Package manager for plugins in flatpak-installed CloudCompare

[dokempf]

I am currently diving into what it would take to make my project available as a Python plugin to CloudCompare. One thing I have stumbled over is the interplay of flatpak-bundled CloudCompare with this plugins package manager. flatpak apps run inside a sandbox where /app is typically read-only. This disables the package manager and the banner rightfully tells us so. In other scenarios, I have circumvented this successfully (e.g. installing into a user directory on Windows), but I do not find a way out of the flatpak sandbox. Am I missing something? Is there a flatpak setting to make the filesystem writeable? Is there a way to control the location of site-packages within the sandbox to a writeable one like e.g. XDG data? Any help is appreciated.

(I am aware of the fact that I can build CC from source to achieve this. But this is neither a good solution for my users, nor do I want to have this as my standard dev workflow for a small plugin)

[tmontaigu]

The python runtime settings allows to add path where it should look for plugins, so the easiest would be to add a path where the flatpak has read access / give read access to it

[dokempf]

How would that work if my plugin introduces a dependency? I would still need to be able to install that into the bundled interpreter somehow? I previously tried to make this work with an interpreter that lives outside the sandbox, but then the pycc and cccorelib packages are missing...

[tmontaigu]

maybe by setting also a PYTHONPATH you could point to the site-packages of a custom venv that has the dependency installed

A cleaner thing would be to add to the plugin the ability to create a virtual env using a menu, and make it create it in a path where the flatpak has read+write access, either a flatpak specific path like ~/.var/app/$FLATPAK_ID or a 'normal' path

In either case it might require to set some permissions on the flatpak using flatseal for example

I'll take a look

[tmontaigu]

Technically, I think that if you create a venv, with a python version that matches the one embeded in the flatpak, in a local dir, install your stuff in it, then allow the flatpak readwrite access to that path, then tell the the runtime to use it (via settings menu) it should/could work

[tmontaigu]

I tried and got and error, but I dont think it's flatpak related, it's more that venv are a bit of a nightmare to handle, but once fixed that should be a solution

[tmontaigu]

How would that work if my plugin introduces a dependency? I would still need to be able to install that into the bundled interpreter somehow? I previously tried to make this work with an interpreter that lives outside the sandbox, but then the pycc and cccorelib packages are missing...

If you were able to start a script and the error is about missing imports, then the problem is likely that numpy is not installed in the environment