cfg defines conditional compiling options, `cfg` attibute parser and evaluator mod cfg_expr; +mod dnf; + +use std::fmt; use rustc_hash::FxHashSet; use tt::SmolStr; pub use cfg_expr::{CfgAtom, CfgExpr}; +pub use dnf::DnfExpr; /// Configuration options used for conditional compilition on items with `cfg` attributes. /// We have two kind of options in different namespaces: atomic options like `unix`, and @@ -40,4 +44,110 @@ impl CfgOptions { self.enabled.insert(atom.clone()); } } + + pub fn apply_diff(&mut self, diff: CfgDiff) { + for atom in diff.enable { + self.enabled.insert(atom); + } + + for atom in diff.disable { + self.enabled.remove(&atom); + } + } +} + +pub struct CfgDiff { + // Invariants: No duplicates, no atom that's both in `enable` and `disable`. + enable: Vec, + disable: Vec, +} + +impl CfgDiff { + /// Returns the total number of atoms changed by this diff. + pub fn len(&self) -> usize { + self.enable.len() + self.disable.len() + } +} + +impl fmt::Display for CfgDiff { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + if !self.enable.is_empty() { + f.write_str("enable ")?; + for (i, atom) in self.enable.iter().enumerate() { + let sep = match i { + 0 => "", + _ if i == self.enable.len() - 1 => " and ", + _ => ", ", + }; + f.write_str(sep)?; + + write!(f, "{}", atom)?; + } + + if !self.disable.is_empty() { + f.write_str("; ")?; + } + } + + if !self.disable.is_empty() { + f.write_str("disable ")?; + for (i, atom) in self.disable.iter().enumerate() { + let sep = match i { + 0 => "", + _ if i == self.enable.len() - 1 => " and ", + _ => ", ", + }; + f.write_str(sep)?; + + write!(f, "{}", atom)?; + } + } + + Ok(()) + } +} + +pub struct InactiveReason { + enabled: Vec, + disabled: Vec, +} + +impl fmt::Display for InactiveReason { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + if !self.enabled.is_empty() { + for (i, atom) in self.enabled.iter().enumerate() { + let sep = match i { + 0 => "", + _ if i == self.enabled.len() - 1 => " and ", + _ => ", ", + }; + f.write_str(sep)?; + + write!(f, "{}", atom)?; + } + let is_are = if self.enabled.len() == 1 { "is" } else { "are" }; + write!(f, " {} enabled", is_are)?; + + if !self.disabled.is_empty() { + f.write_str(" and ")?; + } + } + + if !self.disabled.is_empty() { + for (i, atom) in self.disabled.iter().enumerate() { + let sep = match i { + 0 => "", + _ if i == self.enabled.len() - 1 => " and ", + _ => ", ", + }; + f.write_str(sep)?; + + write!(f, "{}", atom)?; + } + let is_are = if self.disabled.len() == 1 { "is" } else { "are" }; + write!(f, " {} disabled", is_are)?; + } + + Ok(()) + } } -- cgit v1.2.3