YAML files: Statically typed, confusing if poorly formatted

0 users browsing Discussion. | 1 guest | 41 bots

Main » Discussion » YAML files: Statically typed, confusing if poorly formatted

Pages: 1

nyanpasu64

Posted on 19-03-01, 10:17

Post: #26 of 77
Since: 10-31-18

Last post: 1189 days
Last view: 1116 days

(branching off https://helmet.kafuka.org/bboard/thread.php?pid=1719#1719 )

https://github.com/ambv/black places trailing commas after each line in a function call or list literal, to explicitly make swapping/commenting elements easy.

To the contrary, I noticed that ruamel.yaml's YAML formatting makes it hard to comment-out or reorder the first line:

State 1:


orders:
- name: Apples
  price: 1.5
- name: Onions
  price: 2
"2d_identity_matrix":  # quotes optional
- - 1

Diff 2: Newline before [maps within lists]. (Neither PyCharm nor ruamel.yaml does this)


orders:
-
  name: Apples
  price: 1.5
-
  name: Onions
  price: 2
"2d_identity_matrix":
-
  - 1

Diff 3: indenting lists within maps. (PyCharm does this, ruamel.yaml does not)


orders:
  -
    name: Apples
    price: 1.5
  -
    name: Onions
    price: 2
"2d_identity_matrix":
  -
    - 1

- All the above parse to the same tree... I think YAML is too permissive.
- There is no "standard" formatting style. I think State 1 confuses people unfamiliar with YAML (like myself at first).
- ruamel.yaml uses State 1.
- The YAML 1.2 spec uses Option 3, which I like. https://yaml.org/spec/1.2/spec.html#id2760193
- Pycharm does 3 but not 2 (but it's completely configurable).
- I dislike the first 2 are dumb because you have a Dict[List[Dict]] but the innermost dict is only indented 1 level.
--------------------
One thing I like about YAML is (unlike JSON) you can cleanly represent strongly-typed objects (call it tagged unions, or subclass polymorphism, whatever you prefer):


!MyAppConfig
output: !FFmpegOutputConfig
  foo: bar

I use this style in my project https://github.com/jimbo1qaz/corrscope . In Python, I store all config as statically-type-hinted attrs/dataclasses. I use ruamel.yaml to serialize the config to YAML files, and expose the names of the config classes to the YAML dump.
My app auto-generates YAML files which are subsequently hand-edited (primarily GUI-edited, ever since I added one). I believe that !Tagged YAML files are more self-documenting and easier for *me* to cross-reference between YAML and source code (when I use my own program).
But it turns out that ever since moving "ffmpeg vs ffplay" to a runtime flag, I only every dump 1 class to any given parameter, and currently have no uses for polymorphism when loading (other than generating type errors at runtime)... But I still feel it's good to have (I have no issues with it, and don't want to rewrite my working code).
-------
I think https://github.com/ron-rs/ron has a similar "typed/tagged object" feature.
Sadly I feel this "killer feature" of YAML is underutilized.
- https://github.com/dtolnay/serde-yaml/issues/115
Rust Serde added support for *reading* YAML tags. But it *writes* them the "dumb" way, abusing a 1-element map to act as a type tag:


time: TimeOfDay: asdf
# is equal to
time:
  TimeOfDay: asdf

I can't imagine why. Rust is a type-heavy language with tagged unions as a central concept.
Appveyor also abuses a 1-element map as a type tag:


build_script:
  - exit  # run everywhere
  - cmd: exit  # cmd batch
  - sh: exit  # bash
  - ps: exit  # powershell

I could imagine people who only use dynamically typed languages would prefer the above to:


build_script:
  - exit  # run everywhere
  - !cmd exit  # cmd batch
  - !sh exit  # bash
  - !ps exit  # powershell

Pages: 1

Main » Discussion » YAML files: Statically typed, confusing if poorly formatted