The main goal is using Julia with Python, and see how well it works. Also, I compare the performance with Python and Nim. The result is quite interesting. Julia is a bit slower than Nim probably due to the JIT compilation. However, it is much faster than Python. I can load my Julia code in Python, and it works well like Nim's exportpy.
Julia can also use many other languages, here, I only tried some quick C. The experience is shockingly good, just use @ccall and it works. It also has no class/interface, but it can use struct and function similar to Rust and C? Good and bad I guess.
Overall, this is a nice language to use side by side with Python. Nim can also be another one to look into, so I will keep an eye on both. Mojo is probably a better one, but it doesn't run on Windows yet. Next time, I may try a bit more on Bun/Zig vs Deno/Rust.
Also see the HaxeExperiment.
Simply run the allinone.py to see the result. Some setup is required, make sure all the dependencies are installed.
python -m venv .env
.env\Scripts\activate
pip install -r requirements.txt
nim c -d:release --app:lib --out:quicknim.pyd --threads:on --tlsEmulation:off --passL:-static quicknim
gcc -o1 -fPIC -shared -o libquick.so quick.c
python allinone.pyYou will see the result like this:
Benchmark everything in one file
Nim pyd time: 2.429507255554199
Julia in python time: 2.5180130004882812
C: Time taken: 0.904000
Julia @ccall time: 0.9040000438690186
juliacall time: 0.9393115043640137
Notice the overhead with @ccall in Julia and also juliacall in python
Now, let's wait for last benchmark with pure python
Python time: 11.553006410598755
As you can see, Nim is faster than Julia by 0.1s, but Julia in my experience could be a bit easier compared to Nim. I need to put a big question mark here though, but they are far better than Python which ever you choose. C implementation is definitely the fastest, but it is not as easy as Nim or Julia, the source code is also much longer.
See closely, the overhead with @ccall is extremely small, so using C with Julia is quite nice. However, the overhead with juliacall is around 0.04s. Even if we add double overhead, it is still far ahead of the Python's legendary 11.55s, with optimisation it can go down to 7.75s, still more than 3 times slower. Remember that juliacall also converts the data into Python types, and Nim's exportpy does the same. You can use Python in Nim & Julia and also use Nim & Julia in Python, so it is quite flexible here.
Using Nim or Julia side by side with Python is definitely a great choice, not to mention that I have to do quite some data processing in Python, using Julia's built-in DataFrames could give a significant boost in performance especially when I deal with stock data with over 500 files each having 10,000 rows and more. In total, I will have over 3000 files and there can only be more in the future.