1149397 - JS::ubi::Node::edges should return a mozilla::UniquePtr

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Nick Fitzgerald [:fitzgen] [⏰PST; UTC-8]]

It is less than awesome in that the docs say you have to free it with js_free, but that is hard to remember and there are no guarantees. Instead, we should just hand out mozilla::UniquePtr instances with JS::FreePolicy so that this stuff happens automatically.

Comment 1

[Nick Fitzgerald [:fitzgen] [⏰PST; UTC-8]]

Attachment: bug-1149397-edges-return-UniquePtr.patch

Try push: https://treeherder.mozilla.org/#/jobs?repo=try&revision=9267d40202be

Comment 2

[Jim Blandy :jimb]

Comment on attachment 8586450 [details] [diff] [review]
bug-1149397-edges-return-UniquePtr.patch

Review of attachment 8586450 [details] [diff] [review]:
-----------------------------------------------------------------

Great stuff. I have a suggestion, though:

::: js/public/UbiNode.h
@@ +195,5 @@
>      // on |cx| and return nullptr.
>      //
>      // If wantNames is true, compute names for edges. Doing so can be expensive
>      // in time and memory.
> +    virtual UniquePtr<EdgeRange, JS::DeletePolicy<EdgeRange>> edges(JSContext* cx, bool wantNames)

Let's do this instead:

namespace mozilla {
template<>
class DefaultDelete<EdgeRange> : public JS::DeletePolicy<EdgeRange> { };
}

somewhere in this file to establish that as the default deleter for EdgeRange. Note that that default gets picked up here:

https://hg.mozilla.org/mozilla-central/file/c20f8549d631/mfbt/UniquePtr.h#l22

Then just using UniquePtr<EdgeRange> would do the right thing.

It would also improve legibility, without resorting to a typedef.

::: js/src/vm/UbiNode.cpp
@@ -8,5 @@
>  
>  #include "mozilla/Assertions.h"
>  #include "mozilla/Attributes.h"
>  #include "mozilla/Scoped.h"
> -#include "mozilla/UniquePtr.h"

Let's do

  using mozilla::UniquePtr;

within this file, too.

@@ +47,5 @@
>  // All operations on null ubi::Nodes crash.
>  const char16_t* Concrete<void>::typeName() const          { MOZ_CRASH("null ubi::Node"); }
> +mozilla::UniquePtr<EdgeRange, JS::DeletePolicy<EdgeRange>> Concrete<void>::edges(JSContext*,
> +                                                                                 bool) const {
> +    MOZ_CRASH("null ubi::Node");

If this is still too long even after we establish the default deleter, then let's move it out of the specially-formatted block, as we did for ::size, since we can't helpfully show off its symmetry this way anyway.

@@ +215,5 @@
>  
>      if (!r->init(cx, ptr, ::js::gc::MapTypeToTraceKind<Referent>::kind, wantNames))
>          return nullptr;
>  
> +    mozilla::UniquePtr<EdgeRange, JS::DeletePolicy<EdgeRange>> p(mozilla::Move(r));

*sigh* Yep.

Comment 3

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/ba8e68e05149

Comment 4

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/ba8e68e05149

Comment 5

[Pulsebot]

Backout:
https://hg.mozilla.org/mozilla-central/rev/0b202671c9e2

Comment 6

[Carsten Book [:Tomcat]]

sorry had to backout this changes in https://hg.mozilla.org/mozilla-central/rev/0b202671c9e2 since one of this changes seems to have caused a Linux PGO Bustage like:

https://treeherder.mozilla.org/logviewer.html#?job_id=1380632&repo=mozilla-central

Comment 7

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/fff6dec11ff9

Comment 8

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/fff6dec11ff9