Measure LocalLayoutContexts in LayoutTask and LayoutWorkers.

The FreeType instances in layout workers are reasonably large and worth
measuring. The one in LayoutTask is smaller but it's easy to measure at the
same time. Sample output:
```
|    8.33 MiB -- pages
|       8.33 MiB -- url(file:///home/njn/moz/servo/../servo-static-suite/wikipedia/Guardians%20of%20the%20Galaxy%20(film)%20-%20Wikipedia,%20the%20free%20encyclopedia.html)
|          1.32 MiB -- layout-worker-0-local-context
|          1.31 MiB -- layout-worker-1-local-context
|          1.24 MiB -- layout-worker-3-local-context
|          1.17 MiB -- layout-worker-4-local-context
|          1.08 MiB -- layout-worker-2-local-context
|          1.06 MiB -- layout-worker-5-local-context
|          0.78 MiB -- paint-task
|             0.78 MiB -- buffer-map
|          0.38 MiB -- layout-task
|             0.30 MiB -- display-list
|             0.07 MiB -- local-context
```
This required adding a mechanism to WorkQueue to measure worker TLSes.

components/layout/context.rs

@@ -29,6 +29,7 @@ use std::sync::mpsc::{channel, Sender};
 use style::selector_matching::Stylist;
 use url::Url;
 use util::geometry::Au;
+use util::mem::HeapSizeOf;
 use util::opts;
 
 struct LocalLayoutContext {
@@ -37,8 +38,21 @@ struct LocalLayoutContext {
     style_sharing_candidate_cache: RefCell<StyleSharingCandidateCache>,
 }
 
+impl HeapSizeOf for LocalLayoutContext {
+    // FIXME(njn): measure other fields eventually.
+    fn heap_size_of_children(&self) -> usize {
+        self.font_context.heap_size_of_children()
+    }
+}
+
 thread_local!(static LOCAL_CONTEXT_KEY: RefCell<Option<Rc<LocalLayoutContext>>> = RefCell::new(None));
 
+pub fn heap_size_of_local_context() -> usize {
+    LOCAL_CONTEXT_KEY.with(|r| {
+        r.borrow().clone().map_or(0, |context| context.heap_size_of_children())
+    })
+}
+
 fn create_or_get_local_context(shared_layout_context: &SharedLayoutContext)
                                -> Rc<LocalLayoutContext> {
     LOCAL_CONTEXT_KEY.with(|r| {

components/layout/layout_task.rs

@@ -9,7 +9,7 @@
 
 use animation;
 use construct::ConstructionResult;
-use context::{SharedLayoutContext, SharedLayoutContextWrapper};
+use context::{SharedLayoutContext, SharedLayoutContextWrapper, heap_size_of_local_context};
 use css::node_style::StyledNode;
 use data::{LayoutDataAccess, LayoutDataWrapper};
 use display_list_builder::ToGfxColor;
@@ -575,10 +575,28 @@ impl LayoutTask {
         let rw_data = self.lock_rw_data(possibly_locked_rw_data);
         let stacking_context = rw_data.stacking_context.as_ref();
         reports.push(Report {
-            path: path!["pages", format!("url({})", self.url), "display-list"],
+            path: path!["pages", format!("url({})", self.url), "layout-task", "display-list"],
             size: stacking_context.map_or(0, |sc| sc.heap_size_of_children()),
         });
 
+        // The LayoutTask has a context in TLS...
+        reports.push(Report {
+            path: path!["pages", format!("url({})", self.url), "layout-task", "local-context"],
+            size: heap_size_of_local_context(),
+        });
+
+        // ... as do each of the LayoutWorkers, if present.
+        if let Some(ref traversal) = rw_data.parallel_traversal {
+            let sizes = traversal.heap_size_of_tls(heap_size_of_local_context);
+            for (i, size) in sizes.iter().enumerate() {
+                reports.push(Report {
+                    path: path!["pages", format!("url({})", self.url),
+                                format!("layout-worker-{}-local-context", i)],
+                    size: *size
+                });
+            }
+        }
+
         reports_chan.send(reports);
     }
 

components/profile/mem.rs

@@ -224,7 +224,7 @@ impl ReportsTree {
         }
 
         let mebi = 1024f64 * 1024f64;
-        let count_str = if self.count > 1 { format!(" {}", self.count) } else { "".to_owned() };
+        let count_str = if self.count > 1 { format!(" [{}]", self.count) } else { "".to_owned() };
         println!("|{}{:8.2} MiB -- {}{}",
                  indent_str, (self.size as f64) / mebi, self.path_seg, count_str);
 

components/util/mem.rs

@@ -5,6 +5,7 @@
 //! Data structure measurement.
 
 use libc::{c_void, size_t};
+use std::cell::RefCell;
 use std::collections::LinkedList;
 use std::mem::transmute;
 use std::sync::Arc;
@@ -92,6 +93,12 @@ impl<T: HeapSizeOf> HeapSizeOf for Arc<T> {
     }
 }
 
+impl<T: HeapSizeOf> HeapSizeOf for RefCell<T> {
+    fn heap_size_of_children(&self) -> usize {
+        self.borrow().heap_size_of_children()
+    }
+}
+
 impl<T: HeapSizeOf> HeapSizeOf for Vec<T> {
     fn heap_size_of_children(&self) -> usize {
         heap_size_of(self.as_ptr() as *const c_void) +

components/util/workqueue.rs

@@ -36,6 +36,8 @@ enum WorkerMsg<QueueData: 'static, WorkData: 'static> {
     Start(Worker<WorkUnit<QueueData, WorkData>>, *mut AtomicUsize, *const QueueData),
     /// Tells the worker to stop. It can be restarted again with a `WorkerMsg::Start`.
     Stop,
+    /// Tells the worker to measure the heap size of its TLS using the supplied function.
+    HeapSizeOfTLS(fn() -> usize),
     /// Tells the worker thread to terminate.
     Exit,
 }
@@ -45,6 +47,7 @@ unsafe impl<QueueData: 'static, WorkData: 'static> Send for WorkerMsg<QueueData,
 /// Messages to the supervisor.
 enum SupervisorMsg<QueueData: 'static, WorkData: 'static> {
     Finished,
+    HeapSizeOfTLS(usize),
     ReturnDeque(usize, Worker<WorkUnit<QueueData, WorkData>>),
 }
 
@@ -90,6 +93,10 @@ impl<QueueData: Send, WorkData: Send> WorkerThread<QueueData, WorkData> {
                 WorkerMsg::Start(deque, ref_count, queue_data) => (deque, ref_count, queue_data),
                 WorkerMsg::Stop => panic!("unexpected stop message"),
                 WorkerMsg::Exit => return,
+                WorkerMsg::HeapSizeOfTLS(f) => {
+                    self.chan.send(SupervisorMsg::HeapSizeOfTLS(f())).unwrap();
+                    continue;
+                }
             };
 
             let mut back_off_sleep = 0 as u32;
@@ -315,11 +322,32 @@ impl<QueueData: Send, WorkData: Send> WorkQueue<QueueData, WorkData> {
         for _ in 0..self.workers.len() {
             match self.port.recv().unwrap() {
                 SupervisorMsg::ReturnDeque(index, deque) => self.workers[index].deque = Some(deque),
+                SupervisorMsg::HeapSizeOfTLS(_) => panic!("unexpected HeapSizeOfTLS message"),
                 SupervisorMsg::Finished => panic!("unexpected finished message!"),
             }
         }
     }
 
+    /// Synchronously measure memory usage of any thread-local storage.
+    pub fn heap_size_of_tls(&self, f: fn() -> usize) -> Vec<usize> {
+        // Tell the workers to measure themselves.
+        for worker in self.workers.iter() {
+            worker.chan.send(WorkerMsg::HeapSizeOfTLS(f)).unwrap()
+        }
+
+        // Wait for the workers to finish measuring themselves.
+        let mut sizes = vec![];
+        for _ in 0..self.workers.len() {
+            match self.port.recv().unwrap() {
+                SupervisorMsg::HeapSizeOfTLS(size) => {
+                    sizes.push(size);
+                }
+                _ => panic!("unexpected message!"),
+            }
+        }
+        sizes
+    }
+
     pub fn shutdown(&mut self) {
         for worker in self.workers.iter() {
             worker.chan.send(WorkerMsg::Exit).unwrap()