4 ms·

It's still a mess, honestly. They tried to standardize on venv but since it doesn't deal well with Python libraries that depend on non-Python components, conda

by lacker 5y ago

It's still a mess, honestly. They tried to standardize on venv but since it doesn't deal well with Python libraries that depend on non-Python components, conda is still better for many purposes.

anakaine 5y ago
Except when Conda plays hard to get, like in any corporate environment with connection inspection, custom root CAs, or not quite admin privileges. Then Conda is an absolute dog and hard to work with, yet regular old python is fine.
isoprophlex 5y ago
Can you give a specific example? I've had nothing but pain with conda in mixed linux/mac/win environment at work, and actively worked to get it deprecated. We're on plain venvs now, with the occasional setup.py Curious if I missed some useful case for it. - very slow dependency resolvers - using conda in docker is annoying - the worst thing: inconsistencies in downloading binaries and other resources between win/mac/linux when installing packages - conda insisting on messing with bashrc to run their 'activate' crap - beware the fool who runs 'sudo conda whatever' on a multiuser system
- dagw 5y ago
  Curious if I missed some useful case for it. Conda will package and version things like compilers and C libraries etc in a sane way. I can 'conda install' a package and start using it, but if i 'pip install' the same package I'm still left having to hunt down and install a bunch of additional dependencies. For example 'conda install cython' gives me a fully working and consistent cython on all platforms in a way that 'pip install cython' won't. There are still several computational science and data science packages where 'pip install X' on windows simply doesn't work (and 'conda install X' does) due to missing dependencies and other weirdness. Conda can version and manage R as well as Python if you work on projects that use both then you only need one tool. Conda can also install dev tools like Spyder, VSCode or Jupyter(labs) making it easier to offer a single point of entry for an entire dev environment. But basically the big one is simply that Conda will install a lot more numeric packages with a lot fewer headaches on win, linux and mac than pip. Although in pip's defense it has gotten a lot better over the past 2-3 years. And I do agree that the dependency resolver can take a ridiculously long time in some situations.
  - verall 5y ago
    I think this is an interesting case where people just want different things, and so the thing that they want is "better". Some people really just want Python+extras and don't want to worry about getting a fortran compiler. Conda is great for this. Other people really don't want a big meta-install system that bundles a bunch of dependencies. Maybe they're on Linux, they target a single platform, or they want to use a package not in conda that works with a more recent version of the dependencies that conda bundles.
    - dagw 5y ago
      I think this is an interesting case where people just want different things For some people Python is really more of an end user application. They don't want to develop and distribute and deploy software, they want to write scripts and analyze and transform data. They expect the script a colleague emailed them to just work and expect their colleague to be able to open and run their Jupyter notebook as easily as they can open and run an Excel document. For those people Conda comes much closer to solving their problems.
- lacker 5y ago
  Here's an example, let's say you're developing GPU code that needs to work with different CUDA versions on different machines. Look at the `cupy` install instructions for example: https://docs.cupy.dev/en/stable/install.html https://docs.cupy.dev/en/stable/install.html With conda this installation tends to just work, conda will also install the non-python cudatoolkit for you. With pip you have to either make all your python developers set up their c++ environment the same way as well, to install from source, or set a fixed cuda version that all users have to be on. Now `cupy` is just one python library that has non-python dependencies. If your project has several dependencies like this, where conda is a one line install and pip means you have to mess around with your c++ environment, conda is probably the right choice for you. All of the pain you mention with conda is totally correct, though, it's just a question of which sort of pain happens to be worse for the packages you are going to use.

by lacker 5y ago

It's still a mess, honestly. They tried to standardize on venv but since it doesn't deal well with Python libraries that depend on non-Python components, conda is still better for many purposes.

anakaine 5y ago

Except when Conda plays hard to get, like in any corporate environment with connection inspection, custom root CAs, or not quite admin privileges. Then Conda is an absolute dog and hard to work with, yet regular old python is fine.

isoprophlex 5y ago

Can you give a specific example? I've had nothing but pain with conda in mixed linux/mac/win environment at work, and actively worked to get it deprecated. We're on plain venvs now, with the occasional setup.py Curious if I missed some useful case for it. - very slow dependency resolvers - using conda in docker is annoying - the worst thing: inconsistencies in downloading binaries and other resources between win/mac/linux when installing packages - conda insisting on messing with bashrc to run their 'activate' crap - beware the fool who runs 'sudo conda whatever' on a multiuser system

dagw 5y ago
Curious if I missed some useful case for it. Conda will package and version things like compilers and C libraries etc in a sane way. I can 'conda install' a package and start using it, but if i 'pip install' the same package I'm still left having to hunt down and install a bunch of additional dependencies. For example 'conda install cython' gives me a fully working and consistent cython on all platforms in a way that 'pip install cython' won't. There are still several computational science and data science packages where 'pip install X' on windows simply doesn't work (and 'conda install X' does) due to missing dependencies and other weirdness. Conda can version and manage R as well as Python if you work on projects that use both then you only need one tool. Conda can also install dev tools like Spyder, VSCode or Jupyter(labs) making it easier to offer a single point of entry for an entire dev environment. But basically the big one is simply that Conda will install a lot more numeric packages with a lot fewer headaches on win, linux and mac than pip. Although in pip's defense it has gotten a lot better over the past 2-3 years. And I do agree that the dependency resolver can take a ridiculously long time in some situations.
- verall 5y ago
  I think this is an interesting case where people just want different things, and so the thing that they want is "better". Some people really just want Python+extras and don't want to worry about getting a fortran compiler. Conda is great for this. Other people really don't want a big meta-install system that bundles a bunch of dependencies. Maybe they're on Linux, they target a single platform, or they want to use a package not in conda that works with a more recent version of the dependencies that conda bundles.
  - dagw 5y ago
    I think this is an interesting case where people just want different things For some people Python is really more of an end user application. They don't want to develop and distribute and deploy software, they want to write scripts and analyze and transform data. They expect the script a colleague emailed them to just work and expect their colleague to be able to open and run their Jupyter notebook as easily as they can open and run an Excel document. For those people Conda comes much closer to solving their problems.
lacker 5y ago
Here's an example, let's say you're developing GPU code that needs to work with different CUDA versions on different machines. Look at the `cupy` install instructions for example: https://docs.cupy.dev/en/stable/install.html https://docs.cupy.dev/en/stable/install.html With conda this installation tends to just work, conda will also install the non-python cudatoolkit for you. With pip you have to either make all your python developers set up their c++ environment the same way as well, to install from source, or set a fixed cuda version that all users have to be on. Now `cupy` is just one python library that has non-python dependencies. If your project has several dependencies like this, where conda is a one line install and pip means you have to mess around with your c++ environment, conda is probably the right choice for you. All of the pain you mention with conda is totally correct, though, it's just a question of which sort of pain happens to be worse for the packages you are going to use.