Clarify whitespace and newline rules.

[mojombo]

This PR greatly clarifies the use of whitespace and newlines in TOML. A few notes and ramifications:

• Behavior on Windows is noted.
• Key and table names cannot contain any whitespace, even around brackets or dots.
• Keys and dot-separated table parts follow the exact same rules.

[ChristianSi]

Why did you delete these two lines?

[mojombo]

Indenting is covered by the previous statement that "Whitespace is ignored around key names and values.", so I thought it to be redundant. Also, the ability to indent is a weird way to segue into nested tables, and makes it sound as if indentation might carry some semantic value.

[ChristianSi]

I like the definition of what is and what isn't allowed in keys and table name (parts).

A question about the newline handling: this reads to me as if the newline conventions used on the system where a TOML file is created become a permanent part of multiline strings. So assuming the following string comes from a file using Windows newlines:

"""I'm a multiline
string."""

it would be byte-for-byte equivalent with

"I'm a multiline\r\nstring."

Assuming a TOML processor running on Unix reads that file and then creates an updated version, using Unix newlines, this seems to imply it would be forced to serialize the string as

"""I'm a multiline\r
string."""

or

"I'm a multiline\r\nstring."

Not sure if that's indeed the meaning of the spec. But if it is, I must say that I find it quite unintuitive and possibly annoying.

I think it might be better to say approximately the following:

• On reading, every occurrence of CRLF(0x0A 0x0D), standalone LF or standalone CR is converted into a newline.
• On writing, every newline is converted into the newline representation conventionally used on the platform where the TOML generator is running, i.e. either LF (Unix-style) or CRLF (Windows-style).

That's approximately what Python's "universal newline mode" does.

[mojombo]

cc @BurntSushi - I'd love your opinion on this.

[mojombo]

I should also mention that depending on the outcome of #220 the key/table name rules may change. This PR is primarily to get the newline issue nailed down.

[BurntSushi]

What if Newline was defined as either \r\n or \n. This leaves out a lone \r as qualifying as a new line, but I think this OK, unless it's still commonly used somewhere?

[ChristianSi]

👍

\r was once used on Mac, but Mac OS X changed that AFAIK.

[mojombo]

Ok, I've made a few changes to define newline as LF or CRLF and tightened up the copy to reflect that change. cc @BurntSushi @ChristianSi - I'd love your thoughts on the new text.

[mojombo]

@ChristianSi Ok, just pushed a small edit for that.

[BurntSushi]

LGTM. :-)

[mojombo]

Ok, thanks for your feedback on this, everyone. Merging!