Skip to content

strings_and_bytes.rst

Latest commit

 

History

History
62 lines (45 loc) · 2.31 KB

strings_and_bytes.rst

File metadata and controls

62 lines (45 loc) · 2.31 KB
Raw

Strings vs. Bytes in Nevow on python3

Twisted.web decided to have represent several items in their Request class – which is used quite frequently in nevow and exposed to user code – as bytes. Also, at some point nevow has to produce bytes, as that is what needs to go down the line.

In between, however, I'd like to keep out bytes as much as possible and let people work with strings as far as possible. This document attempts to delineate the string/bytes perimeter.

Given that, unfortunately, that perimeter is long and twisted, the plan is to accept bytes and strings in several places (in particular, always for URIs), where bytes, for these purposes, are supposed to be in a (perhaps at some poin configurable) default encoding, which for now is utf-8 independent of the enviroment.

Use utils.toBytes or utils.toString to turn function arguments into strings or bytes as requrired.

Bytes within Twisted we're concerned with

The most important items that are byte strings within request, include:

  • uri
  • keys and values in args (this hurts a lot)
  • header keys (but header values are decoded)
  • prePath -- this is where segments come from; segments, however, are nevow interface and hence strings.
  • the arguments to write (in nevow, we accept strings, too)

At least cred.checkers.InMemoryUsernamePasswordDatabaseDontUse explicitly ASCII-encodes their usernames right now. Since that's what's used in the unit tests, I'm following ASCII-encoded usernames in guard. This seems insane. Anyone actually working with guard should look into this.

Bytes usage within nevow itself

While flatteners still return strings, what is passed on to twisted.web requests' write methods must, of course, be bytes. nevow.appserver.Requests make that translation using utils.toBytes; user code requiring non-UTF-8 encodings needs to translate to bytes itself at this point.

Since renderHTTP can (and is, indeed, encouraged to) write strings and the translation is done within nevow.Request.write (or similar), Page.renderSynchronously and Page.renderString return strings rather than bytes.

This is particularly relevant for unit tests: what is in the FakeRequest's accumulator is bytes.