Run benchmarks with rtoml v0.11?

[samuelcolvin]

I'd love to see how rtoml v0.11 (specifically samuelcolvin/rtoml#78 where the GIL pool was removed from pyo3) effects performance.

[pwwang]

It's incredibly fast! See updated README.md

[samuelcolvin]

Nice! Any chance you could include the benchmark example from cpptoml like you used to, even if that package has been removed?

Multiple benchmark cases are useful.

[pwwang]

It's removed because there's a case that cannot be loaded by rtoml. I will give more details later.

[pwwang]

I found the culprit. It's the item here that when loaded, can't be dumped again with rtoml:

https://github.com/bobfang1992/pytomlpp/blob/47d705b909deee0fa7a3c65624bfa83ea87f578d/benchmark/data.toml#L431-L437

[terrible.capable-tent.wilderness-aromatic.birds.naughty-blind]
bake-scatter-cross = [
    234,
    { attempt-selfish-ghost = 08:43:45, draconian-ski-scintillating = 1911-03-24T12:01:06-01:20, dreary = 266 },
    'innocent broken equable adhesive spiritual title respect'
] # hard-to-find protective
ticket-enthusiastic-delightful = [ 428, 12:49:26, [ 07:10:40, 'neck unadvised finicky stale rest', true ] ]

This is the error:

[TomlSerializationError] values must be emitted before tables

A simple way to reproduce:

import rtoml

toml = """
[a]
b = [234, { c = 1, d = 2, e = 3}, 'abc']
f = [345, "def", [4, "ghi", 456]
"""

loaded = rtoml.loads(toml)
rtoml.dumps(loaded)

Error:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/~/.cache/pypoetry/virtualenvs/toml-bench-TNBybMrA-py3.12/lib/python3.12/site-packages/rtoml/__init__.py", line 62, in dumps
    return serialize(obj, none_value=none_value)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
_rtoml.TomlSerializationError: values must be emitted before tables

[samuelcolvin]

thanks, I'll look into it.

But that's not new, does that mean they've changed their examples?

[pwwang]

That's wired. The file was created 2 years ago.

Should I refer this to a new issue in rtoml repo? I think it fits better there.

[samuelcolvin]

yes please.