Add whitespace stripping in the input

[Kijewski]

In HTML sources indentation can aid readability, but it only bloats the
output. This PR adds the argument "strip" to templates, which lets you
automatically strip the template input:

• "none": No stripping of the input. The generator still strips spaces
  at the back of the input.
• "tail": The current default, i.e. pop a single newline character at
  the end.
• "trim": remove all spaces at the front and back of a line, and remove
  empty lines.
• "eager": like "trim", but replace multiple whitespaces with a single
  space.

The automatic stripping does not try to understand the input at all. It
does not know what xml:space="preserve", <pre> or <textarea>
means.

[vallentin]

I "don't" like the idea of trim. Because whitespace is or can be significant in HTML/CSS, i.e. with inline menus :)

Also, can this be used for #581? Because I looked a bit at that before Christmas, but then I got busy with Christmas and New Year's.

Wait, I misinterpreted it, it's only for whole lines, right? Hmm.

[Kijewski]

It's like trim() is called on every line, then empty lines are dropped, and for the eager case s/\s+/ /g applied.

Of course, white spaces are significant in <pre> and <textarea>. But how often do you see <pre> nowadays? When do you use a <textarea> that is populated by static data, not by {{data}}? For these two cases you cannot apply the automatic stripping.

[djc]

Awesome that you're working on this, did you guys already see rust-lang/rust#92526? It's pretty amazing! This also came up there. I'll link to this PR in from there so that we can feedback from them.

I guess this also needs some good documentation in the book.

My main point of feedback is that I think we'd want a way to change this setting on a crate-wide basis, in the askama.toml configuration. There, we might make it part of an extension/type-based configuration (I think we already have some?) so that people can easily apply this to HTML but not plain text templates. I was going to suggest that we ditch any kind of per-template configuration but it just occurs to me that that might make it harder to test this well (end-to-end) the way you've done here.

[camelid]

Am I correct in thinking that this is all done at compile-time? If so, that makes this extra-exciting :)

[djc]

Am I correct in thinking that this is all done at compile-time? If so, that makes this extra-exciting :)

Yes, that would all be done at compile time.

[Kijewski]

I moved Strip from askama_shared's root into askama_shared::input. That's were the other arguments life.

[djc]

Did you have any thoughts about the configuration approach?

[Kijewski]

Did you have any thoughts about the configuration approach?

Yes, that should be a project-wide configuration. I added the optional strip parameter to the config, too.

But I actually have not idea how to write a test for it. Can you give me some pointers?

[Kijewski]

I'm not convinced anymore that this is the best implementation.

Most likely it would be simpler and better to use Strip::apply() in Generator::write_buf_writable() instead of preformatting the input.

[Kijewski]

I simplified the PR. Now only "none" and "eager" are allowed.

[Kijewski]

My "simplified" version does not work. The "complicated" version https://github.com/djc/askama/commit/766d049536c17a089fd7031e639d9f4999c9e64c did. Reverted and rebased.

[Kijewski]

Superseded by #673.