Does not work with subarrays

[mateogianolio]

A bug with Node.js is causing the offsets of subarrays to not be registered when using the following pattern Below code does not work with subarrays because the ArrayBuffer offset is not taken into account:

v8::Local<v8::Float64Array> array = argument.As<v8::Float64Array>();
void *data = array->Buffer()->GetContents().Data();
double *x = static_cast<double*>(data);

This seems to be fixed by using the following hack:

v8::Local<v8::Float64Array> array = argument.As<v8::Float64Array>();
void *data = array->Buffer()->GetContents().Data();

size_t offset = array->ByteOffset();
size_t data_ptr = reinterpret_cast<size_t>(data);
data_ptr += offset;

double *x = reinterpret_cast<double*>(data_ptr);

However, it appears to be undefined behaviour in C++. Not sure if I should wait for Node.js to patch the bug or if I should incorporate above into the code.

Ping @metabench.

[metabench]

I'd hesitate to call the first example a 'pattern', no offence to its authors intended. It looks like it gets the contents of the underlying buffer, and may not be the right code to match the intention. It's a general area I'm unsure about though, and am asking others, including those who run node.js to consider, and to establish and document what the patterns are.

See nodejs/node#8269

I'm hopeful that those with a thorough understanding of V8 and C++ can agree on what the best practice for this is. I've not yet tried the provided hack. If no-one else has done this yet, I suppose that by default it is best practice as well as being a hack.

[mateogianolio]

Searched the node repo for GetContents() occurrences and found a few interesting results, e.g. this test binding for zlib:

auto buffer = view->Buffer();
auto contents = buffer->GetContents();
auto data = static_cast<unsigned char*>(contents.Data()) + byte_offset;

They cast the underlying buffer to unsigned char* and then add offset.

[mateogianolio]

I got this to work:

void *data = array->Buffer()->GetContents().Data();
size_t offset = array->ByteOffset();
unsigned char *data_offset = static_cast<unsigned char*>(data) + offset;
double *x = reinterpret_cast<double*>(data_offset);

[metabench]

The first line of that example does not have matching brackets. Having removed an extra bracket at the end, I also get a compilation error:

error C2036: 'void *' : unknown size

on the line that corresponds with:

unsigned char* data_offset = static_cast<unsigned char*>(data + offset);

[mateogianolio]

Wow that was sloppy pasting on my part, the + offset should be outside of the cast. Edited the code above.

[metabench]

I took the last two lines of your corrected code and condensed it into:

double *x = reinterpret_cast<double*>(static_cast<unsigned char*>(data) + offset);

I suggest that would be better as it does not leave the unsigned char *data_offset data structure in memory, what's your opinion on this, @mateogianolio?

[mateogianolio]

Looks good!

I'm going to update the code in this repo with the fix and submit a pull request.