Hacker News new | past | comments | ask | show | jobs | submit login

Alternately, take out the footguns.

I saw a team of data scientists brought to their knees by

   pip --user install some-package
because these packages would infect all of their venv's, including anaconda.

When they were just screwing around with Jupyter notebooks to make reports they did OK but when they were trying to create a factory that produced models that could be transplanted into production systems it was a constant struggle and we discovered so many ways you could screw up the configuration of a Python system such as having the output configured to UTF-8 and throwing errors when an invalid character is printed.

Engineering would say "just don't print uncontrolled text" but we had nearly 100 dependencies and some of them would print.

Engineering would say "just use Docker" but we found plenty of times people used Docker images that had broken Pythons, just now there is another layer of indirection to make problems harder to debug.

Then there are all the little things like

   0.1 + 0.2 - 0.3 = 5.551115123125783e-17
The route to reliable quality is to make it easier to do things right than to do things wrong.



Managing python environments makes me want to cry. I totally understand how that happens, I wish there was an easy way to manage it




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: