Auto merge of #58475 - alex:usize-overflow-panic, r=<try>

[WIP] Make usize overflow always have debug-assertions semantics

This is a prototype for rust-lang/rfcs#2635 to enable us to collect some measurements.

src/librustc/ty/query/on_disk_cache.rs

@@ -105,22 +105,21 @@ impl AbsoluteBytePos {
 
 impl<'sess> OnDiskCache<'sess> {
     /// Creates a new OnDiskCache instance from the serialized data in `data`.
-    pub fn new(sess: &'sess Session, data: Vec<u8>, start_pos: usize) -> OnDiskCache<'sess> {
+    pub fn new(sess: &'sess Session, data: Vec<u8>) -> OnDiskCache<'sess> {
         debug_assert!(sess.opts.incremental.is_some());
 
         // Wrapping in a scope so we can borrow `data`
         let footer: Footer = {
-            let mut decoder = opaque::Decoder::new(&data[..], start_pos);
-
             // Decode the *position* of the footer which can be found in the
             // last 8 bytes of the file.
-            decoder.set_position(data.len() - IntEncodedWithFixedSize::ENCODED_SIZE);
+            let mut decoder = opaque::Decoder::new(
+                &data, data.len() - IntEncodedWithFixedSize::ENCODED_SIZE);
             let query_result_index_pos = IntEncodedWithFixedSize::decode(&mut decoder)
                 .expect("Error while trying to decode query result index position.")
                 .0 as usize;
 
             // Decoder the file footer which contains all the lookup tables, etc.
-            decoder.set_position(query_result_index_pos);
+            decoder = opaque::Decoder::new(&data, query_result_index_pos);
             decode_tagged(&mut decoder, TAG_FILE_FOOTER)
                 .expect("Error while trying to decode query result index position.")
         };
@@ -540,7 +539,7 @@ impl<'a, 'tcx: 'a, 'x> ty_codec::TyDecoder<'a, 'tcx> for CacheDecoder<'a, 'tcx,
 
     #[inline]
     fn peek_byte(&self) -> u8 {
-        self.opaque.data[self.opaque.position()]
+        self.opaque.data()[0]
     }
 
     fn cached_ty_for_shorthand<F>(&mut self,
@@ -569,9 +568,9 @@ impl<'a, 'tcx: 'a, 'x> ty_codec::TyDecoder<'a, 'tcx> for CacheDecoder<'a, 'tcx,
     fn with_position<F, R>(&mut self, pos: usize, f: F) -> R
         where F: FnOnce(&mut Self) -> R
     {
-        debug_assert!(pos < self.opaque.data.len());
+        debug_assert!(pos < self.opaque.original_data.len());
 
-        let new_opaque = opaque::Decoder::new(self.opaque.data, pos);
+        let new_opaque = opaque::Decoder::new(&self.opaque.original_data, pos);
         let old_opaque = mem::replace(&mut self.opaque, new_opaque);
         let r = f(self);
         self.opaque = old_opaque;

src/librustc_codegen_llvm/context.rs

@@ -12,13 +12,15 @@ use crate::type_of::PointeeInfo;
 use rustc_codegen_ssa::traits::*;
 use libc::c_uint;
 
+use syntax::ast;
+
 use rustc_data_structures::base_n;
 use rustc_data_structures::small_c_str::SmallCStr;
 use rustc::mir::mono::Stats;
 use rustc::session::config::{self, DebugInfo};
 use rustc::session::Session;
 use rustc::ty::layout::{LayoutError, LayoutOf, Size, TyLayout, VariantIdx};
-use rustc::ty::{self, Ty, TyCtxt};
+use rustc::ty::{self, Ty, TyCtxt, TyKind};
 use rustc::util::nodemap::FxHashMap;
 use rustc_target::spec::{HasTargetSpec, Target};
 use rustc_codegen_ssa::callee::resolve_and_get_fn;
@@ -409,8 +411,14 @@ impl MiscMethods<'tcx> for CodegenCx<'ll, 'tcx> {
         &self.tcx.sess
     }
 
-    fn check_overflow(&self) -> bool {
-        self.check_overflow
+    fn check_overflow(&self, ty: Option<Ty<'tcx>>) -> bool {
+        let type_specific_overflow = match ty {
+            Some(ty) => {
+                ty.sty == TyKind::Uint(ast::UintTy::Usize)
+            },
+            None => false
+        };
+        self.check_overflow || type_specific_overflow
     }
 
     fn stats(&self) -> &RefCell<Stats> {

src/librustc_codegen_ssa/mir/block.rs

@@ -339,7 +339,7 @@ impl<'a, 'tcx: 'a, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                 // NOTE: Unlike binops, negation doesn't have its own
                 // checked operation, just a comparison with the minimum
                 // value, so we have to check for the assert message.
-                if !bx.check_overflow() {
+                if !bx.check_overflow(None) {
                     if let mir::interpret::EvalErrorKind::OverflowNeg = *msg {
                         const_cond = Some(expected);
                     }

src/librustc_codegen_ssa/mir/rvalue.rs

@@ -670,7 +670,7 @@ impl<'a, 'tcx: 'a, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
         // with #[rustc_inherit_overflow_checks] and inlined from
         // another crate (mostly core::num generic/#[inline] fns),
         // while the current crate doesn't use overflow checks.
-        if !bx.cx().check_overflow() {
+        if !bx.cx().check_overflow(Some(input_ty)) {
             let val = self.codegen_scalar_binop(bx, op, lhs, rhs, input_ty);
             return OperandValue::Pair(val, bx.cx().const_bool(false));
         }

src/librustc_codegen_ssa/traits/misc.rs

@@ -12,7 +12,7 @@ pub trait MiscMethods<'tcx>: BackendTypes {
     fn vtables(
         &self,
     ) -> &RefCell<FxHashMap<(Ty<'tcx>, Option<ty::PolyExistentialTraitRef<'tcx>>), Self::Value>>;
-    fn check_overflow(&self) -> bool;
+    fn check_overflow(&self, ty: Option<Ty<'tcx>>) -> bool;
     fn instances(&self) -> &RefCell<FxHashMap<Instance<'tcx>, Self::Value>>;
     fn get_fn(&self, instance: Instance<'tcx>) -> Self::Value;
     fn get_param(&self, llfn: Self::Value, index: c_uint) -> Self::Value;

src/librustc_incremental/persist/load.rs

@@ -200,7 +200,7 @@ pub fn load_query_result_cache<'sess>(sess: &'sess Session) -> OnDiskCache<'sess
     }
 
     match load_data(sess.opts.debugging_opts.incremental_info, &query_cache_path(sess)) {
-        LoadResult::Ok{ data: (bytes, start_pos) } => OnDiskCache::new(sess, bytes, start_pos),
+        LoadResult::Ok{ data: (bytes, _) } => OnDiskCache::new(sess, bytes),
         _ => OnDiskCache::new_empty(sess.source_map())
     }
 }

src/librustc_metadata/decoder.rs

@@ -179,7 +179,7 @@ impl<'a, 'tcx: 'a> TyDecoder<'a, 'tcx> for DecodeContext<'a, 'tcx> {
 
     #[inline]
     fn peek_byte(&self) -> u8 {
-        self.opaque.data[self.opaque.position()]
+        self.opaque.data()[0]
     }
 
     #[inline]
@@ -212,7 +212,7 @@ impl<'a, 'tcx: 'a> TyDecoder<'a, 'tcx> for DecodeContext<'a, 'tcx> {
     fn with_position<F, R>(&mut self, pos: usize, f: F) -> R
         where F: FnOnce(&mut Self) -> R
     {
-        let new_opaque = opaque::Decoder::new(self.opaque.data, pos);
+        let new_opaque = opaque::Decoder::new(self.opaque.original_data, pos);
         let old_opaque = mem::replace(&mut self.opaque, new_opaque);
         let old_state = mem::replace(&mut self.lazy_state, LazyState::NoNode);
         let r = f(self);

src/librustc_mir/build/expr/as_rvalue.rs

@@ -84,7 +84,7 @@ impl<'a, 'gcx, 'tcx> Builder<'a, 'gcx, 'tcx> {
             ExprKind::Unary { op, arg } => {
                 let arg = unpack!(block = this.as_operand(block, scope, arg));
                 // Check for -MIN on signed integers
-                if this.hir.check_overflow() && op == UnOp::Neg && expr.ty.is_signed() {
+                if this.hir.check_overflow(expr.ty) && op == UnOp::Neg && expr.ty.is_signed() {
                     let bool_ty = this.hir.bool_ty();
 
                     let minval = this.minval_literal(expr_span, expr.ty);
@@ -410,7 +410,7 @@ impl<'a, 'gcx, 'tcx> Builder<'a, 'gcx, 'tcx> {
     ) -> BlockAnd<Rvalue<'tcx>> {
         let source_info = self.source_info(span);
         let bool_ty = self.hir.bool_ty();
-        if self.hir.check_overflow() && op.is_checkable() && ty.is_integral() {
+        if self.hir.check_overflow(ty) && op.is_checkable() && ty.is_integral() {
             let result_tup = self.hir.tcx().intern_tup(&[ty, bool_ty]);
             let result_value = self.temp(result_tup, span);
 

src/librustc_mir/hair/cx/mod.rs

@@ -11,7 +11,7 @@ use rustc::hir::Node;
 use rustc::middle::region;
 use rustc::infer::InferCtxt;
 use rustc::ty::subst::Subst;
-use rustc::ty::{self, Ty, TyCtxt};
+use rustc::ty::{self, Ty, TyCtxt, TyKind};
 use rustc::ty::subst::{Kind, Substs};
 use rustc::ty::layout::VariantIdx;
 use syntax::ast;
@@ -218,8 +218,8 @@ impl<'a, 'gcx, 'tcx> Cx<'a, 'gcx, 'tcx> {
         self.tables
     }
 
-    pub fn check_overflow(&self) -> bool {
-        self.check_overflow
+    pub fn check_overflow(&self, ty: Ty<'tcx>) -> bool {
+        self.check_overflow || ty.sty == TyKind::Uint(ast::UintTy::Usize)
     }
 
     pub fn type_is_copy_modulo_regions(&self, ty: Ty<'tcx>, span: Span) -> bool {

src/libserialize/leb128.rs

@@ -114,10 +114,10 @@ pub fn write_signed_leb128(out: &mut Vec<u8>, value: i128) {
 }
 
 #[inline]
-pub fn read_signed_leb128(data: &[u8], start_position: usize) -> (i128, usize) {
+pub fn read_signed_leb128(data: &[u8]) -> (i128, usize) {
     let mut result = 0;
     let mut shift = 0;
-    let mut position = start_position;
+    let mut position = 0;
     let mut byte;
 
     loop {
@@ -136,7 +136,7 @@ pub fn read_signed_leb128(data: &[u8], start_position: usize) -> (i128, usize) {
         result |= -(1 << shift);
     }
 
-    (result, position - start_position)
+    (result, position)
 }
 
 macro_rules! impl_test_unsigned_leb128 {
@@ -176,7 +176,7 @@ fn test_signed_leb128() {
     }
     let mut pos = 0;
     for &x in &values {
-        let (value, bytes_read) = read_signed_leb128(&mut stream, pos);
+        let (value, bytes_read) = read_signed_leb128(&mut stream[pos..]);
         pos += bytes_read;
         assert_eq!(x, value);
     }

src/libserialize/lib.rs

@@ -15,6 +15,7 @@ Core encoding and decoding interfaces.
 #![feature(specialization)]
 #![feature(never_type)]
 #![feature(nll)]
+#![feature(ptr_wrapping_offset_from)]
 #![cfg_attr(test, feature(test))]
 
 pub use self::serialize::{Decoder, Encoder, Decodable, Encodable};

src/libserialize/opaque.rs

@@ -157,59 +157,55 @@ impl Encoder {
 // -----------------------------------------------------------------------------
 
 pub struct Decoder<'a> {
-    pub data: &'a [u8],
-    position: usize,
+    pub original_data: &'a [u8],
+    data: &'a [u8],
 }
 
 impl<'a> Decoder<'a> {
     #[inline]
-    pub fn new(data: &'a [u8], position: usize) -> Decoder<'a> {
+    pub fn new(data: &'a [u8], pos: usize) -> Decoder<'a> {
         Decoder {
-            data,
-            position,
+            original_data: data,
+            data: &data[pos..],
         }
     }
 
     #[inline]
-    pub fn position(&self) -> usize {
-        self.position
+    pub fn data(&self) -> &[u8] {
+        self.data
     }
 
     #[inline]
-    pub fn set_position(&mut self, pos: usize) {
-        self.position = pos
+    pub fn position(&self) -> usize {
+        self.data.as_ptr().wrapping_offset_from(self.original_data.as_ptr()) as usize
     }
 
     #[inline]
     pub fn advance(&mut self, bytes: usize) {
-        self.position += bytes;
+        self.data = &self.data[bytes..];
     }
 
     #[inline]
     pub fn read_raw_bytes(&mut self, s: &mut [u8]) -> Result<(), String> {
-        let start = self.position;
-        let end = start + s.len();
-
-        s.copy_from_slice(&self.data[start..end]);
-
-        self.position = end;
+        s.copy_from_slice(&self.data[..s.len()]);
+        self.advance(s.len());
 
         Ok(())
     }
 }
 
 macro_rules! read_uleb128 {
     ($dec:expr, $t:ty, $fun:ident) => ({
-        let (value, bytes_read) = leb128::$fun(&$dec.data[$dec.position ..]);
-        $dec.position += bytes_read;
+        let (value, bytes_read) = leb128::$fun(&$dec.data);
+        $dec.advance(bytes_read);
         Ok(value)
     })
 }
 
 macro_rules! read_sleb128 {
     ($dec:expr, $t:ty) => ({
-        let (value, bytes_read) = read_signed_leb128($dec.data, $dec.position);
-        $dec.position += bytes_read;
+        let (value, bytes_read) = read_signed_leb128($dec.data);
+        $dec.advance(bytes_read);
         Ok(value as $t)
     })
 }
@@ -245,8 +241,8 @@ impl<'a> serialize::Decoder for Decoder<'a> {
 
     #[inline]
     fn read_u8(&mut self) -> Result<u8, Self::Error> {
-        let value = self.data[self.position];
-        self.position += 1;
+        let value = self.data[0];
+        self.advance(1);
         Ok(value)
     }
 
@@ -277,8 +273,8 @@ impl<'a> serialize::Decoder for Decoder<'a> {
 
     #[inline]
     fn read_i8(&mut self) -> Result<i8, Self::Error> {
-        let as_u8 = self.data[self.position];
-        self.position += 1;
+        let as_u8 = self.data[0];
+        self.advance(1);
         unsafe { Ok(::std::mem::transmute(as_u8)) }
     }
 
@@ -314,8 +310,8 @@ impl<'a> serialize::Decoder for Decoder<'a> {
     #[inline]
     fn read_str(&mut self) -> Result<Cow<'_, str>, Self::Error> {
         let len = self.read_usize()?;
-        let s = ::std::str::from_utf8(&self.data[self.position..self.position + len]).unwrap();
-        self.position += len;
+        let s = ::std::str::from_utf8(&self.data[..len]).unwrap();
+        self.advance(len);
         Ok(Cow::Borrowed(s))
     }