I know everyone is talking about defragging but the Windows registry is still a piece of junk core component on Windows we have to deal with, unlike NTFS getting away with being on an SSD.
I think that fundamentally, a system-wide configuration database isn’t necessarily a bad thing, eapecially with types. It sounds saner than using text with its pitfalls, especially if the format itself sucks (YAML and XML to an extent)
I would agree that JSON is indeed one of the lesser evils, i definitely prefer it over YAML and XML (yes, the latter has schemas…) though there are some things I dislike about it for configuration. Which is fair because it’s a serialization language. But man I could really do without string quoting keys and the no trailing commas (again, fine for serialization), and sometimes I could really use proper comments or references.
There’s a reason there are so many extensions. JSONC,
JSON Schema, JSON References, JSON Pointer, JSON5, J8, all because people want to use it for their purpose… it’s a bit of bastardization, but still better than YAML.
For most user facing configuration purposes, TOML is probably the better choice. But again, I think ultimately the central database approach is better. And while a filesystem is a database, the logic is broken once you put keys into the files.
I know everyone is talking about defragging but the Windows registry is still a piece of junk core component on Windows we have to deal with, unlike NTFS getting away with being on an SSD.
regedit HKLM my ass
I think that fundamentally, a system-wide configuration database isn’t necessarily a bad thing, eapecially with types. It sounds saner than using text with its pitfalls, especially if the format itself sucks (YAML and XML to an extent)
JSON all the things!
I would agree that JSON is indeed one of the lesser evils, i definitely prefer it over YAML and XML (yes, the latter has schemas…) though there are some things I dislike about it for configuration. Which is fair because it’s a serialization language. But man I could really do without string quoting keys and the no trailing commas (again, fine for serialization), and sometimes I could really use proper comments or references.
There’s a reason there are so many extensions. JSONC, JSON Schema, JSON References, JSON Pointer, JSON5, J8, all because people want to use it for their purpose… it’s a bit of bastardization, but still better than YAML.
For most user facing configuration purposes, TOML is probably the better choice. But again, I think ultimately the central database approach is better. And while a filesystem is a database, the logic is broken once you put keys into the files.