Make tests pass on nightly again

[ordian]

I guess the problem is that rust nightly no longer uses jemalloc and we're using heapsize crate, which assumes it does (cc rust-lang/rust/issues/56069). A temporary solution would be to use jemalloc as a global allocator (https://github.com/alexcrichton/jemallocator ?).

[ordian]

This is worth doing soon, since this change will make its way into the rust stable eventually.
I remember @tomaka and @cheme had some thoughts on this.

[cheme]

For short term I can put jemalloc as default allocator with crate jemallocator, I currently need to fork heapsize with very few changes:

diff --git a/Cargo.toml b/Cargo.toml
index ec2747a..b3c2c0e 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -7,6 +7,10 @@ license = "MIT/Apache-2.0"
 repository = "https://github.com/servo/heapsize"
 build = "build.rs"
 
+
+[dependencies]
+jemallocator = "0.1.9"
+
 [target.'cfg(windows)'.dependencies]
 winapi = { version = "0.3.4", features = ["std", "heapapi"] }
 
diff --git a/src/lib.rs b/src/lib.rs
index 7eecbce..85e85e0 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -3,6 +3,13 @@
 #[cfg(target_os = "windows")]
 extern crate winapi;
 
+#[cfg(not(target_os = "windows"))]
+extern crate jemallocator;
+
+#[cfg(not(target_os = "windows"))]
+#[global_allocator]
+static ALLOC: jemallocator::Jemalloc = jemallocator::Jemalloc;
+
 #[cfg(target_os = "windows")]
 use winapi::um::heapapi::{GetProcessHeap, HeapSize, HeapValidate};
 use std::borrow::Cow;
@@ -36,15 +43,7 @@ pub unsafe fn heap_size_of<T>(ptr: *const T) -> usize {
 
 #[cfg(not(target_os = "windows"))]
 unsafe fn heap_size_of_impl(ptr: *const c_void) -> usize {
-    // The C prototype is `je_malloc_usable_size(JEMALLOC_USABLE_SIZE_CONST void *ptr)`. On some
-    // platforms `JEMALLOC_USABLE_SIZE_CONST` is `const` and on some it is empty. But in practice
-    // this function doesn't modify the contents of the block that `ptr` points to, so we use
-    // `*const c_void` here.
-    extern "C" {
-               #[cfg_attr(any(prefixed_jemalloc, target_os = "macos", target_os = "ios", target_os = "android"), link_name = "je_malloc_usable_size")]
-        fn malloc_usable_size(ptr: *const c_void) -> usize;
-    }
-    malloc_usable_size(ptr)
+    jemallocator::usable_size(ptr)
 }

I do not believe it will be a good long term solution (I do not really like the idea to get specific allocator things).

Maybe running more approximate allocation evalution (without querying the allocator).
Edit: Making functionalities requiring to heapsize optional (and bind heapsize use to jemalloc global alloc) could be another option. But I need to check what it implies (if it is even possible).

[tomaka]

The good long term solution is to kill the heapsize crate.

[5chdn]

This is worth doing soon, since this change will make its way into the rust stable eventually.

This is our window to address it:

• 1.32 beta: Dec 6th
• 1.32 stable: Jan 17th

Ideally, we have a fix before Christmas

[5chdn]

not related but relevant? #9167 #10013

[ordian]

not related but relevant? #9167 #10013

It's relevant in a sense that if/when we'll get rid of heapsize crate, it will resolve those issues as well.

[arkpar]

We certainly want to continue using jemalloc on supported platforms. It is much faster than the system allocator.

[cheme]

@arkpar that should not be a problem (I intend to allow setting global allocator on a PR I am currently working on).

Regarding the 10024 PR I believe the test limit does not matter that much and this change should be fine. (being able to switch to the default allocator might prove being useful in the future).

[ordian]

closed by #10432 (heapsize part)