Why do so many tools have JSON config files?(textlog.cc) |
Why do so many tools have JSON config files?(textlog.cc) |
That's the whole post. First I don't know why this is posted on HN. Second I don't see how "JSON config" and "writing things down" are the two opposite options.
The "writing things down bit" is in the context of commenting. I mean, it's literally in the same paragraph.
No, we're just lazy.
I'm in favor of documentation. I write it. I get people pointing at what I've written as examples of what everybody should do. However it is a lot of work. I'm constantly looking things over to be sure it still makes sense. I often wonder if it is really worth it. I hope you follow my example and write documentation, but it better feel like a lot of work.
I don't think that's necessarily lazy, it's just efficient
That's what I am saying about my ruby codes as well. Still my boss wanted proper commit messages
1: Because JSON is a very easy serialization format to work with. I suspect these tools all have configuration classes / objects that are deserialized straight from the config file.
2: I suspect a lot of these tools are written in Javascript, and in Javascript JSON is very easy to work with.
In C#, I find "binary serialization" much easier to program with than JSON.
Then again, the resulting blobs require specialized tooling to read, and they're very hard to work with in other programming languages.
---
But, jokes aside:
I find CSV is great for "rows" because it doesn't repeat field names for every object.
I really like XML when each tag is an object, and fields are attributes. Most people don't understand this and end up making a hug mess; and some serializers do this by default too. IMO, this is why JSON became much more popular.
[1]: https://github.com/edn-format/edn
[2]: https://en.wikipedia.org/wiki/Clojure#Extensible_Data_Notati...
env=staging
db=0.0.0.0
#descriptive comment
etc=true
[section_name]
key=value
key=value
every format/structure that has numbers, strings, booleans, null/none, finite lists of items, and string-indexed records of items already contains JSON
and that's pretty much the barebones you need for a configuration language
(of course you can argue about the syntax)
<whispers>but Lua tables are even better.</whispers>
Shame.
Because XML hasn't been cool for about two decades. And suggesting .ini would you laughed out of the room into retirement.
SERVER=(DESCRIPTION (ADDRESS=(PROTOCOL=TCP)(HOST=MyHost)(PORT=MyPort))(CONNECT_DATA=(SERVICE_NAME=MyOracleSID)));uid=myUsername;pwd=myPassword;They are also apparently very hard to use, to the point that many people can't and several web sites exist that will take your database server, username, and password and assemble a connection string for you. (I can see no issues with that! None!)
It fits into the poor choices we made, <-- Parse error
- `import json` allows read/write.
- easier to use than `import configparser`
- allows writing vis-a-vis `import tomlib`
2. The format is too simple to have ambiguous behavior. No weird “yes” means true, 0 means false, odd rules about comments, blah blah blah. It’s hard to fuck up JSON (but obviously not impossible if you get creative).
3. It errors out early in the parsing if you mess it up.
4. Its data types are present in more or less any language.
5. Most configs are just key/value. JSON does this reasonably well.
6. It is easy to generate and validate JSON documents. For some use cases you don’t need a library (though you should use one).
7. There is only one way to do anything (sane).
8. Everyone is familiar with it.
9. It is dynamic. You do not need to pre-define your sections or keys ahead of time.
10. It can easily be auto formatted to look good with zero risk of changing semantics.
11. Data stores often natively support storing and querying JSON objects.
12. If you are old enough to remember the era when every tool invented its own, often very buggy, config format and parser you will also remember the moment you first saw a JSON config file that was parsed with a standard library parser and thought “finally, this is the modern way”, you will understand why JSON continues being popular. It was the first thing that unambiguously worked compared to what came before it.
This is like asking why people use their keys to open packages: it might not be the right tool for the job but it’s hard to mess up, is the closest thing to you that can get the job done, and everyone (with functioning hands/fingers) can do it with little issue.
I am also certain there is some small but non-zero percentage of people who do it simply because everyone else moralizes about not doing it. Spite is a powerful thing.
ESR had the idea of writing configuration in English. It didn't gain traction at the time, but we have LLMs now. It might be a good idea to revisit the idea of accepting plain english. The LLM output could then be any format that's easy and unambiguous to parse.
Thats the problem isnt it? LLMs arent deterministic. Terrible for prod
Say for instance you have a program that keeps recipes. In your configuration file, there's settings such as metric/imperical units, allergies, diets, type of stove in your kitchen, and some theming (font, size, color...). You put them all in a file that you can parse, but aunt tillie messes up the formatting and the program breaks.
Instead of changing the config by hand, an LLM could supply the diff according to instructions in English. I really don't see why this would be "Terrible for prod". If the LLM screws up, you're simply back to square 1 and aunt tillie will call you just like she would before she had an LLM to fix her computer.
Still a terrible idea for config
However, one annoying thing for TOML was the lack of schema, and the reliance on JSON Schema for that. Which I decided to tackle years ago when I started the TOML Schema project. In the past few months I leveraged code agents to take to the finish line and got something compelling: tomlschema.org
I think this is the real "why do so many tools have JSON config files": because it's just a literal notation, for basic data structures, that looks a lot like what many programming languages natively do, what their data structures natively are. The answer to the post is: it's mechanistic sympathy.
For humans, I do find TOML to be a lot easier to manage. Even if you have a really good editor that takes care of all the quoting/nesting/comma concerns for you, even if you have jsonc or json5 with comments, it's still not as easy/friendly as a big flat file with sections in it.
It should really define sections as something different from dot-separated identifier groups.
1) JSON is pretty darn good for storing configuration. Everything speaks it, and a pretty printed one is very readable/tweakable in a pinch.
2) If you insist that someone manually edit a significant amount of it, you kinda fucked up.
Just my opinion, but it feels like two separate things.
ASN.1 is almost a superset, but lacks a key/value list type; I had made some nonstadard extensions called ASN.1X and one of my new types is a key/value list type, so that makes the data types of ASN.1X a superset of JSON (although the format is different, it makes that all JSON data can be represented using DER if the nonstandard key/value list type of ASN.1X is used).
I don't like JSON that much because of its many problems (some are problems with syntax, others are problems with the data), so I use ASN.1X instead (with the DER format), for my own stuff (but I also deal with JSON because it is common enough).
With JSON, I can format the string as I feel makes sense for that specific data structure, while YAML forces me to into specific indentation patterns, and it still causes me to question every time whether the dashes in arrays should be indented or not. And since people frequently template YAML, the strict indentation has caused a bunch of issues for no good reason, including production outages (sure, shouldn't happen with good practices, but there are so many places without good practices!).
Because JSON syntax generally has only one way to represent each data type and almost every data type uses explicit start and end signifiers (except floats and bools, both of which are short and clearly stand out), it's easy to know what the context of every character is both while reading linearly, and when jumping to specific points. Meanwhile YAML has multiple ways to express almost anything, and frequently the only way to tell the current context is to read ahead before jumping back. This is especially terrible for strings due to optional delineation, because almost every bare text could be a keyword (as demonstrated by the Norway problem).
I know this is not most people's experience, but that's because readability is subjective.
Given what I know about it (single file, no auth, and such by default) I sort of understand the ”file” part I think. But not the ”config” part.
I've shipped a product that used SQLite, and we actively removed configuration from SQLite. It was a lot easier to diagnose issues when configuration was in text files, because non-programmers could kinda-sorta understand them without needing to learn how to use SQL.
ASN.1X is mostly just a list of additional types, although there is also another serialization format called SDER which is between BER and DER (any valid DER is also valid SDER and any valid SDER is also valid BER), and is intended for when you do not quite need a canonical form but still want the simplicity of DER; one of the things that it allows is overlong length encodings (which is useful when the encoder wants to encode items to a file individually but then go back to encode the length afterward).
The additional types include:
- UTF-16 string: Same as BMP string but non-BMP characters are also allowed (as surrogate pairs). The type number is the same as BMP string.
- OBJECT IDENTIFIER RELATIVE TO: Either a absolute or a relative object identifier; what it is relative to might be either fixed or given elsewhere in the data, depending on the schema. This is equivalent to a type given in the appendix of the official specification of ASN.1, except that it is now a standardized type, and the canonical form (even in SDER, so that a reader does not need to check for both cases) is required to use the relative format if possible.
- BCD string (64): A string of 4-bit characters, with the high nybble first in each byte. The characters come from the character set 0 1 2 3 4 5 6 7 8 9 * # + - . space and it should be padded with a space on the end if necessary.
- PC string (65): A string of characters in the PC character set (or a related character set in some cases). Note that control characters can also be used as graphic characters.
- TRON string (66): A string of TRON characters, encoded as TRON-8.
- Key/value list (67): A set of keys (without duplicates) with associated values. The keys and values can be any type allowed by the schema. In canonical form, they must be sorted by keys in the same order that a SET is sorted (but the values are kept with the corresponding keys). (This is the only one of these nonstandard types which is used for representing JSON data; all of the other JSON types correspond to standard ASN.1 types.)
- Out of band (72): The format and usage of this type depends on the communication channel being used, and is intended for including things inside of the ASN.1X data which is separate from the ASN.1X data, such as file descriptors. This type is not intended for storage in files, and some programs that relay messages may need special handling of this type if it is used (for this reason, it should not be IMPLICIT).
- Reference (74): A reference to another node within the same file (schemas may restrict which nodes can be referenced). The encoding is like a relative OID but the first number is how many times to go to the parent node (0 means the reference itself), and then the rest of the numbers are the zero-based index into the node referenced by the previous number.
- Identified data (75): Contains a set, followed by the payload (of any type), followed by an optional key/value list where the keys are OIDs. The set is used to identify the format, and it can contain OIDs, object descriptors (only expected to be used in error messages and stuff like that), and sequences who first element is a OID; and should not have duplicates. This type may be used as the top-level type in a file in order to identify the file format, but can also be used inside of the file in case the existing types are considered to be insufficient for this purpose.
- Rational number (76): Contains two integers, being the numerator and denominator. The denominator must be greater than zero. If it is canonical form, then it must be lowest terms.
- Translation list (77): A key/value list where the keys specify the languages (null means the default), and is expected to be used where it could be replaced by the appropriate value according to the l10n.
- Scientific number (78): Same as a real number except that the number of digits (or bits) is considered to be significant. Decimal numbers must be NR3, and if it is canonical form then there must be exactly one digit before the dot.
There are also additional situations where the standard ASN.1 schema format does not seem to specify (although as I mentioned above, there is currently no standardized schema format for these things), such as: regular expressions for octet strings (and bit strings), constraining types as though it is another type (e.g. limiting a UTF-8 string to a number of bytes instead of (or in addition to) code points), constraints about what control characters are allowed in a General string (I think it is rarely useful to allow all control characters), etc. (Also, I disagree with the standard recommendation to use automatic tags; I think that manual tags are better and make both the schema and the data files clearer and easier to understand.)
This is not reintroducing anything; the BMP type is already there, although its meaning is expanded (the existing BMP type is effectively a constrained subtype of UTF-16). Although most applications probably will not use UTF-16, it might sometimes be useful in some applications where it is more useful to store UTF-16 instead of converting to/from UTF-8.
> relative OIDs already exist
Yes, although I have given a standardized name and semantics to something that is allegedly already a common use (and is one that I often use in my own projects), which the official specification from ITU admits. It uses the existing OID and relative OID types (and the same type numbers as them), and is like a CHOICE between them (implementations may treat it as such).
> BCD strings are just constrained PrintableStrings
The abstract meaning matches that of constrained Visible (not Printable) strings, but the encoding is more compact.
> UTF-8 won for all of the string types
Although it is common (and some other formats don't support other string types), I disagree, and I think that one character set cannot be useful for all purposes, and furthermore that Unicode is not that good and has many problems.
> Reference sounds like an EXTERNAL, OOB sounds like an ANY DEFINED BY
I don't think so. It seem like different to me.
I obviously disagree about the ease of json editing. jq tells me errors, sure. But formats where we don't need bespoke tooling, where notepad.exe work fine, are I think probably what config files should be more like.