3 ms·

>I feel like I'm being punished by the stupid people who make me use it What's the use-case and what alternative would you prefer?

crabbone 3y ago
There are multiple contradictory requirements to different things you could want from communication formats. Below are some examples: * You could want to have a universal tool that can examine and understand the contents of the message (for debugging purposes), but you could also want not to send meta-information about the message (s.a. types or sizes etc.) that is essential for parsing the message. And you cannot have both at the same time. * You could want a message format that maps onto the primitive types of a particular language well, but at the same time you may want it to be universal and map to many other languages well. But this is impossible because different languages will have different primitive types and the need to be generic will act against the need of being specific. * You may want to be able to stream data, but this works against hierarchic data organization. * You may want to be able to write messages into pre-allocated memory buffers w/o having to re-calculate the amount of memory necessary to encode a message, but this makes it very hard / impossible to add custom fields and types. --- Given all this, I don't think that JSON is a good match for any use-case it's currently commonly used for. If I want data transfer, I'd go for something like SQL. If I want configuration, I'd go for Datalog. But then I see value in optimizing the transfer of multiple similar records, whereas someone else may see value in optimizing transfer of hierarchically structured data, which isn't necessary repeating. I tried many formats of this kind, and am yet to find a good one. I'm inclined to think that maybe trying to arrange hierarchical data with different constraints on its organization is just a bad approach to data transfer, that the organization and constraints of such data shouldn't be encoded as part of the format, but interpreted by the users of the format. But, if I really had to do this the best way I can image, I'd still go for Datalog.

crabbone 3y ago

There are multiple contradictory requirements to different things you could want from communication formats. Below are some examples: * You could want to have a universal tool that can examine and understand the contents of the message (for debugging purposes), but you could also want not to send meta-information about the message (s.a. types or sizes etc.) that is essential for parsing the message. And you cannot have both at the same time. * You could want a message format that maps onto the primitive types of a particular language well, but at the same time you may want it to be universal and map to many other languages well. But this is impossible because different languages will have different primitive types and the need to be generic will act against the need of being specific. * You may want to be able to stream data, but this works against hierarchic data organization. * You may want to be able to write messages into pre-allocated memory buffers w/o having to re-calculate the amount of memory necessary to encode a message, but this makes it very hard / impossible to add custom fields and types. --- Given all this, I don't think that JSON is a good match for any use-case it's currently commonly used for. If I want data transfer, I'd go for something like SQL. If I want configuration, I'd go for Datalog. But then I see value in optimizing the transfer of multiple similar records, whereas someone else may see value in optimizing transfer of hierarchically structured data, which isn't necessary repeating. I tried many formats of this kind, and am yet to find a good one. I'm inclined to think that maybe trying to arrange hierarchical data with different constraints on its organization is just a bad approach to data transfer, that the organization and constraints of such data shouldn't be encoded as part of the format, but interpreted by the users of the format. But, if I really had to do this the best way I can image, I'd still go for Datalog.