Repository navigation
pre-allocate for Set() new component for lower memory usage #107
Description
Activity
Yes, it is a good idea!
I also think in terms of implementing multiple key set which will lead us to full marshaling implementation. When we set multiple keys, it make the first iteration, and find all keys first, and estimate how big new buffer should be, and on second iteration fills it with data.
Actually I dug into source and realized the bytes.Buffer provides a 64 byte initial slice for writing into. Given that, I think it's a tossup whether the complexity is worth it to pre-allocate.
- added a commit that references this issue
on Jul 28, 2026 Fixed in v1.4.0 — Set now pre-allocates the output buffer using calcAllocateSpace, reducing allocations from 6 to 1 per call. See: https://github.com/buger/jsonparser/releases/tag/v1.4.0
Extra allocations are done when using an empty
bytes.Bufferhere increateInsertComponents().bytes.Bufferuses a normal grow strategy of successively doubling in size. This could be avoided by counting the length of the needed keys and the final set value and starting with a Buffer that is about the right size.