Why YAML sucks?

heikkiket@programming.dev to Programming@programming.dev – 2 points –

I feel that Yaml sucks. I understand the need for such markup language but I think it sucks. Somehow it's clunky to use. Can you explain why?

72

I used to think json was the best until I found json lines or line delimited json. Thank me later. I use it all the time. You can append until you’re blue in the face. It’s great for log files. Each line is a valid json file.

YAML works great for small config files, or situations where your configuration is fully declarative. Go look at the Kubernetes API with its resources.

People think YAML sucks because everyone loves creating spaghetti config/templates with it.

One reason it tends to become an absolute unholy mess is because people work around the declarative nature of those APIs by shoving imperative code into it. Think complicated Helm charts with little snippets of logic and code all over the place. It just isn't really made for doing that.

It also forces your brain to switch back and forth between the two different paradigms. It doesn't just become hard to read, it becomes hard to reason about.

I don't like a thing, fellas. With that being all I've told you, please explain why I don't like that thing.

Can people stop hating on shit?

FOR FUCKS SAKE, negative reinforcement dopamine has RUINED THE FUCKING NET.

EVERYWHERE I GO there's someone bitching about something, hate circlejerks are unbelievably popular, people just love to hate on stuff.

You're ruining your thought patterns with all these social media negativity bullshit.

Fucking TOML users hate on fucking YAML fucking C++ users hate Rust fucking Rust users hate literally everything under the sun and are insufferable to work with

EVERYONE, fucking CHILL

Can you stop hating on haters? Thanks 😄

No seeds no stems no stress my guy. The Internet is a great place for complaining. Readers can downvote and move on, everyone gets what they want.

Yeah TBH I like yaml. Sure its not the best ever, but its not the worst it could possibly be.

For config its not terrible. For ansible playbooks its again... not terrible.

Why is everyone always hating on something which is just kinda mid.

I dream of a life where I use YAML but all my configs are stuck in XML. People can complain but there's always worse options.

Any language in which whitespace has syntactic value is intrinsically flawed.

Can't speak to your specific issues, but that's why yaml will always suck.

Haskell supports both semantic whitespace and explicit delimiters, and somehow almost everybody that uses the language disagrees with you.

But anyway, for all the problems of YAML, this one isn't even relevant enough to point out. Even if you agree it's a problem. (And I agree that the YAML semantic whitespace is horrible.) If YAML was a much better language, it would be worth arguing whether semantic whitespace breaks it or not.

Yeah but Haskell is mostly used by mathematicians..

Not any language. I code professionally in F# which has semantic whitespace and it has literally never been an issue for me or my team. In contrast to Python, it’s a compiled language and the compiler is quite strict, so that probably helps.

As a serialization format, agree 100%, but would Python really be better if it switched to braces?

Yes, I think so. The downside with Python comes when refactoring the code. There’s always this double checking if the code is correctly indented after the refactor. Sometimes small mistakes creep in.

It’s really hard to tell when Python code is incorrectly indented. It’s often still valid Python code, but you can’t tell if it’s wrong unless you know the intention of the code.

In order languages it’s always obvious when code is incorrectly indented. There’s no ambiguity.

Can address it by writing code that doesn't depend much on indentation, which also makes code more linear and easier to follow.

I think this is just familiarity. I never have issues with indentation, but when refactoring js I'm always like hey who's fucking brace is this

It’s only hard to tell indentation in Python when the code block gets longer than about a screen, which is usually a sign the code should be refactored into smaller methods.

People hate hearing that they are bad coders 😂

You and the other guy are saying to focus on writing code with less indentation and using smaller methods, and you both got downvoted.

I fully agree, small methods all the way, and when that's not possible it's time to refactor into possibility!

Or a sign you should get a bigger screen 😂 (j/k)

would Python really be better if it switched to braces?

Yes. A thousand times, yes.

Yes it would - look at optional braces for short if expressions in C family languages and why it's so discouraged in large projects. Terminating characters are absolutely worth the cost of an extra LoC

I started in C before moving on to C++, Java, Ruby and Python.

I’ve had more bugs from missing braces than from misaligned whitespace because the latter is far more obvious when looking at a block of code.

Compilation error or run time errors? One is a gift and the other is a curse.

False dichotomy. Optional braces are bad practice because they mislead the programmer that is adding an additional clause to the block.

This misleading behavior wouldn't happen in Python, as it would either be invalid syntax, or it would be part of the block.

Indentation problems are pretty obvious to the reader. Even more than missing or unbalanced braces.

They may be obvious to the reader but they may be impossible to see if tabs and spaces are mixed together.

Closing tokens are always clearer.

That misleading behavior does happen in Python. The next programmer that comes along can’t tell if the original programmer fucked it up and didn’t unindent to put a statement outside of the block or if they meant to put it inside the block. I’ve debugged this one too many times and it takes hours each time because it’s impossible to see the bug at all!!

The misleading behavior is about what you expect to execute in the source code you're looking at vs what's actually executed.

What you describe is a logic ambiguity that can happen in any program / language.

I don’t agree. It’s a direct result of whitespace, which does not happen if you don’t use whitespace. For example it can happen in Java and kotlin, but only if you use if statements without braces, which you pretty much never see. If you do see it you know to look out for the exact issue I described. That’s not possible in Python, since there is no alternative.

YAML sucks because, among other things, indenting it is not obvious.

In contrast, the only mistake of Python when it comes to whitespaces was allowing hard tabs, which makes it too easy to mix them if your editor is not configured.

Improper indentation stands out more than missing or unbalanced braces and it's really not an issue to delimit code blocks.

3 more...

I now use Scala 3, and very happy with syntactic whitespace (combined with an intelligent compiler)

3 more...

We all know it sucks. I have no idea why people use it instead of anything else. My workday is jammed with fucking ansible which, while also being so ghetto that we were easily doing more with less in 2002, uses So.much.fucking.yaml . Just when you think ansible couldn't get more clunky and useless and slow, it also is configured in yaml.

Following along with the style of your own post: YAML doesn't suck, because I feel so.

Thanks for asking.

White space/indentation as a construct of the syntax.

It’s why I have a hard time with python.

People have their likes and dislikes. Nothing wrong with that.

Because people over use it. YAML is pretty good for short config files that need to be human readable but it falls apart with complex multi line strings and escaping.

I think there are much better clearly delimited for machine reading purposes formats out there that you should prefer if you're writing a really heavy config file and, tbh, I think for everything else .ini is probably "good enough".

At least use TOML if you like ini, there is no ini spec but TOML can look quite similar.

I agree - YAML is not suitable for complex cases that people use it in, like Terraform and Home Assistant. My pet peeve is a YAML config in a situation that really calls for more abstraction, like functions and variables. I'd like to see more use of the class of configuration languages that support that stuff, like Dhall, Cue, and Nickel.

There is another gotcha which is that YAML has more room for ambiguity than, say, JSON. YAML has a lot of ways to say true and false, and it's implicit quoting is a bit complex. So some values that you expect to be strings might be interpreted as something els.

What YAML does Terraform use? HCL is similar but different enough to YAML.

Oh, thanks for calling that out. I think I may have mixed up some of the frustrations I experienced at an old job.

One pattern I've noticed is people seeking a language that's better than {JSON,XML,INI,etc} at wrangling their slightly complex configuration files, noticing the additional features and type support offered by YAML, and assuming it will be a good solution.

Then, as their configs grow ever larger and more complex, they discover that expressing them in YAML requires large sections of deep nesting, long item sequences, and line wrapping. The syntax style that they saw working well in other places (e.g. certain programming languages) breaks down quickly at that level of complexity, making it difficult for humans to correctly write and follow, and leading to frequent errors.

YAML doesn't suck for small stuff, IMHO. (But it is more complex than necessary for small stuff.)

For things likely to grow to medium-large size or complexity, I would recommend either breaking up the data into separate files, or looking for a different config/serialization language.

YAML had comments and trailing commas, therefore it's objectively better than JSON. If you want a compromise solution that mostly looks like JSON, try JSON5.

Tons of people making Python comparisons regarding indentation here. I disagree. If you make an indentation error in Python, you will usually notice it right away. On the one hand because the logic is off or you're referencing stuff that's not in scope, on the other because if you are a sane person, you use a formatter and a linter when writing code.

The places you can make these error are also very limited. At most at the very beginning and very end of a block. I can remember a single indentation error I only caught during debugging and that's it. 99% of the time your linter will catch them.

YAML is much worse in that regard, because you are not programming, you are structuring data. There is a high chance nothing will immediately go wrong. Items have default values, high-level languages might hide mistakes, badly trained programmers might be quick to cast stuff and don't question it, and most of the time tools can't help you either, because they cannot know you meant to create a different structure.

That said, while I much prefer TOML for being significantly simpler, I can't say YAML doesn't get the job done. It's also very readable as long as you don't go crazy with nesting. What's annoying about it is the amount of very subtle mistakes it allows you to make. I get super anxious when writing YAML.

It's docs are garbage, but the language is quite simple and human readable even for non-techies. I think it's a bit too easy at times, resulting in people just kinda winging it

Somehow it’s clunky to use.

huh?

I find developing GitHub CI in YAML clunky.

I don't find configuring a simple service via YAML config, with a preset showing me and explaining what I can do clunky.

I don’t like YAML because it’s overly complicated. The specification is like 80 pages long. How the hell did they think that was a good idea?

JSON on the other hand is super simple. It doesn’t do more than it needs to.

Just compare this: https://yaml.org/spec/1.2.2/

With this: https://www.json.org/json-en.html

The entire JSON specification is shorter than just the table of contents of the YAML specification!

Another thing I like about JSON is that you can format it however you want with the whitespace. Want everything on one line? Just write everything on one line!

If data can be represented as a JSON, then there’s generally only one way to represent it in JSON (apart from whitespace). In YAML the same data can be represented in 1000s of different ways. You pick one.

I will never forgive JSON for not allowing commas after the last element in a list.

To paraphrase: There are two kinds of markup languages. Those that people complain about and those that nobody uses.

There is no silver bullet that will work perfectly for all use cases and we also don't want to use 100 different tools. So people use things that aren't perfect. But they're good enough. I don't think YAML is perfect and I still use it, because people know it and there are tons of tools already available.

YAML is fine if you use a subset (don't use the advanced features - not like you know those anyway) and use explicit strings (always add " to strings), otherwise things may be cast when you did not intend values to be cast.

Example:

country: NO (Norway) will be cast to country: False, because it'll cast no (regardless from casing) to false, and yes to true.

country: "NO" should not be cast.

Programmers hate everything. You could design a spec which serenades you with angel song and feeds you chocolate dipped grapes and someone would be like: This is awful, my usecase is being a dog.

Sure there aren't many things that are universally loved. I mean I can't really think of anything that doesn't have some flaw.

But that doesn't mean everything is equal! What would you rather program with, Visual Basic or Go? PHP or Typescript? If you polled people there are obvious winners.

Hey would you rather build from wood or steel?

What glue is better: 2 part epoxy or PVA?

Do you prefer soap or bleach as cleaning agent?

Would you rather build from wood or tissue paper?

What glue is better: 2 part epoxy or pritt stick?

Do you prefer soap or ash as a cleaning agent?

Unlike tissue paper yaml is actually fit for purpose. I actually don't know of any lang that literally can't run a program. The most you could stretch what you're saying to is that some esolangs are akin to making bricks of packed tissues to build with. They are art projects not serious submissions though.

I don't like js as much as anyone else but as evidenced by reality it works. Programmers need to stop sniffing their own farts, you have such strong opinions about the most insane shit when at most you should be talking about narrower scopes for use and trade offs.

It sucks the same way Python sucks. Some people just really don't like indentation-based syntax. I'm one of them, so I dislike both formats. However, if you groove on that sort of thing, I don't think YAML is any worse than any other markup.

Oddly, I get along with Haskell, which also used indentation for scoping/delimiting; I can't explain that, except that, somehow, indentation-based syntax seems to fit better with functional languages. But I have no clear argument about why; it's just an oddity in my aesthetics.

You can't say python's whitespace usage is as bad as yaml's. YAML mixes 2 and 4 spaces all the time. Python scripts don't run if you write this kind of crap.

And whitespaces is really just the tip of the iceberg of YAML problems...

Yaml is fundamentally the same as the json and xml it has mostly replaced (and the toml that didn't manage to replace yaml)... it's a data serialization format and just doesn't have any facility for making abstractions, which are the main tool we human use to deal with complexity.

JSON and YAML aren't the same as XML. The attribute/child distinction in XML, and the fact that every object has a tag name associated with it, make it a PITA to map into the data primitives of any programming language I know.

Yes, XML is different than JSON and YAML, but it's not particularly easier or harder to manually read/edit than JSON or YAML are (IMO the are all a pain, each in its own way).

If you want to look at it from the programmer's side (which is not what OP was talking about)... marshalling/unmarshalling has been a solved issue for at least 20yrs now :) just have a library do it for you (do map json/yaml properties to you objects manually?).

You don't need to worry about attributes/child elements: `` and jack will work the same (ok, this may depend on what language/library you pick - the lib I used back in the day worked either way).

If anything, the issue with XML is all the unnecessarily complicated stuff they added to its "core" (eg. CDATA, namespaces, non-standalone documents, ...) and all the unnecessarily complicated technologies/standards they developed around XML (from Xinclude to SOAP and many others)... but just ignore that BS (like the rest of the world does) and you'll mostly be fine :)