Is it just me or is Json.decodeFromBufferedSource ...

# serialization

[ephemient]

Is it just me or is Json.decodeFromBufferedSource much slower than Json.decodeFromStream or Json.decodeFromString?

[ephemient]

I have a use case in my application where I'm serializing/deserializing JSON with compression applied, and thought the new kotlinx-serialization-json-okio artifact would be a nice simplification. but the performance delta was surprising

[jw]

What is your source data?

[jw]

If you're testing on an in-memory string then it absolutely will be slower

[jw]

Okio has to UTF-8 encode the string to bytes (hopefully not part of your test measurement) and then it has to decode those bytes back into character data (definitely part of your measurement)

[jw]

Not to mention all the overhead of shuffling of Segments and position math internally to track the stream.

[jw]

In comparison to a simple string it absolutely will be slower. A string is the ideal representation on which to decode JSON

[jw]

Okio is for when you data source is slow enough that you can amortize the cost of JSON deserializaton over it (making it virtually free) and/or for when you cannot fit the raw string contents in memory.

[ephemient]

the source is a okio.ByteString - sometimes from a database and sometimes in memory

[ephemient]

I do expect some difference to String, but not 2.5-4x like I'm seeing

[jw]

In both of those cases a string is a much better source for kotlinx.serialization