This is a personal blog than a technical one.
It was a lovely Sunday night. Someone on a programming Discord posted an article about a Python to C++ transpiler. I knew this sentiment would come up.
"Still, Python is way slower because of the massive overhead of being an interpreted language..."
Anyone in the tech space has probably heard and read that Python is slow. It is. This picture below should paint a good enough picture of that.
In the picture, also look at Node JS. JavaScript isn't compiled, but can toe-to-toe with speeds compared to compiled languages. This picture is a slide from Anthony Shaw's Restarting Pyjion, a general purpose JIT for Python- is it worth it?
Anthony Shaw is a Microsoft employee and author of CPython Internals. He has also given another talk Why is Python Slow?. I would think he's someone who knows what he's talking about.
On both talks, Anthony has given a good introduction to how Python runs code under the hood. tl;dw: Your code gets converted through multiple steps all the way down to bytecode, which is then continually evaluated by the Python interpreter.
The compilation step to bytecode, which is done ahead-of-time (AOT), could be made significantly faster if it would have been done just-in-time (JIT) instead. In fact, projects such as PyPy (with JIT) and Pyjion use JIT compilation.
Python's performance problems come from it being AOT compiled and less from it being interpreted.
- me
That's what I said. Verbatim. And it looks like I rattled some nest.
"Compiled?"
"Compiled?"
"Compiled???"
And especially this guy.
Compiled?
Python is compiled?
If Python is compiled, then boy... do I have a bridge for you.
That was indeed, an invitation for engagement. So I replied,
Language -> AST -> CFG -> bytecode -> ceval.c
.pyc files are basically compiled bytecode
But go ahead and sell me that bridge.
I don't think he has sold me that bridge, especially if his reply consist of him saying that his "personal definition" of compilation refers to compilation to machine-level compilation. Arguing with a stranger's "personal definitions" over the internetz would be a really productive way to spend a Sunday night, wouldn't it? Improving Python's performance was the topic I really hoped to have after all.
I would just leave here a snippet from the sidebar of the Wikipedia article on Ahead-of-Time Compilation.
Notable Runtimes
- Android Runtime (ART)
- Common Language Runtime (CLR) and Mono
- crt0
- Java virtual machine (JVM)
- Objective-C and Swift
- V8 and Node.js
- CPython and PyPy
- Zend Engine (PHP)
- LuaJIT
If anything, I believe now more than ever that finding a specialist on a programming language do have value.
Top comments (0)