Clang'ed FreeBSD: Builds Quicker, Uses Way Less RAM

[archibald]

I was compiling libreoffice (and the rest of the programs I had installed) on a celeron (a low spec one supported by 3GB of RAM), so on that laptop a 10% speedup would have saved me about an hour. With that said, I'm recompiling things left right and centre to try to help test the new Xorg packages - basically if I want to install something I need to compile it. People who aren't tracking the development branch don't need to do this though, so for them it's unlikely to be an issue.

I suppose one of their concerns could be how much development time is spent waiting for things to compile; from that point of view, anything that allows developers to spend more time working and less time waiting is probably a good thing for them.

[oliver]

Increased compiling speed is 'nice to have' but it's not that important.

I too run gentoo and desktop and servers. And while compiling takes a lot of time sometimes, it doesn't matter. Say Clang is 25% faster, which its not, compiling libreoffice now takes about 30-45 minutes on my phenom x6. I'm not sitting there twiddling my thumbs waiting for it to finish. So if it's done 10 minutes faster, while nice, I probably never notice it.

As for ram usage during compilation, while I'm for memory efficiency in running programms etc, very strongly in favor. But during compilation, who cares. Really, who cares.

[archibald]

It is completely useless to measure compiling times or memory usage for clang while compiling.

Try compiling libreoffice - anything that can speed up that beast is a bonus in my book!

natch, as FreeBSD is intended as a SERVER OS not a desktop OS

This isn't quite true. That's certainly where is gets most of its use though. With the new intel drivers and intel KMS support they're making an effort to become more appealing as a desktop OS. They've got very recent drivers and libraries in their testing branch, and I would expect to see them merged to mainline in the next couple of months (hopefully in time for 9.2).

[septianix]

Under FreeBSD installing software from the ports tree means compiling it (guess where Gentoo got the portage idea from). Compiler performance is more important there. Binary performance seldom makes a real life difference.

Exactly, this is a big plus for FreeBSD users. As for the performance gap it's way overblown in most cases. I have compared two FreeBSD 9 system, one with ports compiled with mostly gcc46 the other with clang. The result: I found no noticeable difference in performance in all the applications I use except for the most CPU intensive tasks I run which is ... compiling!

[DeepDayze]

i love freebsd and their shenanigans. i sucks as a desktop OS but they are a thorn in the side of gnu-zealots.

natch, as FreeBSD is intended as a SERVER OS not a desktop OS

[garegin]

i love freebsd and their shenanigans. i sucks as a desktop OS but they are a thorn in the side of gnu-zealots.

[markg85]

Oh wow. Sorry to say this but it's crap!

It is completely useless to measure compiling times or memory usage for clang while compiling. That's about the same as measuring the cpu usage while compiling and complaining when it's at a solid 100%..

The thing that matters here is if the compiled binary is bigger/smaller compared to gcc and even if that is smaller (which would be better) then it's still a big question if the compiled binary is faster and more memory efficient then the gcc one. I would bet that a GCC compiled binary beats a LLVM one in every possible way though LLVM is progressing rather fast and nicely so that might change sometime. For now GCC is the best one out there on those two.

My 2 cents..

[Awesomeness]

do you get money every time you say in a clang related article that it is Apple sponsored?

I guess he gets as much money from Apple as he gets from Canonical when he writes that Unity is by them and from Red Hat when he mentions that GNOME Shell is mostly by them.

I appreciate that clang may be more CPU and memory efficient than gcc when compiling some C++ programs. That's great.

But, for a distro, the issue of the speed of the resultant binaries as well as well (in some part) the size of those binaries.

Under FreeBSD installing software from the ports tree means compiling it (guess where Gentoo got the portage idea from). Compiler performance is more important there. Binary performance seldom makes a real life difference.

[willmore]

For a distro is this a wise decision?

I appreciate that clang may be more CPU and memory efficient than gcc when compiling some C++ programs. That's great.

But, for a distro, the issue of the speed of the resultant binaries as well as well (in some part) the size of those binaries.

This is similar to the arguements over which compression program to use on distrubtions package files. If 10x more CPU in compression or 10x (or even in combination with) more memory usage during the compression stage is worth it if it produces a slightly smaller package as long as there are no large negatives in the decompression stage.

Compile/compress once run/decompress *many times*. It's pretty clear which side deserves more effort.

[mark_]

do you get money every time you say in a clang related article that it is Apple sponsored?