I didn't. There wasn't much to the blog post. In fact, the blog musings are the same criticism we've heard a thousand times. But please don't suggest that I missed some deep point in a trivial blog post when you really just mean "I disagree".
I even pointed out a real-world concrete example where JSON config is made pleasant to edit by hand for humans: an IDE like VS Code that pulls in JSON schema.
Though since you only doubled down on the blog post's opinion instead of responding to a single point I made, can I say that you missed the points of my comment? You didn't even address my criticism by pitching an alternative. How are we supposed to have a conversation?
I think the debate here is what "editable by humans" means. If you're using a tool for editing JSON like VS Code, I don't believe you're really editing it by hand. You're using a program that has specific things for JSON that give you to have a pleasant experience editing. I believe editing by hand means using a standard plain text editor and it's a lot worse.
Formats that use braces are relatively easy to parse by machines but can be harder to read by humans. Utilizing white space makes things easier for humans to read. A JSON file could have everything on one line. But something like YAML it forces you to use white space and new lines which helps humans to read it.
The only significant thing missing from JSON is comments. JSON5 or convention can solve that, so it's not as bad as you'd expect.
Still, editing JSON directly by an IDE that still exposes its syntax is "editing by hand", just as editing C in Vim with ctags is still editing by hand.
I do have to note though that VSCode uses JSON with extensions. Most obviously, it allows comments. But there are other minor things that make it easier to write (and parse) - e.g. it allows trailing commas in literals.
It's not people to have to worry about. It's the machine generated code. But now that I think about it more, YAML is a bad example since it's a superset of JSON. So a machine could generate really ugly YAML too.
But do you know what sucks even harder? XML. If I can replace an XML configuration file with JSON then screw it.
In my experience, the biggest reason people hate it is because it has too many ways of structuring data, and it allows people who design the schemas to go overboard. Attributes or child nodes? Multiple namespaces or just extend the one you have? CDATA or embedding?
That said, simple XML is as good as simple JSON. XML is fine if you keep it simple when designing the schema. AND of course one can screw up with JSON too. But after almost 20 years of people over-engineering their XML schemas, I can't fault anyone for choosing a simpler data format.
Recently I've had to work with XML config files for a tool called Oozie, which is a data pipeline scheduling tool within the hadoop/spark ecosystem, and it has been soul crushingly tedious for me. Everything feels verbose and opaque, the documentation seems to prioritize enumerating every possible configuration option over providing minimum viable configs for common use cases.
JSON configs often just feel more simple and developer friendly. I'd say this has less to do with technical differences between JSON and XML and more to do with how "modern" software systems have been designed to be more ergonomic for developers/administrators and these modern systems happen to be more likely to use JSON.
It's also fun to hear non-technical people at work talk about "Jason files".
Ha! That one always makes me smile. One of the best exchanges I’ve heard was something like:
“... and we’ll have JSON handle all the serialization”
“Jason? Oh we hired a new engineer?”
I usually go the other way; Jasons I know have their names represented as "JSON" in my mind, and sometimes in written form.
I really wish everyone would think of JSON as a French word. "Jay-sonne" (or jay-SAWN). Not "Jase-in"... it would help keep the confusion to a minimum.
Ha - https://www.youtube.com/watch?v=zhVdWQWKRqM He starts by saying "Jason" but then suggests the French version. Love it!
I've heard people prefer JSON over XML because XML is too complex... and now we have json schemas, jsonpath... the namespaces can't be far away.
I'll admit that XML probably would be better without entities and the crazy whitespace handling.
In JSON they’re plain and simple. You even get to use good ‘ol [ ] delimiters.
In XML? It’s a sea of open and close tags.
Now do an array of arrays. Then hand-edit it.
Like I said, it’s no better. But it’s way easier to read.
People can easily read and edit it with deeply nested structures, attributes, and even malformed data. Frontend frameworks using components and variations like JSX also maintain the same style because it's very natural to use.
Now try taking a typical HTML document and expressing it in JSON. Even an empty page would be completely unmanageable without tooling. People have a strange reaction when they hear "XML" but it's much more structured, usable, and widespread than you think.
Also HTML is not XML. HTML is non-strict (the spec is it’s parser not its format). HTML doesn’t have schema files. HTML doesn’t understand self-closing tags. HTML does have void elements (img, br, input, etc). HTML is designed for humans to write. And it’s a tool fit for its purpose, unlike the abomination that is XML.
Scroll down and I think it shows it perfectly how (X|HT)ML tags are much simpler than JSON syntax once things get complex and nested. It's not that hard to define basic primitives like we do with HTML (which has dozens of tags) but XML also lets you define your own schema if necessary to make things more compact.
I understand the technicalities between HTML vs XML but I don't see how it makes any practical difference when you're editing a bunch of tags in a text file. It's the same thing. The structure looks identical. What is the actual issue that makes HTML easy but XML hard?
Well, until you do need some metadata for that key-value pair. Which is why even in many JSON schemas it's pretty common to get something like { "key": "key", "value": "value", ... } (where ... is usually empty in practice).
The problem with MSBuild, for the most part, isn't XML - it's its own data model (which is not XDM, by the way).
My main point is in xml the closing tags are unnecessary, and you need an opening tag for every value where it’s obvious from nested context what that value is supposed to be. Xml is very redundant, there’s just all this zero entropy text everywhere which conveys no information. I disagree with your metadata in json example, why not just add more data to the value? The data model of msbuild is also annoying, but I’ve worked with some dotnet core projects now and using json as the project format definitely saves typing and actually allows you to intuit what a project is doing rather than just bombarding you with useless text.
And with respect to data and metadata, what you propose is not fundamentally different in terms of verbosity:
<foo metadata="bar">baz</foo>
{
"foo": {"value": "baz", "metadata": "bar"}
}
It’s just crazy to look at tbh, and it was annoying to type on my phone. I’m amazed people are still arguing in favor of it. In reality you just want the markup to transmit the most information in the least amount of bits. This is a measurable quantity and xml objectively sucks at it.
If that were the case, we'd be using binary serialization everywhere. But you yourself are making an argument that ease of writing matters. So does ease of reading. Overly verbose markup is a tax on both, but so is extreme brevity.
Anyway, this particular thread was a discussion to address a very specific point made in the article that overstated XML verbosity over JSON. I'm not actually arguing that XML is perfect, or even "good enough". Its syntax is overly verbose, and its data model has pointless distinctions and arbitrary restrictions. But it also had many good ideas, and it's unfortunate that those get ignored in the quest of simplifying everything - and then later, when the issues that were the original motivation for those ideas are rediscovered, that wheel gets reinvented in dozens of flawed and mutually incompatible ways.
Do you have difficulty in reading a large HTML document? That's as verbose and repetitive as XML but also usually filled with junk comments and malformed tags. If you don't find it hard, then why is XML different?
HTML is not XML and even then few developers today use HTML without some sort of processing involved. HTML is also not a language used for configuration files, so it is really off-topic.
The question is, do people prefer XML configuration over JSON configuration. The answer is that people generally prefer JSON. JSON just maps far better to data types we use in everyday programming. Have you seen some of the common ways of expressing key-value pairs in XML? It's horrible.
> Now try taking a typical HTML document and expressing it in JSON.
That's not the problem anybody is trying to solve here. We're talking about configuration files, not complex documents. But let's say I wanted to solve that problem, the representation would look like this:
Not that bad at all.
Why is it so horrible? Have you ever typed a <ul><li> list? Or <input name="key" value="value">? What exactly is so difficult about this syntax in XML but completely fine in HTML?
The point isn't whether you can represent a document in JSON, it's about how easy it is for a human to manage it. The XML/HTML structure is far easier as things get more complex. The page you linked to shows this quite clearly even with the small "Bulleted List Example" at the bottom.
There's a lot of practical differences when actually implementing configuration with either XML or JSON, due to technical distinctions.
> The point isn't whether you can represent a document in JSON, it's about how easy it is for a human to manage it.
If XML is easier to manage, why are people shifting from XML and to JSON for configuration? You may have some experience with HTML, but have you actually used some of these XML abominations that are used for configuration?
> The XML/HTML structure is far easier as things get more complex.
Configuration files shouldn't be complex. They're mostly key-value pairs, perhaps a couple of lists here and there and a modest amount of nesting. XML is complex to begin with.
Perhaps at some stage, for some use-cases, XML starts becoming simpler to edit. Configuration files generally isn't one of them.
> The page you linked to shows this quite clearly even with the small "Bulleted List Example" at the bottom.
It's an example of representing XML-like structure with JSON, which is fairly easy. Try it the other way around, things get hairy. If you were to represent the data as JSON, you wouldn't write it like that.
I'm just not sure what the argument against this is other than verbosity. Config files shouldn't be complex? Sure, but what if they are? Are people really moving to JSON everywhere or is it just tied to the rise of Javascript and web frameworks?
Even if I would agree with this (which I don't), if it doesn't matter then you should pick JSON because it's easier to work with as a developer and it doesn't require XML parsing as a dependency.
> My evidence is how many people easily edit large complex document structures in HTML already without issue.
This isn't evidence at all. An HTML document is not a configuration file. Even then, people rarely author large HTML documents by hand these days. It's more likely that they're using a simpler language like Markdown to author documents.
> Are people really moving to JSON everywhere or is it just tied to the rise of Javascript and web frameworks?
If Javascript and web frameworks are responsible for the rise of JSON, but XML is better, why didn't they stick with XML? Remember, SOAP was XML. AJAX stands for "asynchronous Javascript and XML". The way to do an HTTP request (before fetch) from Javascript was called (inappropriately) XMLHTTPRequest. JSON on the other hand was just an informal spec with several flaws, yet it won out over XML.
2) JSON does need to be parsed and does add a dependency. Javascript is not the only language out there.
3) HTML not being config files is not the point. The document/tag structure is identical. If you can edit large complex HTML files (regardless of how they are generated) then you can edit large complex XML files. And it's rather easy to do so, against the claims that XML is so hard to work with.
4) There are plenty of XML configs out there, backing just about every piece of software you use. If you only focus on web/js projects then JSON "won out" because it is a simple dump of the in-memory representation rather than a more formal serialization like XML. The missing features like JsonPath and Json Schema are now being added back to turn it into a proper serialization format. When storing configs, these schema features are rather important.
2) A JSON parser is a far smaller and simpler dependency than an XML parser, especially when we're adding XPath to the mix.
3) XML isn't necessarily hard, it's tedious.
4) There's plenty of legacy software out there using all kinds of stuff for config. You don't see a lot of people choosing XML these days. You see YAML or TOML or JSON, even though all of these have issues of their own. JSON happens to be the simplest and most commonly supported.
This still applies then
In my world, config files are machine-parsed but human-generated and modified. JSON is just easier to work with.
const config = require('config.json');
console.log(config.apiKey);
With XML you often have to use things like XPATH etc. to extract the data which is a lot more effort.
I don't think people realize that JSON doesn't actually have a querying system at all, you have to deserialize it to an object to use. There is the coming JsonPath standard but that's not well supported yet and is pretty much the same as XPath.
But you’re not wrong. Yes you have to deserialise it. Hence it being an Object Notation system.
While you say JSON is trying to catch up with XML in regards to XPath, JSON is trying to catch up to XML with things like JSON Schema.
Absolutely. I would say XSD is the strongest advantage of XML over JSON right now. XSD still has many things to improve but JSON schema is even further behind.
You don't have to deserialize it to an object, you just do that because it's convenient in Javascript (and other dynamic language). It's a feature.
Now try using XML like an object, what do you get? A DOM. Which is fine if you wanted a DOM, but I don't want a DOM. I don't need XPath or any of that stuff. These are tools to deal with the complexity of XML, which I don't have, because I am using JSON.
So you're really saying that JSON deserializes to something that is a more natural fit for the language that you're using. And it's true in many cases; but also not so much in others, like when you're dealing with 64-bit integers, or dates, or all those other things that JSON doesn't spec because "complexity". In practice, it just means a proliferation of incompatible ways to represent these things, and utterly insane deserialization behavior in corner cases when implementations try to be "smart" to transparently compensate for JSON lacking something (e.g. https://github.com/JamesNK/Newtonsoft.Json/issues/862).
Conversely, if you are writing in a language that has integral support for XML - say, XQuery, or even VB.NET (https://docs.microsoft.com/en-us/dotnet/visual-basic/program...), the complexity is mostly not there. At the very least, if you control the format - which you have to, if you're in a position to decide what to use - then you can certainly stick to the subset of XML that is not anymore complex than JSON.
Sure, there are corner cases and limitations with JSON. I've never experienced that as a significant issue.
> Conversely, if you are writing in a language that has integral support for XML - say, XQuery, or even VB.NET...
I am not using any of that stuff, nor is there any reason for me to start using it.
> At the very least, if you control the format - which you have to, if you're in a position to decide what to use - then you can certainly stick to the subset of XML that is not anymore complex than JSON.
...or I can just use JSON.
XML (or rather, XDM, which is the appropriate level of abstraction to talk about this) has all those things:
https://www.w3.org/TR/xmlschema-2/#built-in-primitive-dataty...
Nor does it doesn't require an out-of-band schema - you can slap xsi:type on any element. And you can do that without breaking the data model, because namespaces keep data and metadata unambiguously separate, and code can easily deal with the former while being completely oblivious to the latter, unless it needs it.
JSON also has similar higher-level abstraction layers with more metadata. The problem is that nobody can agree on which one to use, or even whether to use one at all, and most code that's deserializing JSON in the wild is not going to be able to distinguish metadata from data.
Sure, you can add information until you arrive at a point where a string attribute or text content will be interpreted as a certain data type, but XML itself doesn't have it.
> JSON also has similar higher-level abstraction layers with more metadata.
JSON has all the basic data types built right in, there's no need for more metadata to do simple things. There's a reasonable mapping to basic data types and structures for almost any language.
> The problem is that nobody can agree on which one to use, or even whether to use one at all, and most code that's deserializing JSON in the wild is not going to be able to distinguish metadata from data.
...which is generally fine because of the aforementioned mapping. Your JSON library doesn't have to (and shouldn't) do any magic.
The only real advantage json has when it comes to deserialization is that it suggests to a human reader that keys are like object properties, leading to a very obvious deserialization strategy. But even that's a bit misleading, since json allows duplicate keys, just like xml, so a pragmatic deserialization library is going to make that feature impossible.
Make no mistake - the culture of straightforward deserialization is hugely valuable! But that's because of its history and other human factors more than the language itself.
Slightly less so as a JSON document, IMHO.
JsonPath is the proposed standard, and it looks pretty much like XPath. And that's before getting into Json Schema.
XML has other problems though.
EDN is a better Lisp-flavored candidate, IMO.
I don't understand what this means.
The problem with XML itself is being unnecessarily verbose (and thus difficult for both human and machine to read) for what's just a way to encode trees. Attributes are arguably XML's self-inflicted gunshot wound in the foot; you mainly need them because of visual noise caused by regular nodes.
mapping a 1-1 s-expression translation on HTML/XML/JSON/YAML etc. solves nothing. YAML has (had) code execution security problem and parser incompatibility issues. HTML/XML actually need namespaces in a few cases. JSON is abused as "compile target".
There can be no one format that works for everything. This is why I like TOML, it is really good at what it tries to do and stops there.
<thing>
<optattr>42</optattr>
...
</thing>
<thing optattr="42">
...
</thing>
https://stackoverflow.com/questions/48998034/does-toml-suppo...
It's even in the example file https://github.com/toml-lang/toml/blob/master/examples/examp...
General purpose until you need to store dates/times, self-referential structures, enums, etc.
json and toml are both serializations of data. I don't think the line you're trying to draw between "configuration file format" and "data structure format" is a well-defined line.
I wouldn't want to use if for config files or passing complex data but then it's not designed for that.
There is a distinction between markup languages e object notation languages. XML, HTML and Markdown are markup, JSON, YAML, TOML and most other are object notations.
(here my distinction, in general, is if they allow unquoted text)
To me it’s an improvement over JSON which is an improvement over XML.
But I'd posit that most - but not all - of the issues with xml config files have little to do with xml per se, and everything to do with the crazily detailed stuff people squeezed in it. You don't have to stick everything in a namespace, and nest even trivial things 3 deep with annotations for obvious types. And if you do those things in json it turns into a mess too.
SOAP is a perfect example of that. But: just because soap is messy doesn't imply xml everywhere has to be that.
But sure, xml has some downsides. Then again, so does json. Oh well!
Cant do that with json.
some_collection:
- first_entry
some_collection:
first_entry
I do know about a number of gotchas, and stepped on some briefly. Yaml is not ideal, but its set of tradeoffs is more suitable for human-editable configs, from my point of view.
It doesn’t matter the language or syntax, if the unit of structure doesn’t fit on the screen, it’s hard to work with.
One example from a larger file:
networks:
macvlan51:
driver: macvlan
driver_opts:
parent: eth0.51
ipam:
config:
- gateway: 192.168.51.1
subnet: 192.168.51.0/24
Edit: OK, it looks like hyphens indicate arrays in the schema but for me it feels like a guess as to which one of these to use for each particular case.
I’ve been editing long sublime-syntax files recently and I do not like it much. But I’m also not sure if JSON would be better for such a thing.
I have never understood this argument. I'm not saying it's the best, but I find it much much more readable than yml or xml.
But yes, if every implementation of JSON moved on to JSON5, I don't think we'd be having this discussion.
Example is below where an application could either extract values at startup, or is extracted when code is built and deployed?
# My App
## Intro about My App
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.
## Prod Server Settings
| Host | Port |Operating System| Type |
| -------------------------|:-------------:|:--------------:|--------|
| [email protected] | 80 | Debian |Web |
| [email protected] | 80 | Debian |Web |
| [email protected] | 1433 | aws |Database|
| webservice.com/api | 80 | aws |Database|
---------------------------------------------------------------------
## Support Information
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.
## Application Roles
|Name | Active Directory Group |Description |
|-----------------------|------------------------|-----------------------------------------|
|System Administrator | MyApp.SysAdmins | Details who should be assigned the role.|
|Finance Administrator | MyApp.FinAdmin | " " |
|Finance Delegate | MyApp.FinDelegate | " " |
|Finance Analyst | MyApp.FinAnalyst | " " |
-------------------------------------------------------------------------------------------|Depends on which kind of humans.
For the regular non tech types I agree JSON is poor. A couple of years ago I created a mini spec for a config file so that very low tech paper designers could configure an interactive book.
For programmers, JSON works fine in my experience.
The author mentions this, and I very strongly agree. What JSON significantly misses is the ability to have comments in your configuration files. I'm not going to hold up the Apache config files as an example of amazing config formats, but... here's an example: https://svn.apache.org/repos/infra/websites/cms/webgui/conf/...
That's awesome. It describes options that are available and not enabled, and what they do. Right. In. The. File. Comparing that to, say, webpack config... there's a pile of research needed to figure out what options are even available, let alone what they do.
All the other arguments for or against it boil down to personal preference, but not having comments is a real miss.
Cloud Formation is a great example of the failures of JSON. Specifying just a single dynamo table with indexes and scaling roles in JSON is a hard to read cluster of a mess. Even with a good idea and json collapse / expand help, large json files are just insufferable to deal with as a human
Programmers will maybe cope better but just being a programmer doesn't make you magically immune to poor readability and lack of comments.
Most of the points in that article apply to programmers as much as non-programmers.
(I agree there is a reasonable debate to be had about programmability - I'm on the fence about this one)
I don't hate it, but I really dislike it. It's definitely better than JSON for this purpose, but it's still quirky and a bit of a pain.
The use of brackets means if something is broken, you'll know explicitly because your IDE will throw a fit. Formats that are white-space dependent can have subtle errors that aren't easily recognizable on first glance.
The 2 problems with JSON are multiline string support and comment support, but JSON5 solves both of those problems. Sure JSON5 isn't JSON, but YAML isn't JSON either. No matter what you're going to need a library to parse a config file, so why not stick to something based on a widely used data-storage format?
{ useFluxCapacitorComment: "Enables time travel. true or false", useFluxCapacitor: true }
I don't mind the formatting, but I'm a professional developer, so my tolerance for complex UIs is abnormal.
Written on mobile, so that JSON may look like garbage.
E.g: npm/yarn will complain loudly if you add unexpected lines in package.json IIRC.
But comments are often code smell—a crutch for poor identifier naming and design. If your names and design are well thought out you rarely need comments and the consumer of your code should be able to understand it using the statically checked source alone—-versus manually having to parse your unchecked human language documentation.
Comments are important when they express things that code cannot.
Examples:
// See ticket XYZ-123 for discussion and rationale.
// We are sidestepping a bug in Docker 3.x here.
// TODO: Allow queue size per service version. Currently we're using max size for all.
Use as explanation of intent.
Use as clarification of code.
Use as warning of consequences.
https://www.amazon.com/Philosophy-Software-Design-John-Ouste...
["DoStuff", "EvenMore"/, "ops go back :("/]
Sure, but how often does that happen?
JSON is a fine format for storing configuration data, and any reasonably structured data. So is XML, protocol buffets, what have you.
JSON is a poor format for humans. For a configuration file that an actual human would edit by hand, and read and try to understand later, JSON is a pain.
Comments are mostly useless in a machine-generated and machine-consumed file. They are indispensable for humans, because they are a link to what machines are not doing.
Nice formatting is useless for a machine-consumed file, because they process a stream of bytes. They are indispensable for humans who recognize text very differently.
This boils down to one thing: config files are source code, and need all the amenities that make source code comfortable for humans.