LLVM Clang 14 Lands An "Amazing" Performance Optimization

[oiaohm]

Yeah, you can think of it as limiting the freedom of immediate downstream contributors for the benefit of all those downstream of them. It's trading local restrictions for more freedom, globally.

However, I think it's silly to moralize about open source licenses. They each have their benefits and drawbacks, and which appears best depends entirely on where you sit and what your goals are.

Moralize model does make sense to work out why different choices are being made. If the downstream parties are not moral and will take open source work and proceeded to closed source it, monetise it and finally directly compete with the open source work so effect running it out of market a non copyleft license is not a good move. Yes companies willing todo the bad behavour generally is like disobey the basic old rule as well "Don't kill the goose that lays the gold egg" Yes the upstream open source project is normally the one laying the golden egg.

One way copyleft is limiting freedom but in some areas of the market that limit freedom is very much like a general world laws of do not kill the project you are taking advantage of.

If everyone in this world was 100% moral we would not need to punish people for murder. The reality is the world is morally imperfect.

Lot of people who want to push the moral as the reason to use X license in the process forget we are living in a moral imperfect world. Depending on how morally imperfect the people you are dealing and your own moral standards with serous-ally alters your license choice.

[mdedetrich]

What do you mean by "worse"? Legally? Financially? Choice wise not enough software?

It was very unpopular at the time, back then (at least in America) proprietary software was seen as the go to, red hat was the exception that pushed things forward that the industry needed.

[coder]

Coremark is the odd one out. Wonder what that is about.

I think it's nearly impossible to implement a compiler optimization that delivers real gains without hurting at least a few benchmarks on at least a few platforms.

It could be something like that Coremark was manually optimized in ways that it doesn't benefit from the optimization where it would help, but applying the optimization does create additional register pressure -> more spilling than before.

Of course, that's just hypothetical. Obviously, any notable regressions should be studied to see if mitigations can be added for such cases.

[coder]

The GPL does not just give freedom, it protects freedom for everyone, so that your code is a gift that keeps on giving.

Yeah, you can think of it as limiting the freedom of immediate downstream contributors for the benefit of all those downstream of them. It's trading local restrictions for more freedom, globally.

However, I think it's silly to moralize about open source licenses. They each have their benefits and drawbacks, and which appears best depends entirely on where you sit and what your goals are.

[oiaohm]

now that pipewire is becoming required for screen capture and audio on Linux, if it was GPL it would force nearly all applications on GNU+Linux to be GPL. even firefox is GPL incompatible. libraries like that pretty much have to be MIT, same as Wayland and X.Org.

Not all copyleft licenses are as nasty. LGPL and MPL in the copylefts in majority of cases would work. Pure GPL is not meant to be used in applications in most cases. GPLv2 with the Linux kernel has the Linus exception that limits it reach.

[oiaohm]

Interesting, that hasn't really been my experience or for other people contributing to open source on company time I've read about.

It does depend who you talk to. Allwinner case to get driver source code and many others. There are quite a few companies where you end up depending on the copyleft license to in fact get the company to release their source code and to consider working with other parties.

jabl not all companies are created equal. There are a lot of morally inferior companies who have gag orders on their staff. Yes its like with wine you could talk to the wine developers and the staff at codeweavers and it was all nice stories. Then you had the TransGaming fork of wine where you could not talk to their developers and they were not submitting upstream and got really upset when wine main project changed to lgpl where they could not keep on adding on their custom only code in secret. Seeing this was when I came aware how these companies work yes the morally inferior are more likely to legally gag their workers.

The markets where you are needing copyleft because of morally inferior companies you don't hear many stories about. Not that they don't exist contractually if their developers want to keep their job they cannot talk about how their company treats open source submits.

jabl there is a dark underside that we should not ignore. Looking at the normally morally superior companies stories is less than 1/4 of the developers in fact working with open source code.

Yes there is a problem where people start saying that people should not use copyleft while completely ignoring they are in a market where copyleft is kind of a require feature.

Another place you see it is the amount of hardware support the Linux kernel has vs freebsd. Linux kernel has more because there are a lot parties up-streaming code who would not without the gplv2 requirements. Yes there are a lot of hardware companies who want to be morally questionable. There are also some who are morally superior in hardware companies as well.

[novideo]

There are exceptions even wrt to Redhat, i.e. pipewire (the new audio/video stack being developed there) is MIT licensed https://github.com/PipeWire/pipewire

now that pipewire is required for screen capture and audio on Linux, if it was GPL it would force nearly all applications on GNU+Linux to be GPL. even firefox is GPL incompatible. libraries like that pretty much have to be MIT, same as Wayland and X.Org.

[LuukD]

Licenses and other detours aside, for any modern compiler to receive runtime improvements, in a single patchset, sized ~3-12%, is big. Atleast I am amazed! Kudos to Djordje.
In his benchmark, even asmfish improves (even though it is mostly assembly). Coremark is the odd one out. Wonder what that is about.

[Rallos Zek]

The license is non-copyleft (BSD and Apache), which is morally superior, at least.

Non-copyleft licenses are for cuckolds and cowards.

[jabl]

https://www.libreoffice.org/about-us/licenses

What are you attempting to be a joke. OpenOffice under Apache 2.0 vs LibreOffice under Mozilla Public License you should have a serous look at. Yes Mozilla Public License is a copyleft license. Yes libreoffice has end up as the higher quality code and more paid full time developers.

I find it hard to imagine a more irrelevant example of one type of license being superior than another, than LibreOffice (LO) vs Apache OpenOffice (AOO). It has everything to do with the history how those projects came into being and more or less absolutely nothing to do with the license. LO started out essentially as a bunch of community-supported patches on top of Sun OpenOffice to make it work better on then current Linux distros, eventually for all practical matters growing into the "real" upstream. AOO was never more than a dump over the wall by Oracle since they weren't interested in OO anymore. Why should a community have formed around that one, considering that anyone interested in any kind of OpenOffice community had already congregated around LO?

In reality, there are plenty of examples of very successful copyleft and very successful permissively licensed projects. It's as if the choice of license isn't as centrally important as internet forum warriors make it out to be, imagine that..

Legal departments in many companies are more willing to sign off for their staff working on copyleft projects sharing their work out side the company than BSD or Apache ones.

Interesting, that hasn't really been my experience or for other people contributing to open source on company time I've read about.