incorrect URL parsing

[trevnorris]

A regression was introduced between v0.10 and v0.12 for the URL parsing of an http.get() request. Basically, multi-byte characters are decoded as 'binary' instead of either

1. Being decoded as UTF-8
2. Properly decoded into their '%' counterparts.

Test and additional information is located at nodejs/node-v0.x-archive#25634 (comment)

[rvagg]

@trevnorris so to clarify, you're saying that io.js is impacted by this also (since we're not strictly an 0.12 fork that's not obvious)?

[trevnorris]

Yes. The test in the linked issue has the same result in io.js. A possible solution would be to simply parse the string with encodeURI() before turning it into a buffer using binary encoding.

[vkurchatkin]

This could be related: #1693

[bnoordhuis]

It's probably the result of this change. Before f674b09, headers and the status line were parsed as UTF-8, now they're parsed as ISO-8859-1.

[trevnorris]

@bnoordhuis Parsing the string this way follows more closely to the spec. Even though browsers may show the unicode characters in the URL bar, checking the network request shows it also decodes them before firing the request. io.js' http module would also barf on this request since we decode incoming headers using ISO-8859-1.

@vkurchatkin It does look like the same issue. IMO the options are to let the user know they need to encode their header strings before sending them, or we should consider doing that automatically before turning them into a buffer.

[Fishrock123]

Duplicate of #1693. See also: #2629

[vkurchatkin]

I wouldn't say it's duplicate. There are two problems with UTF8: parsing and writing, and they seem unrelated.