fix whitespace creation

[yoshuawuyts]

This should fix #72

patch notes

• now creates strings on the server, leading to a 10x speed improvement for rendering
• detects if in Electron, Node, or browser and runs the right version
• fixes random text node creation in lists

[yoshuawuyts]

Ok, so right now things are not solved. The current state is:

• browser tests pass, behavior is patched
• server tests fail, we now rely on .lastChild in the critical path and min-document does not recognize that
• this means that running code on the server could already fail, but it wasn't caught by the tests

What I'm thinking we should do:

• unpublish the last version: it's broken and oh no that's not good
• merge the patch in this PR to master
• split browser & server code, and replace server code with https://github.com/shuhei/pelo/blob/master/index.js

We should get a 10x server render improvement in the process, but also do a semver major release. With that happening we could get cheeky and also add #69.

I have a feeling this is a fair amount of changes, so input would be greatly appreciated! Thanks heaps!

---

IRC dump

02:55 <yoshuawuyts> URGH
02:55 <yoshuawuyts> ok, so I got this whole mad thing working in the browser - fixed the regression
02:55 <yoshuawuyts> (talking about `bel`)
02:55 <yoshuawuyts> but now the server is failing
02:56 <yoshuawuyts> and worse: it's failing because of a pre-existing bug in `bel` - e.g. min-document is an incomplete DOM implementation, and stuff like .nodeValue and .lastChild does not exist
02:57 <yoshuawuyts> so either: 1) we do a revert and pretend nothing changed. This will disallow child-node reordering tho. Or 2) we merge and break server rendering
02:57 <yoshuawuyts> I'm not thrilled about either option
02:59 <yoshuawuyts> well, technically we're not breaking server rendering - it's already broken - it just makes it, err, well more apparent
03:00 <yoshuawuyts> could be a cool opportunity to complete move to https://github.com/shuhei/pelo btw
03:00 <yoshuawuyts> option 3?
03:00 <yoshuawuyts> (anyone got thoughts on this?)
03:05 <sethvincent> would it be worth working on min-document or an alternative?
03:07 <yoshuawuyts> sethvincent: pelo uses string concatenation which provides a 20x speed improvement over object based approaches. I think if we're going to change anything specifically for server stuff that'd be the clearest path
03:07 <sethvincent> hmm, that makes sense

[yoshuawuyts]

woot, it works!

[yoshuawuyts]

Ok, so we're now running pelo in node / electron main process, and regular ol' bel code in electron renderer / browser. This means we got a 10x or so perf improvement for rendering in Node. Yay!

Semver major because the intermediate representation on the server also changed. Thanks!

[mantoni]

That little whitspace fix got out of hand, right? 😆

[yoshuawuyts]

v5.0.0