Rollup of 3 pull requests

src/doc/rustc/src/codegen-options/index.md

@@ -257,9 +257,8 @@ them in parallel. Increasing parallelism may speed up compile times, but may
 also produce slower code. Setting this to 1 may improve the performance of
 generated code, but may be slower to compile.
 
-The default, if not specified, is 16. This flag is ignored if
-[incremental](#incremental) is enabled, in which case an internal heuristic is
-used to split the crate.
+The default, if not specified, is 16 for non-incremental builds. For
+incremental builds the default is 256 which allows caching to be more granular.
 
 ## remark
 

src/libcore/convert/mod.rs

@@ -41,6 +41,7 @@
 #![stable(feature = "rust1", since = "1.0.0")]
 
 use crate::fmt;
+use crate::hash::{Hash, Hasher};
 
 mod num;
 
@@ -746,3 +747,10 @@ impl From<!> for Infallible {
         x
     }
 }
+
+#[stable(feature = "convert_infallible_hash", since = "1.44.0")]
+impl Hash for Infallible {
+    fn hash<H: Hasher>(&self, _: &mut H) {
+        match *self {}
+    }
+}

src/librustc_mir/monomorphize/partitioning.rs

@@ -107,19 +107,11 @@ use rustc_middle::mir::mono::{InstantiationMode, MonoItem};
 use rustc_middle::ty::print::characteristic_def_id_of_type;
 use rustc_middle::ty::query::Providers;
 use rustc_middle::ty::{self, DefIdTree, InstanceDef, TyCtxt};
-use rustc_span::symbol::Symbol;
+use rustc_span::symbol::{Symbol, SymbolStr};
 
 use crate::monomorphize::collector::InliningMap;
 use crate::monomorphize::collector::{self, MonoItemCollectionMode};
 
-pub enum PartitioningStrategy {
-    /// Generates one codegen unit per source-level module.
-    PerModule,
-
-    /// Partition the whole crate into a fixed number of codegen units.
-    FixedUnitCount(usize),
-}
-
 // Anything we can't find a proper codegen unit for goes into this.
 fn fallback_cgu_name(name_builder: &mut CodegenUnitNameBuilder<'_>) -> Symbol {
     name_builder.build_cgu_name(LOCAL_CRATE, &["fallback"], Some("cgu"))
@@ -128,7 +120,7 @@ fn fallback_cgu_name(name_builder: &mut CodegenUnitNameBuilder<'_>) -> Symbol {
 pub fn partition<'tcx, I>(
     tcx: TyCtxt<'tcx>,
     mono_items: I,
-    strategy: PartitioningStrategy,
+    max_cgu_count: usize,
     inlining_map: &InliningMap<'tcx>,
 ) -> Vec<CodegenUnit<'tcx>>
 where
@@ -148,11 +140,10 @@ where
 
     debug_dump(tcx, "INITIAL PARTITIONING:", initial_partitioning.codegen_units.iter());
 
-    // If the partitioning should produce a fixed count of codegen units, merge
-    // until that count is reached.
-    if let PartitioningStrategy::FixedUnitCount(count) = strategy {
+    // Merge until we have at most `max_cgu_count` codegen units.
+    {
         let _prof_timer = tcx.prof.generic_activity("cgu_partitioning_merge_cgus");
-        merge_codegen_units(tcx, &mut initial_partitioning, count);
+        merge_codegen_units(tcx, &mut initial_partitioning, max_cgu_count);
         debug_dump(tcx, "POST MERGING:", initial_partitioning.codegen_units.iter());
     }
 
@@ -480,27 +471,78 @@ fn merge_codegen_units<'tcx>(
     // the stable sort below will keep everything nice and deterministic.
     codegen_units.sort_by_cached_key(|cgu| cgu.name().as_str());
 
+    // This map keeps track of what got merged into what.
+    let mut cgu_contents: FxHashMap<Symbol, Vec<SymbolStr>> =
+        codegen_units.iter().map(|cgu| (cgu.name(), vec![cgu.name().as_str()])).collect();
+
     // Merge the two smallest codegen units until the target size is reached.
     while codegen_units.len() > target_cgu_count {
         // Sort small cgus to the back
         codegen_units.sort_by_cached_key(|cgu| cmp::Reverse(cgu.size_estimate()));
         let mut smallest = codegen_units.pop().unwrap();
         let second_smallest = codegen_units.last_mut().unwrap();
 
+        // Move the mono-items from `smallest` to `second_smallest`
         second_smallest.modify_size_estimate(smallest.size_estimate());
         for (k, v) in smallest.items_mut().drain() {
             second_smallest.items_mut().insert(k, v);
         }
+
+        // Record that `second_smallest` now contains all the stuff that was in
+        // `smallest` before.
+        let mut consumed_cgu_names = cgu_contents.remove(&smallest.name()).unwrap();
+        cgu_contents.get_mut(&second_smallest.name()).unwrap().extend(consumed_cgu_names.drain(..));
+
         debug!(
-            "CodegenUnit {} merged in to CodegenUnit {}",
+            "CodegenUnit {} merged into CodegenUnit {}",
             smallest.name(),
             second_smallest.name()
         );
     }
 
     let cgu_name_builder = &mut CodegenUnitNameBuilder::new(tcx);
-    for (index, cgu) in codegen_units.iter_mut().enumerate() {
-        cgu.set_name(numbered_codegen_unit_name(cgu_name_builder, index));
+
+    if tcx.sess.opts.incremental.is_some() {
+        // If we are doing incremental compilation, we want CGU names to
+        // reflect the path of the source level module they correspond to.
+        // For CGUs that contain the code of multiple modules because of the
+        // merging done above, we use a concatenation of the names of
+        // all contained CGUs.
+        let new_cgu_names: FxHashMap<Symbol, String> = cgu_contents
+            .into_iter()
+            // This `filter` makes sure we only update the name of CGUs that
+            // were actually modified by merging.
+            .filter(|(_, cgu_contents)| cgu_contents.len() > 1)
+            .map(|(current_cgu_name, cgu_contents)| {
+                let mut cgu_contents: Vec<&str> = cgu_contents.iter().map(|s| &s[..]).collect();
+
+                // Sort the names, so things are deterministic and easy to
+                // predict.
+                cgu_contents.sort();
+
+                (current_cgu_name, cgu_contents.join("--"))
+            })
+            .collect();
+
+        for cgu in codegen_units.iter_mut() {
+            if let Some(new_cgu_name) = new_cgu_names.get(&cgu.name()) {
+                if tcx.sess.opts.debugging_opts.human_readable_cgu_names {
+                    cgu.set_name(Symbol::intern(&new_cgu_name));
+                } else {
+                    // If we don't require CGU names to be human-readable, we
+                    // use a fixed length hash of the composite CGU name
+                    // instead.
+                    let new_cgu_name = CodegenUnit::mangle_name(&new_cgu_name);
+                    cgu.set_name(Symbol::intern(&new_cgu_name));
+                }
+            }
+        }
+    } else {
+        // If we are compiling non-incrementally we just generate simple CGU
+        // names containing an index.
+        for (index, cgu) in codegen_units.iter_mut().enumerate() {
+            cgu.set_name(numbered_codegen_unit_name(cgu_name_builder, index));
+        }
     }
 }
 
@@ -879,13 +921,7 @@ fn collect_and_partition_mono_items(
     let (codegen_units, _) = tcx.sess.time("partition_and_assert_distinct_symbols", || {
         sync::join(
             || {
-                let strategy = if tcx.sess.opts.incremental.is_some() {
-                    PartitioningStrategy::PerModule
-                } else {
-                    PartitioningStrategy::FixedUnitCount(tcx.sess.codegen_units())
-                };
-
-                partition(tcx, items.iter().cloned(), strategy, &inlining_map)
+                partition(tcx, items.iter().cloned(), tcx.sess.codegen_units(), &inlining_map)
                     .into_iter()
                     .map(Arc::new)
                     .collect::<Vec<_>>()

src/librustc_session/session.rs

@@ -758,6 +758,13 @@ impl Session {
             return n as usize;
         }
 
+        // If incremental compilation is turned on, we default to a high number
+        // codegen units in order to reduce the "collateral damage" small
+        // changes cause.
+        if self.opts.incremental.is_some() {
+            return 256;
+        }
+
         // Why is 16 codegen units the default all the time?
         //
         // The main reason for enabling multiple codegen units by default is to

src/libstd/sys/cloudabi/thread.rs

@@ -4,8 +4,8 @@ use crate::io;
 use crate::mem;
 use crate::ptr;
 use crate::sys::cloudabi::abi;
+use crate::sys::stack_overflow;
 use crate::sys::time::checked_dur2intervals;
-use crate::sys_common::thread::*;
 use crate::time::Duration;
 
 pub const DEFAULT_MIN_STACK_SIZE: usize = 2 * 1024 * 1024;
@@ -22,27 +22,36 @@ unsafe impl Sync for Thread {}
 impl Thread {
     // unsafe: see thread::Builder::spawn_unchecked for safety requirements
     pub unsafe fn new(stack: usize, p: Box<dyn FnOnce()>) -> io::Result<Thread> {
-        let p = box p;
+        let p = Box::into_raw(box p);
         let mut native: libc::pthread_t = mem::zeroed();
         let mut attr: libc::pthread_attr_t = mem::zeroed();
         assert_eq!(libc::pthread_attr_init(&mut attr), 0);
 
         let stack_size = cmp::max(stack, min_stack_size(&attr));
         assert_eq!(libc::pthread_attr_setstacksize(&mut attr, stack_size), 0);
 
-        let ret = libc::pthread_create(&mut native, &attr, thread_start, &*p as *const _ as *mut _);
+        let ret = libc::pthread_create(&mut native, &attr, thread_start, p as *mut _);
+        // Note: if the thread creation fails and this assert fails, then p will
+        // be leaked. However, an alternative design could cause double-free
+        // which is clearly worse.
         assert_eq!(libc::pthread_attr_destroy(&mut attr), 0);
 
         return if ret != 0 {
+            // The thread failed to start and as a result p was not consumed. Therefore, it is
+            // safe to reconstruct the box so that it gets deallocated.
+            drop(Box::from_raw(p));
             Err(io::Error::from_raw_os_error(ret))
         } else {
-            mem::forget(p); // ownership passed to pthread_create
             Ok(Thread { id: native })
         };
 
         extern "C" fn thread_start(main: *mut libc::c_void) -> *mut libc::c_void {
             unsafe {
-                start_thread(main as *mut u8);
+                // Next, set up our stack overflow handler which may get triggered if we run
+                // out of stack.
+                let _handler = stack_overflow::Handler::new();
+                // Finally, let's run some code.
+                Box::from_raw(main as *mut Box<dyn FnOnce()>)();
             }
             ptr::null_mut()
         }

src/libstd/sys/hermit/thread.rs

@@ -5,11 +5,10 @@ use crate::fmt;
 use crate::io;
 use crate::mem;
 use crate::sys::hermit::abi;
+use crate::sys::stack_overflow;
 use crate::time::Duration;
 use core::u32;
 
-use crate::sys_common::thread::*;
-
 pub type Tid = abi::Tid;
 
 /// Priority of a task
@@ -49,26 +48,32 @@ impl Thread {
         p: Box<dyn FnOnce()>,
         core_id: isize,
     ) -> io::Result<Thread> {
-        let p = box p;
+        let p = Box::into_raw(box p);
         let mut tid: Tid = u32::MAX;
         let ret = abi::spawn(
             &mut tid as *mut Tid,
             thread_start,
-            &*p as *const _ as *const u8 as usize,
+            p as *mut u8 as usize,
             Priority::into(NORMAL_PRIO),
             core_id,
         );
 
-        return if ret == 0 {
-            mem::forget(p); // ownership passed to pthread_create
-            Ok(Thread { tid: tid })
-        } else {
+        return if ret != 0 {
+            // The thread failed to start and as a result p was not consumed. Therefore, it is
+            // safe to reconstruct the box so that it gets deallocated.
+            drop(Box::from_raw(p));
             Err(io::Error::new(io::ErrorKind::Other, "Unable to create thread!"))
+        } else {
+            Ok(Thread { tid: tid })
         };
 
         extern "C" fn thread_start(main: usize) {
             unsafe {
-                start_thread(main as *mut u8);
+                // Next, set up our stack overflow handler which may get triggered if we run
+                // out of stack.
+                let _handler = stack_overflow::Handler::new();
+                // Finally, let's run some code.
+                Box::from_raw(main as *mut Box<dyn FnOnce()>)();
             }
         }
     }

src/libstd/sys/unix/thread.rs

@@ -3,11 +3,9 @@ use crate::ffi::CStr;
 use crate::io;
 use crate::mem;
 use crate::ptr;
-use crate::sys::os;
+use crate::sys::{os, stack_overflow};
 use crate::time::Duration;
 
-use crate::sys_common::thread::*;
-
 #[cfg(not(target_os = "l4re"))]
 pub const DEFAULT_MIN_STACK_SIZE: usize = 2 * 1024 * 1024;
 #[cfg(target_os = "l4re")]
@@ -43,7 +41,7 @@ unsafe fn pthread_attr_setstacksize(
 impl Thread {
     // unsafe: see thread::Builder::spawn_unchecked for safety requirements
     pub unsafe fn new(stack: usize, p: Box<dyn FnOnce()>) -> io::Result<Thread> {
-        let p = box p;
+        let p = Box::into_raw(box p);
         let mut native: libc::pthread_t = mem::zeroed();
         let mut attr: libc::pthread_attr_t = mem::zeroed();
         assert_eq!(libc::pthread_attr_init(&mut attr), 0);
@@ -65,19 +63,28 @@ impl Thread {
             }
         };
 
-        let ret = libc::pthread_create(&mut native, &attr, thread_start, &*p as *const _ as *mut _);
+        let ret = libc::pthread_create(&mut native, &attr, thread_start, p as *mut _);
+        // Note: if the thread creation fails and this assert fails, then p will
+        // be leaked. However, an alternative design could cause double-free
+        // which is clearly worse.
         assert_eq!(libc::pthread_attr_destroy(&mut attr), 0);
 
         return if ret != 0 {
+            // The thread failed to start and as a result p was not consumed. Therefore, it is
+            // safe to reconstruct the box so that it gets deallocated.
+            drop(Box::from_raw(p));
             Err(io::Error::from_raw_os_error(ret))
         } else {
-            mem::forget(p); // ownership passed to pthread_create
             Ok(Thread { id: native })
         };
 
         extern "C" fn thread_start(main: *mut libc::c_void) -> *mut libc::c_void {
             unsafe {
-                start_thread(main as *mut u8);
+                // Next, set up our stack overflow handler which may get triggered if we run
+                // out of stack.
+                let _handler = stack_overflow::Handler::new();
+                // Finally, let's run some code.
+                Box::from_raw(main as *mut Box<dyn FnOnce()>)();
             }
             ptr::null_mut()
         }

src/libstd/sys/vxworks/thread.rs

@@ -3,11 +3,9 @@ use crate::ffi::CStr;
 use crate::io;
 use crate::mem;
 use crate::ptr;
-use crate::sys::os;
+use crate::sys::{os, stack_overflow};
 use crate::time::Duration;
 
-use crate::sys_common::thread::*;
-
 pub const DEFAULT_MIN_STACK_SIZE: usize = 0x40000; // 256K
 
 pub struct Thread {
@@ -31,7 +29,7 @@ unsafe fn pthread_attr_setstacksize(
 impl Thread {
     // unsafe: see thread::Builder::spawn_unchecked for safety requirements
     pub unsafe fn new(stack: usize, p: Box<dyn FnOnce()>) -> io::Result<Thread> {
-        let p = box p;
+        let p = Box::into_raw(box p);
         let mut native: libc::pthread_t = mem::zeroed();
         let mut attr: libc::pthread_attr_t = mem::zeroed();
         assert_eq!(libc::pthread_attr_init(&mut attr), 0);
@@ -53,19 +51,28 @@ impl Thread {
             }
         };
 
-        let ret = libc::pthread_create(&mut native, &attr, thread_start, &*p as *const _ as *mut _);
+        let ret = libc::pthread_create(&mut native, &attr, thread_start, p as *mut _);
+        // Note: if the thread creation fails and this assert fails, then p will
+        // be leaked. However, an alternative design could cause double-free
+        // which is clearly worse.
         assert_eq!(libc::pthread_attr_destroy(&mut attr), 0);
 
         return if ret != 0 {
+            // The thread failed to start and as a result p was not consumed. Therefore, it is
+            // safe to reconstruct the box so that it gets deallocated.
+            drop(Box::from_raw(p));
             Err(io::Error::from_raw_os_error(ret))
         } else {
-            mem::forget(p); // ownership passed to pthread_create
             Ok(Thread { id: native })
         };
 
         extern "C" fn thread_start(main: *mut libc::c_void) -> *mut libc::c_void {
             unsafe {
-                start_thread(main as *mut u8);
+                // Next, set up our stack overflow handler which may get triggered if we run
+                // out of stack.
+                let _handler = stack_overflow::Handler::new();
+                // Finally, let's run some code.
+                Box::from_raw(main as *mut Box<dyn FnOnce()>)();
             }
             ptr::null_mut()
         }

src/libstd/sys/windows/thread.rs

@@ -1,10 +1,9 @@
 use crate::ffi::CStr;
 use crate::io;
-use crate::mem;
 use crate::ptr;
 use crate::sys::c;
 use crate::sys::handle::Handle;
-use crate::sys_common::thread::*;
+use crate::sys::stack_overflow;
 use crate::time::Duration;
 
 use libc::c_void;
@@ -20,7 +19,7 @@ pub struct Thread {
 impl Thread {
     // unsafe: see thread::Builder::spawn_unchecked for safety requirements
     pub unsafe fn new(stack: usize, p: Box<dyn FnOnce()>) -> io::Result<Thread> {
-        let p = box p;
+        let p = Box::into_raw(box p);
 
         // FIXME On UNIX, we guard against stack sizes that are too small but
         // that's because pthreads enforces that stacks are at least
@@ -34,21 +33,27 @@ impl Thread {
             ptr::null_mut(),
             stack_size,
             thread_start,
-            &*p as *const _ as *mut _,
+            p as *mut _,
             c::STACK_SIZE_PARAM_IS_A_RESERVATION,
             ptr::null_mut(),
         );
 
         return if ret as usize == 0 {
+            // The thread failed to start and as a result p was not consumed. Therefore, it is
+            // safe to reconstruct the box so that it gets deallocated.
+            drop(Box::from_raw(p));
             Err(io::Error::last_os_error())
         } else {
-            mem::forget(p); // ownership passed to CreateThread
             Ok(Thread { handle: Handle::new(ret) })
         };
 
         extern "system" fn thread_start(main: *mut c_void) -> c::DWORD {
             unsafe {
-                start_thread(main as *mut u8);
+                // Next, set up our stack overflow handler which may get triggered if we run
+                // out of stack.
+                let _handler = stack_overflow::Handler::new();
+                // Finally, let's run some code.
+                Box::from_raw(main as *mut Box<dyn FnOnce()>)();
             }
             0
         }

src/libstd/sys_common/thread.rs

@@ -1,18 +1,7 @@
 use crate::env;
 use crate::sync::atomic::{self, Ordering};
-use crate::sys::stack_overflow;
 use crate::sys::thread as imp;
 
-#[allow(dead_code)]
-pub unsafe fn start_thread(main: *mut u8) {
-    // Next, set up our stack overflow handler which may get triggered if we run
-    // out of stack.
-    let _handler = stack_overflow::Handler::new();
-
-    // Finally, let's run some code.
-    Box::from_raw(main as *mut Box<dyn FnOnce()>)()
-}
-
 pub fn min_stack() -> usize {
     static MIN: atomic::AtomicUsize = atomic::AtomicUsize::new(0);
     match MIN.load(Ordering::SeqCst) {

src/test/codegen-units/partitioning/incremental-merging.rs

@@ -0,0 +1,42 @@
+// ignore-tidy-linelength
+// We specify -C incremental here because we want to test the partitioning for
+// incremental compilation
+// compile-flags:-Zprint-mono-items=lazy -Cincremental=tmp/partitioning-tests/incremental-merging
+// compile-flags:-Ccodegen-units=3
+
+#![crate_type = "rlib"]
+
+// This test makes sure that merging of CGUs works together with incremental
+// compilation but at the same time does not modify names of CGUs that were not
+// affected by merging.
+//
+// We expect CGUs `aaa` and `bbb` to be merged (because they are the smallest),
+// while `ccc` and `ddd` are supposed to stay untouched.
+
+pub mod aaa {
+    //~ MONO_ITEM fn incremental_merging::aaa[0]::foo[0] @@ incremental_merging-aaa--incremental_merging-bbb[External]
+    pub fn foo(a: u64) -> u64 {
+        a + 1
+    }
+}
+
+pub mod bbb {
+    //~ MONO_ITEM fn incremental_merging::bbb[0]::foo[0] @@ incremental_merging-aaa--incremental_merging-bbb[External]
+    pub fn foo(a: u64, b: u64) -> u64 {
+        a + b + 1
+    }
+}
+
+pub mod ccc {
+    //~ MONO_ITEM fn incremental_merging::ccc[0]::foo[0] @@ incremental_merging-ccc[External]
+    pub fn foo(a: u64, b: u64, c: u64) -> u64 {
+        a + b + c + 1
+    }
+}
+
+pub mod ddd {
+    //~ MONO_ITEM fn incremental_merging::ddd[0]::foo[0] @@ incremental_merging-ddd[External]
+    pub fn foo(a: u64, b: u64, c: u64, d: u64) -> u64 {
+        a + b + c + d + 1
+    }
+}

src/tools/compiletest/src/runtest.rs

@@ -2513,7 +2513,7 @@ impl<'test> TestCx<'test> {
                     .filter(|s| !s.is_empty())
                     .map(|s| {
                         if cgu_has_crate_disambiguator {
-                            remove_crate_disambiguator_from_cgu(s)
+                            remove_crate_disambiguators_from_set_of_cgu_names(s)
                         } else {
                             s.to_string()
                         }
@@ -2563,6 +2563,16 @@ impl<'test> TestCx<'test> {
 
             new_name
         }
+
+        // The name of merged CGUs is constructed as the names of the original
+        // CGUs joined with "--". This function splits such composite CGU names
+        // and handles each component individually.
+        fn remove_crate_disambiguators_from_set_of_cgu_names(cgus: &str) -> String {
+            cgus.split("--")
+                .map(|cgu| remove_crate_disambiguator_from_cgu(cgu))
+                .collect::<Vec<_>>()
+                .join("--")
+        }
     }
 
     fn init_incremental_test(&self) {