Why is FreeBSD deprecating GCC in favor of Clang/LLVM?

http://unix.stackexchange.com/questions/49906/why-is-freebsd-deprecating-gcc-in-favor-of-clang-llvm

by wslh • 13 years ago

134 159 13 years ago

unix.stackexchange.com

saurik10• 13 years ago

It continues to be disappointing (which I think is the least strong word that could be used) that clang's website attempts to demonstrate the superiority of clang by comparing against the error messages from gcc 4.2, a version of gcc that wasn't even the most current when that comparison was published. This then leads to people accepting this as fact (especially amongst developers targeting Apple products who don't even seem to realize versions of gcc later than 4.2 exist, as Apple refused to continue work on gcc due to the licensing changes) an turn up in answers like this one as the "most notable" technical advantage of the product.

In fact, the error messages of the two systems are largely a tradeoff at this point, and in a project I work on that compiles using both gcc and clang (targeting MinGW and OSX) simultaneously, on multiple occasions gcc has returned the more useful error messages (even if you discount the times when clang outright crashes while trying to compile my code, something gcc hasn't done to me in a decade). Even to the extent to which clang is better (which is mostly how it will guess what you meant and then compile the rest of the file under that assumption), it is kind of insulting in an open source context to treat that as a competitive advantage when I'm certain gcc would have loved that patch as a contribution.

aspensmonster2• 13 years ago

as Apple refused to continue work on gcc due to the licensing changes

This. As stated in the accepted answer in the linked article, licensing issues are the real reason. Technical differences are ephemeral. The competition between GCC and LLVM has a habit of inducing improvments in both projects. By all means, at any given point in time one project can have an edge over another in this niche or that edge case. But fundamentally this is a philosophical divide between Free Software and Open Source Software.

JoshTriplett2• 13 years ago

This is a licensing debate, but please don't conflate that with a divide between "Free Software" and "Open Source". Both GCC and LLVM/clang qualify as both Free Software and Open Source software. Both terms encompass both copyleft and non-copyleft (all-permissive) licensing.

GCC uses a copyleft license; LLVM/clang uses an all-permissive license. FreeBSD prefers to avoid dependencies on copyleft code, especially in their core; thus, they naturally prefer LLVM/clang.

jevinskie2• 13 years ago

Technical differences are ephemeral.

Hardly the case when you want to extend/embed your compiler frontend/backend. In this case, LLVM/Clang shine. They are written in fairly clean and comprehensible C++ libraries. Most CLI tools are well under 1k lines! GCC is a real pain to even comprehend, let alone extend/embed!

saurik10• 13 years ago

(Right. I think we are in agreement, but to be very clear for others who might only read these comments: the accepted answer is from where I quoted the "most notable" technical benefit.)

hcarvalhoalves• 13 years ago

clang's website attempts to demonstrate the superiority of clang by comparing against the error messages from gcc 4.2

Even if it that doesn't hold true anymore, I think it's valid to highlight since it's one of the motivations, probably one of the original creators' pain points with GCC.

The website reads:

    Some of the goals for the project include the following:

    End-User Features:
    - Expressive diagnostics (examples)

eonil• 13 years ago

Clang is targeting specific market - free and non-GPLv3 compiler (with OBJC support).

The sales catchphrase superior error message is still true for the people in the market. And comparison with compiler in GPLv3 is meaningless to them.

You're just out of target.

kps2• 13 years ago

You could always submit a patch.

   talks about the experience provided by the command line compiler, contrasting
  -Clang output to GCC 4.2's output in several examples.
  +Clang output to that of GCC 4.2 (current when the Clang project began)
  +in several examples.
   <!--

rsynnott4• 13 years ago

clang's website attempts to demonstrate the superiority of clang by comparing against the error messages from gcc 4.2, a version of gcc that wasn't even the most current when that comparison was published.

This is the latest version available by default on MacOS and FreeBSD (for licensing reasons; subsequent versions are GPLv3), the two operating systems where clang sees most use.

saurik10• 13 years ago

At which point you may as well just point out "clang is maintained; unlike gcc 4.2, which at best will only receive back-ported compilation and security fixes, clang is under active development" ;P. You don't even need to motivate "better": it just needs to be "all least sort of as good". In the end, the only legit reason for this change is licensing, and any references to technical superiority between clang and gcc as a whole are red herrings at best and disingenuous at worst.

6ren2• 13 years ago

gcc 4.2.1 was the last version to use GPLv2. Later versions switched to GPLv3, which the BSD guys couldn't swallow. From skimming the answer, I think it's pretty clear that the switch wasn't due to technical reasons, but that Clang/LLVM is now "good enough".

It seems reasonable for FreeBSD to favour BSD licensing.

saurik10• 13 years ago

I am not exactly certain what you are disagreeing with from what I said (as I agree that the only legitimate reason is licensing), so I guess I'll justt point out that the answer you apparently only "skimmed" has a section on "technical benefits" that specifically listed a "most notable" example being improved error messages.

cmccabe2• 13 years ago

Well, I like the clang error messages better than gcc's, and I use recent versions. And to refer to the major architectural differences between the two as something that could be bridged by a single "patch" is a little silly.

gcc is still better at optimization, though. For now.

[edit: wow, people are really confused here. Guys, LLVM was designed from the start to have a clean separation between passes. You can compile some other language to the LLVM bytecode and then just use the code generator, for example. It even includes a JIT! In contrast, gcc is a huge monolith. The projects are written in different languages, too... C++ versus C.

tl;dr is that when you say that "the only difference is the license," you sound like someone saying "the only difference between Ruby and Java is the file extension." Pointy haired.]

illumen• 13 years ago

gcc is in C++ for a while now.

gcc has been becoming more modular over previous releases, and is becoming more modular in the future. For example, there are plugins since 2010.

Your information is outdated, so it makes your opinion seem uninformed.

saurik10• 13 years ago

When a parse failure occurs, making an educated guess and continuing is not a "major architectural difference". (As for your edit, I now challenge you to tell us how these architecture passed and the JIT help FreeBSD's use cases. That said, my comment you are responding to just expresses bother about the "most notable" technical benefit listed by this SO answer, so I'm not even certain why you feel talking about other things is relevant.)

lbenes• 13 years ago

The real disappointment to me is that this decision was made over an irrational licensing issue instead of technical merits. If Clang was superior to GCC in the vast majority of benchmarks, then I would support this decision. But that’s not the case, GCC is still leading in many benchmarks and can be an order of magnitude faster when the commonly used OpenMP library is used.[1] BSD users are the losers here.

[1] http://www.phoronix.com/scan.php?page=article&item=llvm_33sv...

D9u• 13 years ago

I've been using clang/LLVM in my FreeBSD system for several months now, and I've noticed that clang builds faster than gcc, and I've not had any issues with reliability, so I don't see what the problem is here.

Also, as noted in the comments on StackExchange, BSD licensing vs GPL may be reason enough to discontinue the use of gcc in FreeBSD systems.

lmm• 13 years ago

less-optimizing compilers usually build faster than more-optimizing compilers.

jorgecastillo• 13 years ago

I would say that the license is to a certain degree also a technical reason to use LLVM. Projects, corporations or people can integrate LLVM into their product with less hassle than with GCC. You can't integrate GCC into a product directly, GCC can only be used as an independent binary. I would think that this would make LLVM more enticing to developers. Obviously this depends on the intended use if all you need is a compiler GCC might still be the better choice.

superbaconman• 13 years ago

Politics really does destroy all that is holy. in this case free software. should I be able to extend free software for profit or not? I would say yes... but there are some cases in which I would say no. for a compiler I would say no one should be able to make money off an extension. it just doesn't make since.

icarus1272• 13 years ago

I was thinking this as well. I see the GPL, and Copy Left licensing in general, as a sort of regulatory force in free software. If you want to participate the GPL will put constraints on you to play nice with others in the FOSS space.

Given what happened in the financial markets as they were deregulated I am extremely suspicious of the notion that removing the constraints of Copy Left will actually be positive in the long run. I have no doubt that it will help companies like Apple but I don't see the benefit for the wider community of developers.

jcase3• 13 years ago

I'm not sure this is true (anymore?). There are many open source projects who are thriving under non copyleft licenses.

If someone wanted to be a dick about it they could release modified GPL code as one big diff and publish it on some obscure webpage only their customers have access to. Instead we see more and more companies and individuals becoming friendly open source citizens. Perhaps some because the GPL mandates it but I would like to think mainly because they feel it's right thing to do—copyleft or not.

voltagex_• 13 years ago

The GPL is supposed to put constraints on you, but I'm still waiting for a source code release from Billion for their router firmware (Busybox) and Philips have removed access to their GPL source trees.

gaadd33• 13 years ago

Why do you feel that? There's no benefit to a company in maintaining an ever evolving and complex fork of a project as compared to contributing back the non business essential parts so that their fork can be smaller.

sophacles2• 13 years ago

What I don't understand is that the same people who get up in arms about government, and bitch and moan about "rights to the fruit of my labor", don't like the notion that someone else can claim those very rights by putting a gnu license on something. At that point they get all whiney about unfairness to their little business and how they could just make money if the could appropriate someone else's labor against their terms (aka the thing they hate about taxes). It's an odd cognitive dissonance.

pnathan2• 13 years ago

Excuse me. I am an existence proof of the falsity of your statement. I don't like government much, and I do think I have a right to the fruit of my labor. And I'm more than happy to use GPL software when it is appropriate, and I'm quite happy to license my own labor out as AGPL3.

anonymoushn• 13 years ago

Taxation deprives the victim of property. Distributing derivative works of OSS does not deprive the OSS authors of anything. Distributing works that use OSS as a library does not deprive the OSS authors of anything. Running a service which does not distribute OSS but depends on the OSS internally does not deprive the OSS authors of anything.

Of course, people can use whatever license they want. I personally favor licenses that maximize the rights of my users. Hopefully, though, you can now see that the combination of positions you mentioned is a consistent one.

thwarted• 13 years ago

the same people who get up in arms about government, and bitch and moan about "rights to the fruit of my labor", don't like the notion that someone else can claim those very rights by putting a gnu license on something

I misread this at first. I think this makes the point more cogent:

the same people who get up in arms about government, and bitch and moan about "rights to the fruit of my labor", don't like the notion that someone can claim "the rights to the fruit of their" labor by putting a gnu license on something

Reply

jcase3• 13 years ago

Not sure if you mean for profit and proprietary or just making money in general. For some reason though a lot of people think you can't sell GPL'd software which you absolutely can.

Reply

foobarbazqux12• 13 years ago

Why would I pay for a copy of some GPL software when I could just wait for someone else to buy it and then get it from them for free?

Reply

wes-exp• 13 years ago

You can sell GPL software. It's just really hard to actually make money :-P

Reply

rickjames282• 13 years ago

The only thing you have a say in is code that you have copyright on.

Reply

guelo• 13 years ago

Would the world be a better place if Apple weren't allowed to Tivoize the iDevices because all the free software in them were licensed under GPLv3? I'd say yes.

Reply

jevinskie2• 13 years ago

Who is to say all free software developers want to prevent Tivoization? If all developers wanted to prevent that, they would use GPLv3! Clearly many developers like the MIT/BSD licenses. Are you advocating that these developers switch to GPLv3 or... what is your alternative?

Reply

MBCook2• 13 years ago

Torvalds himself has said that he things Tivoization should be allowed, but many other big kernel developers disagree.

Reply

zhemao• 13 years ago

I'm not sure what you're talking about. You think if the software is GPLv3, Apple won't put restrictions in their hardware in order to be able to shift the software with OSX? More likely they will just not use that software and go write their own (which is, after all, one of the reasons behind their support of the LLVM project in the first place). If all open source software was GPLv3, Apple would ship only proprietary software on their systems, so we'd go from OSX using free software with proprietary modifications, to OSX using completely proprietary software. Hardly much of an improvement.

Reply

vvhn• 13 years ago

What's FreeBSD's adoption of llvm/clang got to do with Apple's non-adoption of GPLv3 ? (Linux is a GPLv3 holdout as well btw).

The *BSDs prefer a non-GPL licensed alternative if they can get one.

Reply

egil• 13 years ago

Linux is licensed as GPL v2 (only), so a switch to gpl v3 would require concent from all contributors and be practically impossible.

Reply

rsynnott4• 13 years ago

What on earth does this have to do with Clang? You realise that Apple adopting GPLv3 GCC wouldn't require all iOS software to be licensed under GPLv3, right?

Reply

pilif• 13 years ago

We can clearly see that since the move to GPL3, Apple is reimplementing a lot of free software previously licensed under GPL2 (I wonder how long the default shell remains outdated bash. I wonder what they will move to). And unlike clang, not all of their replacements are free software.

Case in point: their SMB stack.

To me this shows that Apple would just not have used (and contributed to) any free software if it was all GPL3. The world would be a worse place. In fact, the world has gotten considerably worse, Apple-wise, when we lost Samba in OSX.

The patent clauses in GPL3 make that license completely unsuitable for any company (with cautious lawyers) that owns significant patents. I can really kind if understand Apple here.

Reply

belorn5• 13 years ago

Just to put it into perspective, the the patent clauses in GPLv3 was initially copied from the apache license almost word for word.

The only significant change that FSF did was adding an additional point regarding patent agreements. They describe that: If you have an expressed permission to practice a patent or covenant not to sue for patent infringement, such agreement is seen as identical as if you were holding the patent yourself.

If you are in a company that has significant amount of patents, and who would prefer keeping a patent rather than gaining benefit from free and open source software, then one need to keep an eye on any software one want to distribute that has apache, GPL or any other licenses with a patent clause in them.

If you are in a company that has significant amount of patent agreements, then one need to keep an eye on any software one want to distribute that got GPLv3 software in it. Apple do have a few such agreements.

Reply

plywoodtrees• 13 years ago

gcc, for the sake of wanting to make it hard to call it from proprietary front-ends, is architecturally tangled. (See http://comments.gmane.org/gmane.comp.gcc.devel/59296 - the "new methods" mentioned there never emerged). Proposals to refactor it have been vetoed by rms.

As a result, gcc is substantially harder to understand and work on than other compilers of similar scope.

As a result, new development effort, both personal and corporate, is going into clang.

As a result, even if clang is only approximately at parity with gcc now, people expect it to improve faster in future, and therefore to be a better choice now.

All of these contribute to what http://lists.freebsd.org/pipermail/freebsd-questions/2012-Ju... says are non-licence reasons to replace gcc:

  1) GCC's continuously increasing propensity to generate "bad code",

  2) The inability of GCC mamintainers to fix _long-standing_ bugs, some
     have been identified for over a decade, and have not been fixed.

  3) The continuously increasing trend of introducing 'non standard' features,

  4) The growing need to 'write around' correct/valid code that GCC will not 
     compile.

  5) The fact that the GCC code is 'unmaintainable' -- *NO*ONE* (other than 
     someone who has been working with GCC internals for "forever" --a decade
     at an absolute minimum) has any chance of 'understanding' what it is
     doing internally.

Reply

sanxiyn3• 13 years ago

"New methods" did emerge, and the result is GCC plugin. http://gcc.gnu.org/wiki/plugins

GCC plugin needs to be GPL-compatible, and GCC includes a license check similar to Linux kernel module tainting. (Instead of tainting, you can't run plugin at all.) http://gcc.gnu.org/onlinedocs/gccint/Plugin-API.html

Reply

LeeLorean• 13 years ago

In a word - Apple. IOS and OSX borrow heavily from BSD

Apple has made more money form free software than anyone else, and they can pressure FreeBSD to stay away from GPL3

Reply

mbreese• 13 years ago

In a word - no. FreeBSD has been around a lot longer than Apple's switch to Darwin. FreeBSD has a long history of preferring BSD licensed code to GPL licensed code. Apple has just financed clang/llvm. Apple has their reasons for using clang, and FreeBSD has theirs. FreeBSD didn't need any pressure from Apple to make the switch.

Reply

rsynnott4• 13 years ago

they can pressure FreeBSD to stay away from GPL3

Yes; Apple can pressure the FreeBSD project to stay away from GPLv3, through their evil Apple magic. In fact, they are presumably pressuring the whole world to stay away from GPLv3, as it has seen almost no adoption outside the FSF! Naughty, naughty Apple. If not for mean ol' Apple, the FreeBSD project would no doubt rush to embrace GPLv3, discarding their horrible old BSD license!

Reply

stox• 13 years ago

Be glad for the rise of Clang/LLVM. When there is a serious competitor, quality will stay higher and the projects will evolve faster. ie. Clang will force gcc to become better. gcc will force Clang to become better. End result, we all WIN!

Reply

hobs• 13 years ago

or more commonly I see with the big projects: 1. Big open source software project has almost all of the market share for specific task. 2. Userbase gets tired of old/bad/needless/shitty management stuff in the underlying whatever. 3. New hotness(or several) appears on stage 4. One emerges as main successor to (1) 5. Goto 1

Reply

octo_t• 13 years ago

Not in the case of big software projects (and your C compiler is undoubtedly a big software project). Its taken about 10 years and millions of man hours to get clang/llvm on a performance parity with gcc.

Any new C compiler has to either just be a C -> optimised C compiler (in which case why not just write an LLVM pass) or duplicate all of the optimisations which exist in LLVM/gcc.

Reply

MichaelMoser123• 13 years ago

I did not understand the 'political' issue: The GPLv3 license does not apply to executable files created by gcc; you can even use gcc to create proprietary products. What is the problem of using gcc 4.7 in order to compile BSD if the license does not apply to the executable - BSD in this case ?

Reply

merijnv• 13 years ago

FreeBSD is not just a kernel, it also ships with a userland, which includes a C compiler (obviously).

In other words, FreeBSD ships with a gcc binary, which means FreeBSD is distributing gcc, which means all clauses in the GPLv3 apply to them.

Any group/company/whatever wishing to ship FreeBSD that doesn't strip out the compiler will be affected similarly.

Reply

sanxiyn3• 13 years ago

While Clang's GCC compatibility is quite good, 12.1% of Debian archive still fails to build with Clang. It is an almost drop-in replacement, but not a drop-in replacement. http://clang.debian.net/

Reply

mitchty2• 13 years ago

Why would it be drop in? The default for clang is c99, for gcc it is still c89, if this was never specified in makefiles you're in for a world of possible hurt. Most "errors" I see on that page amount to boring differences of how the compilers treat C, or how clang doesn't support some random gcc extension, or how clang differs in its interpretation of implementation defined behavior.

Reply

sanxiyn3• 13 years ago

I am just pointing this out because some people on this thread are claiming "I have been using Clang for N months and had no problem". I believe them, but it won't necessarily work for you, because Clang is not a drop-in replacement.

Clang doesn't need to be a drop-in replacement and I think it shouldn't be, but let the buyers beware.

Reply

0x092• 13 years ago

The statistics on that page do a good job of explaining that in many (most?) of the cases clang will never quite be a drop-in replacement, because the failure is in actuality uncovering a bug in either GCC or the software being compiled.

Sylvestre Ledru spent some time on this in particular in his talk at this past FOSDEM[1], taking the above as evidence that clang is now mature enough to support the Debian archive.

[1] https://archive.fosdem.org/2013/schedule/event/debian_clang/

Reply

Why is FreeBSD deprecating GCC in favor of Clang/LLVM?

Related Stories