add arrayBuffer/Blob support to httpDo. implement loadBytes

[Spongman]

re: #2693, closes #2674.

this PR adds arrayBuffer and Blob support to the httpDo method, and adds a convenience method loadBytes for loading a Uint8Array specifically.

adds some tests for loadBytes, including one for ensuring that the error callback is called when a 404 occurs.

[Spongman]

ugh. this test only fails on travis. i can't tell why...

[Spongman]

there seems to be an issue with the mocha test framework where an Error thrown in an inner Promise causes the test to fail rather than be caught in the outer Promise's catch clause. disabling this 404 test for now... also, this doesn't seem to happen on Windows, just on linux.

[limzykenneth]

Looks great, I'll check it out later this week and maybe have a crack at the mocha problem as well, weird that it differs between windows and linux though...

[Spongman]

yeah, i'm not really sure how to debug this, since it only happens within PhantomJS, not in the browser. I tried forcing the Promise polyfill in the browser, but it still didn't repro the problem.

[limzykenneth]

Ok, I've tested all that I can think of and here are a few points:

Not sure why I didn't caught this before (maybe because I switched to Firefox recently) but if an error callback isn't provided, simply throwing the returned error will in Firefox cause it to fail silently while in Chrome and Safari cause it to throw an "Unhandled Promise Rejection" error. For some context of why see this question https://stackoverflow.com/questions/30715367/why-can-i-not-throw-inside-a-promise-catch-handler#30741722 The suggested method in the question using setTimeout gave a very unhelpful message so I wouldn't suggest that. I think maybe substituting throw err with console.error(err) could work but if you have a better solution do go for it.

---

Next is about the mocha problem which I think I figured out. It seems that webkit haven't implement finally for Promise yet which is why it failed in PhantomJS and not most browsers. If the test for that can be rewritten it would be great. Normally I use sinon.js to test that but it may not be worth to bring the whole thing in just for this test. You can decide how to proceed.

---

The rest may not be best for here as they are more general feature request, I can move it to a separate issue once this is merge.

• It would be nice to be able to specify type for the httpDo overload that accept Request object otherwise anything we cannot auto detect its type will be parsed as text.

• We probably need a new example for httpDo documentation that demonstrate its use without using the Request object as that may be a bit too high level for most not familiar with these standard.

Once the first two points are checked off this can be merged.

[Spongman]

Fantastic! I'll switch to ’console.error’ and workaround the missing finally.

[limzykenneth]

Excellent, thanks!