Alignment rules for linting and serialization

[stasm]

In #24 (comment), @zbraniecki, @Pike, @flodolo and I had a discussion about the standard formatting and whitespace alignment of values in FTL.

The argument raised by @flodolo against pretty-aligning of values was that it breaks the blame and creates diff noise.

@zbraniecki then created an example file and expressed concerns about its readability without value-aligning:

0.1 syntax: https://github.com/projectfluent/fluent-rs/blob/0.1/tests/workload-low.ftl
0.2 syntax: https://github.com/projectfluent/fluent-rs/blob/master/tests/workload-low.ftl

To remind principles:
- the file should be readable without prior knowledge of FTL
- the file should be easy to scan through
- the file should be editable with minimal FTL knowledge

For one, I'd say we did manage to make it look cleaner (less sigils!) But also, I believe the attribute alignment significantly helps with readability, so I'm not sure if I agree that preserving blame is worth it.

I'm creating a new issue to give this topic its own place for discussion.

[stasm]

For comparison, here's the same file without any special alignment of values: https://gist.github.com/stasm/020a827c37004b44dbd61fd3a23817fe

[Pike]

Thanks for the example, I was looking for one. I finished up the brand.ftl part in https://gist.github.com/Pike/c1301c217e8fdb4246e55334049a5863#file-workload-low-ftl-L747

[stasm]

Thanks, I fixed my gist as well. (The vim macro that I used only removed the extra white space after the =.)

[Pike]

On a general note, I'm not a fan of putting linting rules into the implementation of serializers. It's mostly an even political opinion on mine, humans are right, machines are wrong. I somewhat know that there's a lot of linting in fashion today, but that might be something to implement at a different level than our core serializers.

On the examples here:

The example is a bit weird, in that the alignment rules are not coming from the IDs, but from the attributes. As such, the indention is determined by the grammar and not by the file. You'd only hit a change in indention across unrelated entries if you'd start using placeholder as attribute or something. Which is why I also put the brand part on the things to realign.

https://bugzilla.mozilla.org/attachment.cgi?id=8829980&action=diff#a/mobile/android/locales/en-US/mobile/aboutDownloads.ftl_sec2 might be an interesting other use-case.

Also, practically, we see alignment almost never on the values, but on the =, so

foo       = Foo

instead of

foo =       Foo

The latter is what the current examples show, and I find the unindented file easier to read. It might be just because of = being closer to the value.

Also, kinda specific to our chosen example, there's a tighter pattern of

some_id_thing
    .value = Value

and then the .accesskey stuff is a bit off in indention, which makes it almost easier to ignore.