Merge pull request #379 from swgillespie/zero_div_zero_nan

implement 0.0/0.0 -> NaN lint as described in #370

src/lib.rs

@@ -47,6 +47,7 @@ pub mod loops;
 pub mod ranges;
 pub mod matches;
 pub mod precedence;
+pub mod zero_div_zero;
 
 mod reexport {
     pub use syntax::ast::{Name, Ident, NodeId};
@@ -88,6 +89,7 @@ pub fn plugin_registrar(reg: &mut Registry) {
     reg.register_late_lint_pass(box matches::MatchPass);
     reg.register_late_lint_pass(box misc::PatternPass);
     reg.register_late_lint_pass(box minmax::MinMaxPass);
+    reg.register_late_lint_pass(box zero_div_zero::ZeroDivZeroPass);
 
     reg.register_lint_group("clippy_pedantic", vec![
         methods::OPTION_UNWRAP_USED,
@@ -152,5 +154,6 @@ pub fn plugin_registrar(reg: &mut Registry) {
         types::TYPE_COMPLEXITY,
         types::UNIT_CMP,
         unicode::ZERO_WIDTH_SPACE,
+        zero_div_zero::ZERO_DIVIDED_BY_ZERO,
     ]);
 }

src/zero_div_zero.rs

@@ -0,0 +1,50 @@
+use rustc::lint::*;
+use rustc_front::hir::*;
+
+use utils::{span_help_and_lint};
+use consts::{Constant, constant_simple, FloatWidth};
+
+/// ZeroDivZeroPass is a pass that checks for a binary expression that consists
+/// of 0.0/0.0, which is always NaN. It is more clear to replace instances of
+/// 0.0/0.0 with std::f32::NaN or std::f64::NaN, depending on the precision.
+pub struct ZeroDivZeroPass;
+
+declare_lint!(pub ZERO_DIVIDED_BY_ZERO, Warn,
+              "usage of `0.0 / 0.0` to obtain NaN instead of std::f32::NaN or std::f64::NaN");
+
+impl LintPass for ZeroDivZeroPass {
+    fn get_lints(&self) -> LintArray {
+        lint_array!(ZERO_DIVIDED_BY_ZERO)
+    }
+}
+
+impl LateLintPass for ZeroDivZeroPass {
+    fn check_expr(&mut self, cx: &LateContext, expr: &Expr) {
+        // check for instances of 0.0/0.0
+        if_let_chain! {
+            [
+                let ExprBinary(ref op, ref left, ref right) = expr.node,
+                let BinOp_::BiDiv = op.node,
+                // TODO - constant_simple does not fold many operations involving floats.
+                // That's probably fine for this lint - it's pretty unlikely that someone would
+                // do something like 0.0/(2.0 - 2.0), but it would be nice to warn on that case too.
+                let Some(Constant::ConstantFloat(ref lhs_value, lhs_width)) = constant_simple(left),
+                let Some(Constant::ConstantFloat(ref rhs_value, rhs_width)) = constant_simple(right),
+                let Some(0.0) = lhs_value.parse().ok(),
+                let Some(0.0) = rhs_value.parse().ok()
+            ],
+            {
+                // since we're about to suggest a use of std::f32::NaN or std::f64::NaN,
+                // match the precision of the literals that are given.
+                let float_type = match (lhs_width, rhs_width) {
+                    (FloatWidth::Fw64, _)
+                    | (_, FloatWidth::Fw64) => "f64",
+                    _ => "f32"
+                };
+                span_help_and_lint(cx, ZERO_DIVIDED_BY_ZERO, expr.span,
+                    "constant division of 0.0 with 0.0 will always result in NaN",
+                    &format!("Consider using `std::{}::NAN` if you would like a constant representing NaN", float_type));
+            }
+        }
+    }
+}