[analog31]

Granted, I'm just a scientific programmer, but my workplace has a full blown software team maintaining a multi million line codebase. That codebase is rebuilt every night, and as I understand it, you're not allowed to submit a change that breaks anything. And they have people whose job is to keep their tools working.

What people casually think of as "Python" is really a huge dynamic ecosystem of packages. Imagine that there are 60k packages, each with 1k lines of code... that's a 60 million line codebase, and it can't be checked for breaking changes. Short of continually testing your code against the latest versions of packages, you're going to hit some bumps if you haul an old code out of the vault and try to fire it up on a new system.

I don't know how Javascript developers handle this.

I handle it by running Python inside an isolated environment -- WinPython does this for me -- and occasionally having to fix something if a new version of a package causes a breaking change.

The drawback of my method is deployment -- there is no small nugget of code that I can confidently share with someone. They have to install their own environment for running my stuff, or take their chances, which usually ends badly.

[axelf4]

> you're going to hit some bumps if you haul an old code out of the vault and try to fire it up on a new system.

Most package managers have lockfiles that allow for some degree of determinism. Of course if a library introduces breaking changes you're going to have to rewrite, but only when explicitly upgrading the dependency.

[tarsinge]

You should try Dockerizing your project. Then other people just have to type docker run and it’ll work everywhere.

[mmrezaie]

Well in sceintific projects you are working in HPC platforms which docker doesn't exists and there are many reasons for it (it is not just technical).

[yrro]

Singularity may offer a way forward (though I guess you are aware of this already)

[fock]

and nowadays most HPC centers run on spack or easybuild anyways. I adopted easybuild 2 years ago, since then switched to spack and see that our local HPC center is also using it for their new modules...

[mmrezaie]

When I am working on large projects like this and the miryad of problems which python and its friends bring to us, I tend to see if projects like "nuitka" can solve those problems. For example if you need to scale you jobs ust imagine how many sysopen_at syscalls one single python script causes, and usually multiply that to the number of cores your job will have. Nuitka solves most of those problems by packaging everythin. It is almost as staticallly linking your code to run in the cluster.