Mirror of forum.nim-lang.org

13394 :: Cborious: Fast standards-compliant CBOR library for Nim 2.x

• elcritch (orginal) [2025-09-12T06:29:01+02:00] view original

  Announcing Cborious for Nim 2.x!

  • What: Fast, standards-compliant CBOR library for Nim with streaming I/O, canonical/deterministic modes (RFC 8949), and compile-time encode/decode derivation inspired by msgpack4nim.
  • Highlights: Zero-dependency Streaming reader/writer; canonical/deterministic encoding; object-as-array or map modes; liberal decode for interoperability; optional self-describe tag; easy custom tags.
  • Streaming: Uses std/streams interface for compatibility
  • Performant: Based on the design of msgpack4nim msgpack library for high performance
  • Install: atlas use https://github.com/elcritch/cborious or nimble install https://github.com/elcritch/cborious
  • Quick use:
    • toCbor / fromCbor helpers for simple types and objects.
    • CborStream for incremental pack/unpack or reusing a stream.
    • cborPack / cborUnpack for packing specific types; override to provide custom serde for a type

  Example:

  
  import cborious
  
  type Msg = object
    greeting: string
    value: int
  
  let enc = toCbor(Msg(greeting: "hi", value: 42))
  let dec = fromCbor(enc, Msg)
  Run

  • Docs/Tests: nim docs, nim test
  • Link: https://github.com/elcritch/cborious

  Feedback welcome—API ergonomics, tag coverage, and real‑world use cases especially!

  Performance

  The performance seems pretty decent:

  
  # nim c -d:release -r tests/bench_cbor.nim
  Benchmarking with iters=10000
  cborious:           one-shot size=41 bytes repr=@[164, 98, 105, 100, 24, 42, 100, 110, 97, 109, 101, 104, 78, 105, 109, 32, 85, 115, 101, 114, 102, 97, 99, 116, 105, 118, 101, 245, 102, 115, 99, 111, 114, 101, 115, 133, 1, 2, 3, 5, 8]
  cbor_serialization: one-shot size=41 bytes repr=@[164, 98, 105, 100, 24, 42, 100, 110, 97, 109, 101, 104, 78, 105, 109, 32, 85, 115, 101, 114, 102, 97, 99, 116, 105, 118, 101, 245, 102, 115, 99, 111, 114, 101, 115, 133, 1, 2, 3, 5, 8]
  cbor_em:            one-shot size=41 bytes repr=@[164, 98, 105, 100, 24, 42, 100, 110, 97, 109, 101, 104, 78, 105, 109, 32, 85, 115, 101, 114, 102, 97, 99, 116, 105, 118, 101, 245, 102, 115, 99, 111, 114, 101, 115, 133, 1, 2, 3, 5, 8]
  --- Results (encode + decode round-trip) ---
  cborious:              avg=61100 ns/op total=611 ms
  cbor_serialization:    avg=108131 ns/op total=1081 ms
  cbor_em:               avg=103738 ns/op total=1037 ms
  
  cbor_serialization/cborious: 1.77x for 10000 iterations
  cbor_em/cborious: 1.70x for 10000 iterations
  Run

  
  # nim c -d:release -r "tests/bench_cbor_simple.nim"
  Benchmarking with iters=10000
  cborious:           one-shot size=25 bytes repr=@[135, 24, 42, 24, 100, 245, 24, 100, 251, 64, 5, 191, 9, 149, 170, 247, 144, 250, 64, 73, 14, 86, 130, 1, 2]
  cbor_serialization: one-shot size=88 bytes repr=@[167, 102, 70, 105, 101, 108, 100, 48, 24, 42, 102, 70, 105, 101, 108, 100, 49, 24, 100, 102, 70, 105, 101, 108, 100, 50, 245, 102, 70, 105, 101, 108, 100, 51, 24, 100, 102, 70, 105, 101, 108, 100, 52, 251, 64, 5, 191, 9, 149, 170, 247, 144, 102, 70, 105, 101, 108, 100, 53, 250, 64, 73, 14, 86, 102, 70, 105, 101, 108, 100, 54, 162, 102, 70, 105, 101, 108, 100, 48, 1, 102, 70, 105, 101, 108, 100, 49, 2]
  cbor_em:            one-shot size=25 bytes repr=@[135, 24, 42, 24, 100, 245, 24, 100, 251, 64, 5, 191, 9, 149, 170, 247, 144, 250, 64, 73, 14, 86, 130, 1, 2]
  --- Results (encode + decode round-trip) ---
  cborious:              avg=30962 ns/op total=309 ms
  cbor_serialization:    avg=207246  ns/op total=2072 ms
  cbor_em:               avg=46876  ns/op total=468 ms
  
  cbor_serialization/cborious: 6.69x for 10000 iterations
  cbor_em/cborious: 1.51x for 10000 iterations
  Run

---

• kolsky (orginal) [2025-09-13T03:31:07+02:00] view original

  Dazzling, I remember when I started Rust, I tried to encode some data with cbor, to increase performance.

  Found only discontinued serde_cbor, others were too complicated to use. I understand Rust developers implementing concise code, but you could write two classes of code. One with whatever custom, concise operations you want, and one for simple one-liner operation, like this library implements with toCbor(), fromCbor() 👍🏻👍🏻

---

• elcritch (orginal) [2025-09-13T06:43:33+02:00] view original

  That led me down a bit of a rabbit hole looking into Rust's CBOR world. It looks rough... The top cbor crates are archived, and the new ones look all over the place. All 3 Nim CBOR libs look to be more complete and easier to use. Obviously Cborious is the most complete! It implements self-describing CBOR ;)

  Actually I quite enjoy that a few hours of prompting GPT5 on a side project can be so productive with Nim. Truth be told I was using making Cborious as a way to get over "writers block" in the mornings! Helps me overcome procrastination.

---