Array.prototype.concat should allocate output storage only once.

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Andreas Kling]

Array.prototype.concat should allocate output storage only once.

[Andreas Kling]

Created attachment 229273 [details]
Patch

[Darin Adler]

Comment on attachment 229273 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=229273&action=review

> Source/JavaScriptCore/runtime/ArrayPrototype.cpp:421
> +    unsigned finalArraySize = 0;

This is unsigned.

> Source/JavaScriptCore/runtime/ArrayPrototype.cpp:423
> +    for (size_t i = 0; i <= argCount; ++i) {

Yet this is size_t. What guarantees we don’t overflow the unsigned?

> Source/JavaScriptCore/runtime/ArrayPrototype.cpp:428
> +        curArg = exec->uncheckedArgument(i);

A little lame to do this one extra time at the end of the loop but then discard the result.

> Source/JavaScriptCore/runtime/ArrayPrototype.cpp:431
> +    JSArray* arr = constructEmptyArray(exec, nullptr, finalArraySize);

Go nuts and call this array instead of arr?

[Andreas Kling]

(In reply to comment #2)
> (From update of attachment 229273 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=229273&action=review
> 
> > Source/JavaScriptCore/runtime/ArrayPrototype.cpp:421
> > +    unsigned finalArraySize = 0;
> 
> This is unsigned.
> 
> > Source/JavaScriptCore/runtime/ArrayPrototype.cpp:423
> > +    for (size_t i = 0; i <= argCount; ++i) {
> 
> Yet this is size_t. What guarantees we don’t overflow the unsigned?

Let's Checked<> it.

[Andreas Kling]

Committed r167249: <http://trac.webkit.org/changeset/167249>

[Oliver Hunt]

Comment on attachment 229273 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=229273&action=review

>> Source/JavaScriptCore/runtime/ArrayPrototype.cpp:423
>> +    for (size_t i = 0; i <= argCount; ++i) {
> 
> Yet this is size_t. What guarantees we don’t overflow the unsigned?

Why <= argCount? That means we always walk off the end of the argument list.  What am i missing?

>> Source/JavaScriptCore/runtime/ArrayPrototype.cpp:428
>> +        curArg = exec->uncheckedArgument(i);
> 
> A little lame to do this one extra time at the end of the loop but then discard the result.

ah i see. do {} while loop instead maybe?

> Source/JavaScriptCore/runtime/ArrayPrototype.cpp:442
>              for (unsigned k = 0; k < length; ++k) {
> -                JSValue v = getProperty(exec, curObject, k);
> +                JSValue v = getProperty(exec, currentArray, k);

currentArray is an object, i would do a canGetFastIndex, and fastGet -- so two loops but the get should be much faster.

[WebKit Commit Bot]

Re-opened since this is blocked by bug 131621

[Andreas Kling]

Created attachment 229291 [details]
Patch

Derp, let's not read past the last argument, shall we. Wish we had a debug-ews to tell me I'm doing dumb stuff sometimes.

[Oliver Hunt]

Comment on attachment 229291 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=229291&action=review

> Source/JavaScriptCore/runtime/ArrayPrototype.cpp:444
> -                JSValue v = getProperty(exec, curObject, k);
> +                JSValue v = getProperty(exec, currentArray, k);

Follow on patch to optimise this get

[Andreas Kling]

Committed r167255: <http://trac.webkit.org/changeset/167255>