Compilation speed

[notramo]

A simple, 3 line "hello world" program in Pony is taking ~5 seconds to compile. And every time I compile the same code, it's 5 seconds. On the same machine, Zig compiles a similar "hello world" program in <1s with cold cache, and in milliseconds with incremental compilation, and a Roc "hello world" program starts in <50ms with roc run.

Zig uses incremental compilation. Roc uses a dev backend with hot reloading, and cold compilation is very fast too. Hylo will probably have an interpreter for instant feedback.

Are there any plans for <1s feedback with Pony, whatever mechanism it would be? The goal is to be able to run my code <1s after saving it.

[SeanTAllen]

There is no goal for less than 1 second compilation. There is a goal to make compilation as fast as possible while providing the various functionality and static guarantees that we want. If pure compilation speed is your goal for a language of your choice then Pony will also assuredly not meet what you need. Go for example, is a language where compilation speed has always been a motivating goal and they have decided to not include features due to the impact on compilation times. Pony aims first for a robust set of static guarantees and other language features. While we consider compilation speed important, we do not place it at as high of a level of other languages in terms of evaluating trade-offs.

I would suggest using the --debug and --thin-lto for the fastest compilation times. The defaults that the compiler ships with (modulo one thing) are aimed at the default creating binaries that will have the fastest generated code. Both --debug and --thin-lto trade runtime performance for compilation speed.

[notramo]

Go is a very bad example for this. Zig was not designed with compilation speed as a primary goal. It was just added later, because good language design allowed for incremental compilation. Hylo is also a language with no "fast compilation" in the list of design principles. It will have an interpreter for instant development feedback, because compilation will likely remain slow.

I'm not saying that trading some language features for compilation speed would be good. I'm asking whether Pony architecture would allow for fast feedback loop during development. Maybe incremental compilation could cache enough things. Zig can cache very efficiently, however, in the case of Crystal, it's not possible, because the types of variables inside functions depend on every call site, and changing one part of a program might require recompiling many other parts. Is Pony similar to Zig or Crystal in this aspect?

[notramo]

An interesting feature in Roc is that it can run incorrect and partially written code in development mode (roc run), to allow running the program and let it crash, because it does not matter during development. But the checker displays the errors, and the build fails, so errors do not slip into compiled programs.

[redvers]

Closer to Crystal, as Pony has to evaluate the whole program for consistency.

Pony's philosophy is that correctness > all.

I personally can't imagine a scenario where incorrect and partially written code being allowed to run is a positive - even, especially during development.

[notramo]

Closer to Crystal, as Pony has to evaluate the whole program for consistency.

Is some of these computations/checks cacheable or incrementally checkable?

I personally can't imagine a scenario where incorrect and partially written code being allowed to run is a positive - even, especially during development.

This depends on the language design. It's acceptable for Roc, because it's a FP language, so the parts of the program are very well isolated. Based on your description of Pony, it would probably not be a good fit.