this post was submitted on 07 Jun 2023
136 points (94.2% liked)
Asklemmy
43978 readers
590 users here now
A loosely moderated place to ask open-ended questions
If your post meets the following criteria, it's welcome here!
- Open-ended question
- Not offensive: at this point, we do not have the bandwidth to moderate overtly political discussions. Assume best intent and be excellent to each other.
- Not regarding using or support for Lemmy: context, see the list of support communities and tools for finding communities below
- Not ad nauseam inducing: please make sure it is a question that would be new to most members
- An actual topic of discussion
Looking for support?
Looking for a community?
- Lemmyverse: community search
- sub.rehab: maps old subreddits to fediverse options, marks official as such
- [email protected]: a community for finding communities
~Icon~ ~by~ ~@Double_[email protected]~
founded 5 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
That's what I mean by “static linking with extra steps”. This problem was already solved a very long time ago. You only get these version conflicts if your dependencies are dynamically linked, and you don't have to dynamically link your dependencies.
Yes, you don't have to dynamically link dependecies, but you don't want to recompile your app just to change dependency version.
Don't I? Recompiling avoids ABI stability issues and will reliably fail if there is a breaking API change, whereas not recompiling will cause undefined behavior if either of those things happens.
That's why semver exists. Major-update-number.Minor-update-number.Patch-number Usually, you don't care about patches, they address efficency of things inside of lib, no Api changes. Something breaking could be in minor update, so you should check changelogs to see if you gonna make something about it. Major version most likely will break things. If you'll understand this, you'll find dynamic linking beneficial(no need to recompile on every lib update), and containers will eliminate stability issues cause libs won't update to next minor/major version without tests.
What's so horribly inconvenient about recompiling, anyway? Unless you're compiling Chromium or something, it doesn't take that long.
Still, it's going to take some time, every time some dependency(of dependency(of dependency)) changes(cause you don't wanna end up with critical vulnerability). Also, if app going to execute some other binary with same dependency X, dependency X gonna be in memory only once.
Compared to the downsides of using a container image (duplication of system files like libc, dynamic linking overhead, complexity, etc), this is not a compelling advantage.
That seems like a questionable design choice.
I mean, you could have GUI for some CLI tool. Then you would need to run binary GUI, and either run binary CLI from GUI or have it as daemon. Also, if you are going to make something that have more than one binary, you'll get more space overhead for static linking than for containers
Man, that's underestimating compiling time and frequency of updates of various libs, and overestimating overhead from dynamic linking (it's so small it's calculated in CPU cycles). Basically, dynamic linking reduces update overhead, like with static linking you'll need to download full binary every update, even if lib is tiny, while with dynamic you'll have to download only small lib.
Yes, I've seen that pattern before, but:
If they're meant to run on the same machine and are bundled together in the same container image, I would call that a questionable design choice.
Well, I have only my own experience to go on, but I am not usually bothered by compile times. I used to compile my own Linux kernels, for goodness' sake. I would just leave it to do its thing and go do something else while I wait. Not a big deal.
Again, there are exceptions like Chromium, which take an obscenely long time to compile, but I assume we're talking about something that takes minutes to compile, not hours or days.
No, I'm not. If you're not using JIT compilation, the overhead of dynamic linking is severe, not because of how long it takes to call a dynamically-linked function (you're right, that part is reasonably fast), but because inlining across a dynamic link is impossible, and inlining is, as matklad once put it, the mother of all other optimizations. Dynamic linking leaves potentially a lot of performance on the table.
This wasn't the case before link-time optimization was a thing, mind you, but it is now.
Okay, but I'm much more concerned with execution speed and memory usage than with how long it takes to download or compile an executable.
In the time i was thinking about some kind of toolkit installed though distrobox. Distrobox, basically, allows you to use anything from containers as if it was not. It uses podman, so i guess it could be impossible to use docker for GUI, although i cant really tell.
Yes, but static linking means you'll get security and performance patches with some delay, while dynamic means you'll get patches ASAP.
Some claim this doesn't work in practice because of the ABI issues I mentioned earlier. You brought up Semver as a solution, but that too doesn't seem to work in practice; see for example OpenSSL, which follows Semver and still has ABI issues that can result in undefined behavior. Ironically this can create security vulnerabilities.
Yeah, but there's by lot more security improvement by having ability to apply fix for severe vulnerability ASAP than weakening from possible incompativilities. Also, i wonder why i never brought it up, shared libs are shared, so you can use them across many programming languages. So, no, static is not the way to replace containers with dynamic linking, but yes, they share some use cases.
Um, we're talking about undefined behavior here. That creates potential RCE vulnerabilities—the most severe kind of vulnerability. So no, a botched dynamically-linked library update can easily create a vulnerability worse than the one it's meant to fix.
Shared libraries are shared among processes, not programming languages.
You still can use them in any programming language