I should depend on raw-body about body

[jonathanong]

probably depend - raw-body doesn't actually parse the string. i was thinking about making a body-parser module, but i guess this module is what that would be. people in connect/express just want too many options, it's pretty annoying.

i'm not sure what's the point of using the string decoder vs. buffer.toString() though. only issue i can think about that is getting the correct charset from the request body if it isn't utf8.

also, not including string parser is easier for me since then i can implement the string parser however i want. see: https://github.com/koajs/body-parser/blob/master/index.js

[Raynos]

@jonathanong So I don't know whether string decoder of Buffer.concat(buffers) is better. should have a bench mark.

I copied @izs code from npm-www which uses StringDecoder which is used to handle two bytes UTF8 properly since body += String(chunk) is wrong if chunk has a trailing byte.

[Raynos]

@jonathanong btw I'm going through express & connect and rewriting them using small modules. If you have any modules to recommend using you should add them to the wiki page !

[jonathanong]

yeah, i don't know which one is faster, but i prefer fewer lines of code. i doubt any performance differences are significant, though.

[jonathanong]

sick! i actually wanted to do that, but unfortunately a lot of the middleware rely on connect patches like res.on('header').

most of the remaining connect issues are sessions, but sessions right now are such a mess that i don't want to touch them. for koa, tj's been thinking about just making tests for a session "spec" which implementors can use instead of relying on connect.session(). it should be easier to make better session middleware that way. let me know if you're interesting in doing that for connect/express.

[Raynos]

@jonathanong look at generic-session & redsess & level-session. that's what I use, I like them.

but of course it doesn't have that crazy asynchronously load session for each req before route and asynchronously save session for each res.end()

[isaacs]

Buffer.concat is not better unless for some reason you already HAVE a list of buffers. If you're collecting a stream into a string, then StringDecoder is better in both time and space.

[Raynos]

@isaacs I broke out your npm-www things, should I PR them back into npm-www ?

[jonathanong]

is the impact significant?

also, why even bother using string decoder then? why not just do req.setEncoding()?

[Raynos]

@jonathanong req.setEncoding will set the encoding on the source namely the req. Using stringDecoder in body is setting the encoding on the destination, namely the streaming body parser.

If you pipe the req to multiple location then maybe not all locations want the data chunks to be UTF8

[jonathanong]

oh, that's true. there was a bug where someone decided to set the encoding for no reason.

[jonathanong]

actually, that wouldn't matter. you assume the data is in buffers as well.

either way, i'll add encoding support to raw-body.

[Raynos]

@jonathanong I did a code review of raw-body and added some issues! Once those are handled I'll update this to depend on raw-body instead.

[Raynos]

I updated the code to depend on Raynos/raw-body

Waiting for @jonathanong to publish a version of raw-body.

Added @jonathanong as a collaborator aswell. You can make alterations. If your not sure about stuff make PRs and I'll review it.

[Raynos]

Now depends on raw-body form npm.

Published as v4.0.2