Has anybody been able to build a statically linked binary that shows a Vulkan surface? I’ve put some context around this problem in the video. I understand that the vulkan driver has to be loaded dynamically - so it’s more of a question whether a statically built app can reliably load and talk with it. I think it should be possible but haven’t actually seen anyone make it work. I’m aware of “static-window9” by Andrew Kelley but sadly it doesn’t work any more (at least on my Gentoo machine T_T).
(I’m also aware of AppImages but I don’t think they’re the “proper” solution to this problem - more like a temporary bandaid - better than Docker but still far from perfect)
Can’t see why you dislike appimages, sure not 100% size efficient - but for one off binaries youre probably not spending much time optimizing anyways.
Not that you couldnt make an appimage 2.0 solving all your issues, but we’d just be back to that package manager xkcd all over again
Oh, I don’t dislike them. I love them actually. I just wish that vulkan drivers loved them as much as I do :P
Has anybody been able to build a statically linked binary
The question should by why you’d want to. Careful if your reply is something about ‘one binary to work on a very diverse arrangement of library pinnings’ because the next question would be ‘why would you think that’s either achievable or valuable as a goal’; and toss in a ‘why try to ship the same binary in several different repos anyway’ bonus question.
In short, if your biggest problem is how to build a binary that works everywhere, you have a lot of questions about responsible build/release processes to answer, and they will be embarrassing for you.
I think of demoscene, game jams, desktop pets and other sorts of small home-brew software written by & for small groups of people. There is nothing embarrassing about one-off programs. I understand that what you said is just a figure of speech and you don’t really think so (and thank you for a well thought comment) but it still makes me kind of sad to see this kind of shaming. It’s discouraging people from being creative.
I’d say my bar is already low. A toy GUI program that draws a triangle and plays a sound should work on my friend’s machine (with a different set of system libraries) and it shouldn’t randomly break when OS update bumps some library version. AFAIK something like that would require a static binary (INTERP is a no-go) and an ability to dlopen stuff at runtime.
Writing software that “just works” shouldn’t be as hard as it is.
Sounds like flatpaks/appimages with extra steps
Includes all dependencies? ✔️
A single file? ✔️
Independent of host libraries? ✔️
Limited learning curve? ✔️
Not sure how appimages handle it internally, but with flatpaks you can even be storage efficient with layers, whereas 100s of static binaries could contain an awful lot of duplicates.
You know it’s bad when Linux YouTubers are arguing against Linux ports because Proton is just so much more functional for Linux gamers.
I mean, why support another API if you have already implemented one, right? Pretty sure there is a Flatpak Version of Proton, wo you can somehow bundle it as well
I’ll also add that I’m aware of glibc’s stance on dynamic linking from static binaries. I don’t buy the whole NSS argument. It’s easily solvable by a basic request/response protocol through some local socket. IMO that argument is just a cheap excuse to justify status quo.
I have no experience with this, but I figured a Rust library might have tried to solve it (static linking is very much the norm here) and I found that
ashcan statically link the “Vulkan loader”. I don’t know, what that actually means, for example whether it would still loadlibxcbat runtime. Might be worth looking into what they do…See the “Optional linking” section here for their description: https://crates.io/crates/ash#optional-linking
