if i miss a quote in json, it's an invalid file. I fix it and move on. Maybe just pass it through a beautifier or something.
if I mess up indentation in yaml (which can easily happen after a copy and paste), I get a valid file that means something completely different. And no beautifier can help with that, because it's a valid file. I have to look really closely to find the error, because tools can't help. And when I do, I usually have to change multiple lines, instead of just adding a comma.
XML is a bit verbose but otherwise easy to understand. JSON5 supports comments.
And neither requires me to explain weird formatting nuances to devops engineers.
YAML is a pain to read (lists of structures are very messy), can't be auto-formatted, and is full of weird "gotchas" (Norway, errant tabs, etc.) if you don't do things "the right way."
Requiring the use of whitespace in formatting is wrong. End of.
I instant exit on XML conf files. I have no idea how to parse the info, maybe I should. But there are too many tags and my eyes instantly glaze over when I see it!
Agreed about INI for simple stuff. Not good for arrays and nested things though. Usually use binary for that type of config (with clear documentation). Most binary config files I use are plain old C structures. I'm not a web person so no need to make the config plain text.
JSON if it also has an online schema, and a reference to it in the JSON file. That way some editors can check for errors and hint to the user about it.
But the good old flat text-data, the ini format is consistent, readable, and easily understandable by less-technical users.
Using whitespace was a mistake. I know all the kids love it but it's just wrong. I've never had as much difficulty explaining json or xml to users as I have yaml.
We seem to have something in common: there is a serialisation form, we strongly dislike. But what I cannot understand is: why the heck would anyone torture anyone else to read or even write XML? XML is the absolutely worst configuration language I can imagine. I mean: when is something an attribute, when a tag on its own? What is even a list? And don't forget to include a full HTTP URI for the namespace, otherwise the tag is not defined.
By the way: all valid JSON is valid yaml as well. So in theory, you can use yaml as JSON with comments.
If the purpose of the software is to work in closed, offline environment with Mr. NOTECH operating it via "line 5: rotations per minute; line 6: temperature in Fahrenheit" commands, then trust me, it's going to get the job done.
"unbeatable" and "get the job done" are very different claims. If you just have bare unlabeled numbers on "magic lines" in a file - sure you can work with that. But I'd hardly say it's "unbeatable" compared to "anything else" for readability and clarity. It's the type of thing programming classes have students do so they don't have to introduce libraries or other concepts.
It's very likely that Mr. NOTECH would be able to understand "TEMP_F: ###" just as easily as "put the temperature on line 6". And it would likely be MUCH easier for Mr. NOTECH to look at a config file and read back to you what the temperature setting is. Especially if they haven't worked with the software for long. As a bonus it would be easier for Mr. NOTECH to know whether the temperature should be in F or C. Or even Kelvin.
I absolutely am. You're simply overthinking it - the scenario I provided makes plaintext *.txt perfect.
Sure, we can spend the rest of the day inventing scenarios, where Mr. NOTECH will be manipulated remotely by aliens, and what then, but that's an exercise in creativity, not a situation that needs to be taken into account.
Don't go get the guillotine for me here but - HCL is really nice, and allows for a configuration to handle a bunch of cases that the developer may not have forseen at dev time.
I like how elixir approaches it: configs are just elixir script files that have a module for some common conventions. Gives you a lot of power, and no goofy new syntax
That works if you have a small, disciplined team. As the team grows and it becomes harder to enforce standards, config files that are run through the language compiler easily get filled with programming statements that shouldn't be in a config file.
Assuming you can trust whoever wrote the config file, I prefer if it's executable code.
Not really fussed about what language. The point is it's really handy to have simple logic checks like a basic if statement on an environment variable. I also like being able to split my config files up into multiple files - maybe with an environment variable to check which files will be loaded in.
If you want the file to be directly human readable/editable:
If you never need to look at it or edit it manually:
I hate yaml with a passion. for stuff such as ansible it becomes an uneditable mess (one copy and paste away from destroying the whole file)
luckily, yaml is a superset of json, so you could just write json and feed it through a yaml parser and it'll work
if i miss a quote in json, it's an invalid file. I fix it and move on. Maybe just pass it through a beautifier or something.
if I mess up indentation in yaml (which can easily happen after a copy and paste), I get a valid file that means something completely different. And no beautifier can help with that, because it's a valid file. I have to look really closely to find the error, because tools can't help. And when I do, I usually have to change multiple lines, instead of just adding a comma.
IDK what the best is, but I know for certain it’s not YAML.
I don't know which is the best, but yaml is the worst.
Uh no, the worst is a tie between XML and JSON.
XML because the syntax is hard to read and even harder to write, and JSON because you can't do comments. WTF.
XML is a bit verbose but otherwise easy to understand. JSON5 supports comments.
And neither requires me to explain weird formatting nuances to devops engineers.
YAML is a pain to read (lists of structures are very messy), can't be auto-formatted, and is full of weird "gotchas" (Norway, errant tabs, etc.) if you don't do things "the right way."
Requiring the use of whitespace in formatting is wrong. End of.
Somebody has to say it, so I'm taking on the duty:
If whitespace is a problem, you use the wrong editor.
Oh, yeah, nothing like telling a dba they're using the wrong editor when they're trying to configure something.
If your config format requires specific editors you're using the wrong format.
you can write json with comments and pass it through a yaml parser. try it.
I instant exit on XML conf files. I have no idea how to parse the info, maybe I should. But there are too many tags and my eyes instantly glaze over when I see it!
TOML looks good but it's rarely used
TOML does look good. Wow! I havent looked at it at all before.
Makes certain levels of nesting painful
You can say this is a design choice, and you shouldn't ever go too deep in config, and I'd agree, but it is a limiting factor to be aware of
For simple stuff, INI is pretty good.
I must admit I've written stuff that uses a JSON config file, but I might finish implementing YAML instead. Any day now...
Agreed about INI for simple stuff. Not good for arrays and nested things though. Usually use binary for that type of config (with clear documentation). Most binary config files I use are plain old C structures. I'm not a web person so no need to make the config plain text.
You could use TOML. It's pretty much an extension of INI
JSON if it also has an online schema, and a reference to it in the JSON file. That way some editors can check for errors and hint to the user about it.
But the good old flat text-data, the ini format is consistent, readable, and easily understandable by less-technical users.
I like xml, but it isn't the easiest to read.
I hate yaml with a burning passion. At least with JSON you can compress it all down.
A compliant yaml parser will read json, as they're essentially siblings.. so just write your configs in json and they'll work, if you dislike yaml.
Thanks for the tip!
I'm no fan of XML. It looks messy and the schema is difficult to write, imo. XML works best as a markup language.
YAML could have been okay if it were stricter with what they consider strings, and used tabs instead of spaces.
Using whitespace was a mistake. I know all the kids love it but it's just wrong. I've never had as much difficulty explaining json or xml to users as I have yaml.
We seem to have something in common: there is a serialisation form, we strongly dislike. But what I cannot understand is: why the heck would anyone torture anyone else to read or even write XML? XML is the absolutely worst configuration language I can imagine. I mean: when is something an attribute, when a tag on its own? What is even a list? And don't forget to include a full HTTP URI for the namespace, otherwise the tag is not defined.
By the way: all valid JSON is valid yaml as well. So in theory, you can use yaml as JSON with comments.
I mainly work with windows, and powershell works great with xml.
.el (Emacs Lisp), nothing beats using Lisp to configure a Lisp environment.
For my last self-made program, I just wrote a config.py, and call
import configfrom other code files and write my config in Python.That one which I can automate.
I mainly use Ansible or Puppet. So YAML isn't that bad if you fully understand it.
Puppet language sux IMHO.
YAML. This is the way.
Normal text file
Depends on the purpose of the software. In certain specific situations a plaintext *.txt is unbeatable.
An unstructured text document can be "unbeatable" for configuration how?
The "PURPOSE" is the keyword.
If the purpose of the software is to work in closed, offline environment with Mr. NOTECH operating it via "line 5: rotations per minute; line 6: temperature in Fahrenheit" commands, then trust me, it's going to get the job done.
"unbeatable" and "get the job done" are very different claims. If you just have bare unlabeled numbers on "magic lines" in a file - sure you can work with that. But I'd hardly say it's "unbeatable" compared to "anything else" for readability and clarity. It's the type of thing programming classes have students do so they don't have to introduce libraries or other concepts.
It's very likely that Mr. NOTECH would be able to understand "TEMP_F: ###" just as easily as "put the temperature on line 6". And it would likely be MUCH easier for Mr. NOTECH to look at a config file and read back to you what the temperature setting is. Especially if they haven't worked with the software for long. As a bonus it would be easier for Mr. NOTECH to know whether the temperature should be in F or C. Or even Kelvin.
Unbeatabe, then.
I see, you're not being serious.
I absolutely am. You're simply overthinking it - the scenario I provided makes plaintext *.txt perfect.
Sure, we can spend the rest of the day inventing scenarios, where Mr. NOTECH will be manipulated remotely by aliens, and what then, but that's an exercise in creativity, not a situation that needs to be taken into account.
That depends on if a human is reading/writing it, or a program.
https://nestedtext.org/
It’s like yaml but simple, consistent, untyped, and you never need to escape any characters, ever.
Types and validation aren’t going to be great unless they’re in the actual code anyway.
Don't go get the guillotine for me here but - HCL is really nice, and allows for a configuration to handle a bunch of cases that the developer may not have forseen at dev time.
I like how elixir approaches it: configs are just elixir script files that have a module for some common conventions. Gives you a lot of power, and no goofy new syntax
That works if you have a small, disciplined team. As the team grows and it becomes harder to enforce standards, config files that are run through the language compiler easily get filled with programming statements that shouldn't be in a config file.
Assuming you can trust whoever wrote the config file, I prefer if it's executable code.
Not really fussed about what language. The point is it's really handy to have simple logic checks like a basic if statement on an environment variable. I also like being able to split my config files up into multiple files - maybe with an environment variable to check which files will be loaded in.