Bring in 5.1 - 5.0 abi compatibility

[matthewloring]

No description provided.

[ofrobots]

Good work! Looking at the diff from v8.h in 5.0, there is still this ABI difference remaining:

@@ -7149,7 +7323,7 @@ class Internals {
1 * kApiPointerSize + kApiIntSize;
static const int kStringResourceOffset = 3 * kApiPointerSize;
- static const int kOddballKindOffset = 4 * kApiPointerSize;+ static const int kOddballKindOffset = 5 * kApiPointerSize;
static const int kForeignAddressOffset = kApiPointerSize;

[matthewloring]

That should be fixed now.

[ofrobots]

Looking at some of the other include/ files, a few more differences. These should be easy to fix however:

--- a/deps/v8/include/v8-debug.h+++ b/deps/v8/include/v8-debug.h@@ -18,13 +18,11 @@ enum DebugEvent {
Exception = 2,
NewFunction = 3,
BeforeCompile = 4,
- AfterCompile = 5,+ AfterCompile = 5,
CompileError = 6,
- PromiseEvent = 7,- AsyncTaskEvent = 8,+ AsyncTaskEvent = 7,
};
--- a/deps/v8/include/v8-platform.h+++ b/deps/v8/include/v8-platform.h@@ -152,9 +152,9 @@ class Platform {
*/
virtual uint64_t AddTraceEvent(
char phase, const uint8_t* category_enabled_flag, const char* name,
- uint64_t id, uint64_t bind_id, int32_t num_args, const char** arg_names,- const uint8_t* arg_types, const uint64_t* arg_values,- unsigned int flags) {+ const char* scope, uint64_t id, uint64_t bind_id, int32_t num_args,+ const char** arg_names, const uint8_t* arg_types,+ const uint64_t* arg_values, unsigned int flags) {
return 0;
}

[matthewloring]

These should now be compatible.

[verwaest-zz]

Swapping this back around will be bad for performance, since we perform range checks on LAST_SPECIAL_RECEIVER_TYPE and LAST_CUSTOM_ELEMENTS_RECEIVER. If JS_OBJECT_TYPE falls above, all JS_OBJECT_TYPE objects (most common JavaScript object) will take the slow path for e.g., .hasOwnProperty.

[matthewloring]

I added custom checks for JS_OBJECT_TYPE at the comparison points to LAST_CUSTOM_ELEMENTS_RECEIVER. Using this contrived benchmark:

varo={};for(vari=0;i<10000000;i++){Object.assign(o,{i: i});for(varkino){if(!o.hasOwnProperty(k)){console.log('fail');}}}

I observed:

mattloring-macbookpro:node mattloring$ time node-6.1.0 ~/Desktop/slow-test.js
real 0m1.985s
user 0m1.961s
sys 0m0.019s
mattloring-macbookpro:node mattloring$ time out/Release/node-old ~/Desktop/slow-test.js
real 0m25.168s
user 0m24.408s
sys 0m0.095s
mattloring-macbookpro:node mattloring$ time out/Release/node-fast ~/Desktop/slow-test.js
real 0m2.104s
user 0m1.960s
sys 0m0.016s

@ofrobots@verwaest What do you think?

[ofrobots]

I think this is not right. JS_SPECIAL_API_OBJECT_TYPE used to be <= LAST_CUSTOM_ELEMENTS_RECEIVER before your change, but is > after your change. JS_OBJECT_TYPE is invariant here.

I think the expression you want is:

OR (
LESS_EQUAL (instance_type, LAST_CUSTOM_ELEMENTS_RECEIVER
EQUAL(instance_type, JS_SPECIAL_API_OBJECT_TYPE))

[matthewloring]

Fixed.

[ofrobots]

LGTM to me. Let's get @verwaest to review as well.

[targos]

There is now v8/v8@ea5e96f in the 5.1 branch. Do you have to make changes for that?

[verwaest-zz]

LGTM as well.

[matthewloring]

I've updated the patch to support v8/v8@ea5e96f.

[ofrobots]

LGTM. @jeisinger PTAL as well.

[jeisinger]

LGTM

[targos]

@matthewloring If everything is ready, could you squash the commits so I can pick them up in nodejs#7016 ?

[matthewloring]

@targos All squashed.

[targos]

The patch does not apply cleanly on my branch

[ofrobots]

Actually this PR needs more work. Intercepted enumerator tests are failing for me.

@matthewloring I think the problem is that JS_SPECIAL_API_OBJECT_TYPE has been reordered wrt. LAST_CUSTOM_ELEMENTS_RECEIVER. API objects (with interceptors) no longer look like they require custom elements iteration breaking enumerator tests.

@targos: this is not likely to be ready before next week.

[ofrobots]

Looking at this more closely, I think the removal of PromiseEvent was actually an API change that is missing from the official list (/cc @natorion). I think we will have bring this back as a revert of https://codereview.chromium.org/1833563002.

[ofrobots]

nit: align with the previous line.

[ofrobots]

LGTM from my side. Given that that the roots list was reordered; it would be good to get signoff from @verwaest or @jeisinger as well.

[jeisinger]

lgtm

[ofrobots]

Launched a CI: https://ci.nodejs.org/job/node-test-commit/3684/
@targos this is rebased on top of your branch.

[matthewloring]

Closing this as it has landed in core.