They only dedup runtimes, not individual dependencies.

Not sure I follow you entirely, but I think we agree.

No, not quite. Flatpak is containers - it just stuffs every dependency that an application needs in a directory with no way to deduplicate or update independently. Gobo is a bit more nuanced, since dependencies are shared between applications when the versions match.

I think the main premise is that every version of every software has its own installation prefix. This allows you to mix&match different versions, perform atomic upgrades, etc. You can think of it as a proto-Nix. TBH I don’t see much point in it now that Nix(OS) and Guix exist, or, if you don’t like their purity, stal/IX.

The article is very light on details, but the numbers don’t seem to check out at all. Back-of-the napkin math (assuming a square 1km × 1km solar array and total sun luminosity of 3.83e26 W):

1 km ^ 2 * (3.83e26 W) / (4 * π * (1 AU) ^ 2) * 1 year to TWh ≈ 11.94 TW·h

This is a “measly” 12 TWh of TOTAL energy delivered to the array over a year - not accounting for solar panel efficiency losses (20-24%) or the elephant in the room of transmitting this energy back to earth. For context, China alone consumed around 39 PWh (39000 TWh) of energy from fossil fuels just over the course of one year, 2023. The entire world consumed 55 PWh (55000 TWh) of oil energy in 2023 alone. It’s not even comparable to the annual consumption of oil. If we consider the aforementioned factors, assuming 24% solar panel efficiency and an extremely generous 50% power transmission efficiency, we get:

1 km ^ 2 * (3.83e26 W) / (4 * π * (1 AU) ^ 2) * 24% * 50% ≈ 163.43 MW

Which is literally nothing on a national scale - it’s less than a percent of the Three Gorges Dam output.