GCC 4.9 is doing “some seriously crazy shit” according to Linus Torvalds

https://lkml.org/lkml/2014/7/24/584

by jdoliner • 12 years ago

213 238 12 years ago

lkml.org

e40• 12 years ago

More and interesting conversation happens here:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61904

frik• 12 years ago

The bad compiler versions are 4.5.0 (when debug_insn came in) to 4.8.3 and 4.9.0 and 4.9.1.

wfunction2• 12 years ago

"Here's a small (auto-)reduced testcase for this specific issue"

Do you happen to know how that is done? How is a test case "auto"-reduced? I've never done that before.

kzrdude2• 12 years ago

For example http://embed.cs.utah.edu/creduce/ ; the slides from a presentation in 2012 are interesting: http://embed.cs.utah.edu/creduce/pldi12_talk.pdf

a3_nm2• 12 years ago

https://gcc.gnu.org/wiki/A_guide_to_testcase_reduction

dochtman• 12 years ago

The LLVM project recommends delta:

http://delta.tigris.org/

CyberShadow• 12 years ago

For D we use DustMite (disclaimer: I'm the author). It also works to some extent with C/C++ (thanks to the languages' syntax being similar enough).

https://github.com/CyberShadow/DustMite/wiki

rwmj• 12 years ago

I think it's a case of "bugs happen". They fixed this one pretty quickly.

HCIdivision17• 12 years ago

The tone difference is certainly stark! The bug report feels like a process is in play, while the mail has a tone of a rant in a closed room.

I often take it for granted the compiler is always right (and it darn well just about is), but bugs like this look exactly like mis-cast magic to me. Learning how the compiler makes decisions seems like a rabbit hole, but may be fascinating enough to be worth it. (And think about being a maintainer of such a complex beast!)

davidgerard• 12 years ago

Fixed in tree: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61904

reidrac2• 12 years ago

I find interesting that the tone in the bug report is completely different (no "colourful language").

kuon• 12 years ago

Because one don't fuck with bug report!

Aardwolf• 12 years ago

What's up with the many insults? A bit of professionalism amongst developers would be nice...

test1235• 12 years ago

Sounds like regular ol' Linus to me ...

stickydink• 12 years ago

This is just Linus, unfortunately

Svip4• 12 years ago

So it will get posted on HN and other news websites.[0] Notice how no one else in the thread cares about Linus' language.

[0] He is after all warning about the use of gcc-4.9.0

Argorak5• 12 years ago

Notice how no one else in the thread cares about Linus' language.

Except Sarah Sharp and others.

rsynnott4• 12 years ago

As in, Torvalds being mean about the compiler, or something else? That seems pretty par for the course in the context of a badly-broken thing that people tend to expect never to be broken.

Sharlin• 12 years ago

Also note that his tone in the actual gcc bug report is professional and matter-of-factly: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61904

gkya6• 12 years ago

Professionalism? Torvalds is a teenager with too much privileges.

dasil003• 12 years ago

Torvalds has earned his privileges as much as any developer alive or dead. In fact, if his abstract power ranks up there with Fortune 500 CEOs and national leaders, I'd say he deserves his privileges more than most of them as well.

sjwright3• 12 years ago

If he was designing a video game, your assessment would be viable. As it happens though, Torvalds writes code that makes half the world function reliably; some would say his exceptionally direct attitude has served him well in this endeavor.

watwut3• 12 years ago

Professionals come in various shapes. Torvalds is not the most polite man in the worlds, far from that. However, he achieved a lot, whether because or despite of it. So, yeah he reserve respect and deserve to be called professional. Although not to be called "the most polite man".

Argorak5• 12 years ago

Linus wants to keep this tone and considers it important.

http://arstechnica.com/information-technology/2013/07/linus-...

chippy• 12 years ago

Torvalds doesn't want a "code of conduct," saying that "venting of frustrations and anger is actually necessary, and trying to come up with some ‘code of conduct’ that says that people should be ‘respectful’ and ‘polite’ is just so much crap and bullshit.”

This is interesting when one compares it with other projects such as Django which have gone the whole other way in addressing modern views of how to approach diversity, gender issues, conduct etc.

notacoward3• 12 years ago

It might be fine for Linus to adopt such a tone, because he knows enough to moderate it up and down according to circumstances. It makes him more expressive. Note that the bug in question really does indicate some serious mistakes in the compiler, with serious consequences, and along with the invective he provides plenty of information. The problem is that hundreds of other kernel developers imitate his most severe tones all the time without such technical detail and even for technical issues where Linus himself wouldn't express disdain that strongly. That lack of "dynamic range" makes them less expressive, so nobody can tell when they're expressing justifiable concern/outrage and when they're just being dicks out of habit.

It's like the second amendment of developer communication. That sort of verbal weaponry should be available, but its use still needs to be justified.

walshemj6• 12 years ago

Non native speakers often don't have the instinctual grasp of slang and swearwords in English and so tend to use the F C and S words a lot without fully understanding the severity of the words.

"seriously crazy shit" seems a bit restrained for Linus though

Svip4• 12 years ago

No, we understand them all right. We are just not generally as prudish as English-speakers. Perhaps it's more a culture difference than a language barrier.

Dewie• 12 years ago

He's been living in the US since, what, 1996? He probably understands the cultural severity of the words he uses.

wsc981• 12 years ago

If you think those English curse words are severe, you should never hear the Dutch curse words [0].

---

[0] http://en.wikipedia.org/wiki/Dutch_profanity

[1] http://stuffdutchpeoplelike.com/2011/07/26/dutch-swears-with...

caf2• 12 years ago

You'll notice that the insults are entirely directed at an inanimate piece of software.

sjwright3• 12 years ago

I don't know. If there's any kind of software that can get emotional and hurt, it's a compiler.

kabdib3• 12 years ago

"Hey, Michaelangelo, what is that piece of shit you're carving?"

I don't think so.

n8m2• 12 years ago

Thank you!

walshemj6• 12 years ago

If it's your product you take any criticism personally especially if it some hack who messed up and blamed your baby instead of coughing to your cockup - as the tech editor of the guardian did to one of my products!

recalibrator2• 12 years ago

Linus is just being Linus. I've been following him since Red Hat 5.2 and the open source culture has always been "colourful".

His words shouldn't be taken too personally, just as getting bitten by a grizzly bear shouldn't be taken personal.

dalke4• 12 years ago

Interesting analogy, but in a way I don't think you meant. Problem bears are translocated or killed. That's not due to taking things personally but to prevent future bad interactions. If your analogy has merit, shouldn't we do the same (metaphorically) with Torvalds?

Tyrannosaurs5• 12 years ago

That may be so but I'm pretty sure that pretty much every study that's been done on this says insulting people is demotivating and generally destructive behaviour.

Linus can believe what he likes but the evidence contradicts him.

kabdib3• 12 years ago

Man, I wouldn't tolerate an email like that from anybody. It's right up there with some of the most abusive management I've ever known.

Linus might be "a great guy" or "just Linus being Linus" but email like this has no place in a cooperative environment.

_delirium2• 12 years ago

open source culture has always been "colourful"

Some open source culture, yes. However there are plenty of friendly open-source communities as well. If you want to contribute to free-software and don't want to interact with a Stallman/Torvalds/Drepper/deRaadt type personality in doing so, that's a perfectly legitimate preference, and you can find lots of projects that fit the bill!

tonfa• 12 years ago

If you read the gcc bug, he is rather professional.

macspoofing• 12 years ago

First time with Linus?

smcl2• 12 years ago

I initially felt the same, but imagine if you'd spent countless hours getting to the bottom of an obscure bug that turned out to be a completely bone-headed compiler problem outside your control. Sometimes I think Linus goes a bit far, but this time I think he's justified in his venting his irritation

anthracis417• 12 years ago

People in IT are assholes. I've only worked in retail, food, biotech and IT but IT has been my least favorite because of my coworkers. I see the shitty attitude is just an in-general IT thing and accepted.

IvyMike• 12 years ago

imagine if you'd spent countless hours getting to the bottom of an obscure bug that turned out to be a completely bone-headed [...] problem outside your control

I imagine everyone on HN has been in this situation many times during their careers.

tachyonbeam• 12 years ago

I've had to deal with bone-headed obscure bugs in the Linux kernel... Strange feature regressions, kernel-bundled drivers that used to work but have mysteriously become broken over a year ago and were never fixed because nobody gives a damn about making older hardware work. Seems unfair to treat someone like shit for a crime Linus must have been guilty of himself at some point. Unless he somehow thinks that Linux is so perfectly tested that no kernel ever gets released with really annoying bug that crash production systems. As far as I can see, the GCC people are fairly responsive, they're going to fix that bug, and that's probably the best one can hope for.

droopyEyelids• 12 years ago

He isn't a peer though. It's like how the hermit at the top of the mountain will smack the novice martial artist for making little mistakes. That doesn't make the novice resentful. Instead, he's glad for the training.

cottonseed• 12 years ago

I'm reminded of this:

http://philip.greenspun.com/ancient-history/professionalism-...

parax• 12 years ago

Just use LLVM: http://llvm.org/

kzrdude2• 12 years ago

And if you find one bug in llvm, switch back?

LeoNatan25• 12 years ago

Move to MSVC++ :P

a3_nm2• 12 years ago

People are working on it http://llvm.linuxfoundation.org but it is not a trivial thing to do.

keeperofdakeys• 12 years ago

Unfortunately the linux kernel still relies on some gnu C extensions for some low-level control of the output. I don't know if comparable stuff exists in llvm and clang, and if so, if linus has any plans on supporting them. However this is just bugs in a new compiler, and the Linux kernel is probably one of those things that hits cornercases.

exDM69• 12 years ago

I'm not sure which C extensions do you mean, but Clang supports most of the same GNU extensions that GCC does. Indeed, it's one of the goals of Clang to be compatible with GCC whenever possible. However, there might be some things that are not (yet?) supported.

And the kernel can be compiled using LLVM and Clang, but not quite in all configurations. But compiling mainline kernel for x86 using LLVM was achieved some years ago.

gillianseed• 12 years ago

A lot (if not the majority) of GCC extensions used by Linux exist in GCC because the Linux kernel developers requested them in order to have a great level of control in terms of the generated machine code, so it's not exactly surprising that the Linux kernel relies on a lot of said GCC extensions.

Many of these extensions which the Linux kernel relies on have been added to LLVM, but obviously not all, and from the looks of it some never will, which means either Linux needs to refrain from using said extensions, or building Linux using LLVM will continue to require patching.

_stephan• 12 years ago

As if LLVM had no bugs.

userbinator2• 12 years ago

The optimiser in LLVM does some even more "seriously crazy shit" - usually revolving around obscure edge cases involving officially undefined behaviour. Linus would definitely not be happy with that.

AgathaTheWitch• 12 years ago

Linus Torvalds says something thread on HN.

Prediction before clicking: Half of the thread, or more, will not be about what he says, but about his tone and "professionalism."

Never change HN. Never change.

aeberbach• 12 years ago

Linus has earned the right to that tone, care about Linux or not. Now when some unknown emulates it, that's the time to take note (and ignore that person ever after).

currysausage• 12 years ago

Quod licet Iovi, non licet bovi.

PakG1• 12 years ago

You know what, here's a thought. HN is a place that emphasizes civility. Would Linus be allowed to act like that here, or would we say that he's earned the right to act like that here? Double standards suck because something has to break to accommodate them, and often the things that break are the things that everyone else likes when the world is a nicer and more innocent place.

jburwell• 12 years ago

This attitude is breeding the jerks and aholes in this industry. It is far too common to encounter teams and communities where in professionalism is accepted on a sliding scale of "accomplishment". If you accept this type of disrespectful/demeaning behavior from one "special" person then you will get it from everyone. Eventually, people tire of this type of nonsense and leave. You may be able to backfill their talent, but regular churn per mantle inhibits the creativity and velocity of the group. Linus is extremely fortunate that folks are paid very well to take it, and many very large companies (e.g. IBM, Red Hat, Intel, etc) have a significant portion of their long term livilihoods dependent on Linux.

1ris• 12 years ago

Seriously. If a community/somebodey cares more about how something is said more than what is said, said community/person lacks the allegedly so important professionalism.

Goopplesoft• 12 years ago

Would this have really made the top of HN if it wasn't for the tone though?

mzwartbol• 12 years ago

shots fired

Pacabel6• 12 years ago

The most striking thing is the number of downvoted gray comments. Probably 10% of them are like that, at the moment!

It kind of gets distracting when so many are like that. With that many being grayed out, I know a lot of them aren't obvious commercial spam or anything like that. So I end up trying to read them all, which of course is awkward because of how they've been grayed out.

nabla93• 12 years ago

I'm really baffled about this. Maybe I'm from older hacker generation (and I'm also a Finn). I just did not see personal insults going on. I would like to see more neutral discussion about cultural differences (between hackers and nations) and not opinions and own cultural biases stated as normative.

Here is my personal cultural bias:

1. Insulting people personally is not same thing as insulting their ideas or their actions. If someone insults my actions or my ideas, I may get emotional, but I don't take it as an insult. For me it means that I'm being challenged.

2. Cursing is not unprofessional or insulting in itself. Cursing is to be used for emphasis and to get people involved emotionally (good thing). Cursing has its place to signal frustration and get trough people. Too much cursing is like underlining everything.

3. (Anglo American) business culture is not necessarily something to be emulated. Being overly polite invites all kinds of side stepping, passive aggressive behavior and ineffectiveness. I feel that "being professional" is used here as thought terminating cliche.

4. Some hacker cultures can be both fiercely competitive and cooperative at the same time. Being heavily criticized and challenged is part of that process. Kernel hackers compete to get their ideas incorporated into the same codebase. Javascript hackers borrow snippets from each other and do their own things (no need to be as competitive as forking is easier). This might lead to different hacker cultures. What is accepted in one culture is not accepted in other.

5. Management by perkele[1] is management strategy that can be very effective with the 'Shut Up and Show Them the Code' hacker culture.

6. Pacabel may have a point. Younger generations may no be used to their work or behavior being harshly criticized and take it as personal insult. https://news.ycombinator.com/item?id=8089706

[1] https://en.wikipedia.org/wiki/Management_by_perkele#Examples...

kghose4• 12 years ago

I think that the overly critical culture of many European nations prevents younger people from trying out or voicing ideas because humans are human and don't like being humiliated , especially in public. This is something I think we should not emulate.

aninhumer• 12 years ago

If someone insults my actions or my ideas, I may get emotional, but I don't take it as an insult.

The fact that people get emotional in response is exactly the problem. Knowing that an insult wasn't technically directed at them doesn't stop them feeling shitty, and being unproductive as a result.

Younger generations may no be used to their work or behavior being harshly criticized

Younger generations are used to their work being criticised constructively and dispassionately.

yohanatan12• 12 years ago

Cursing is the symptom of a small vocabulary and little self-restraint.

danford• 12 years ago

It's like people don't think the same stuff is happening behind closed doors at other large tech companies.

Forget to turn off a mic at a political conference? You get worse stuff than this.

peterwwillis• 12 years ago

My theory about this is that Linus treats his mailing list like a semi-public IRC chatroom (I mean, it is his mailing list for his kernel, so it makes sense) and HNers are not accustomed to that kind of cross-over if they've even ever idled on an irc channel run by hackers. It usually gets much more crass in there, so in that context these kind of comments make sense.

But when you consider Linus is responsible for, I dunno, billions? trillions? in spending on technology based on his OS, it's also easy to understand people getting nervous or offended by IRC-style banter. It's like seeing the CEO of a large company make jokes about retards. Pretty off-putting.

(And no, of course Linus is not a "CEO of Linux" and he doesn't owe anybody anything, i'm just saying I get how people could feel shocked or offended by his behavior)

dang• 12 years ago

This comment has added far more off-topic gunk to the thread than anyone else's.

Please don't.

quarterto• 12 years ago

Can anyone give a layperson's explanation of what exactly GCC 4.9 is doing wrong here? I'm not fluent in Assembly and/or compiler internals. What's changed from 4.8?

kryptiskt• 12 years ago

It is doing two things wrong, it is moving a constant value to the stack and it's doing it before increasing the stack pointer, so it's momentarily outside the stack area. If an interrupt hits during the instructions the stack is broken, the interrupt service routine will increase the stack for its own use and overwrite that value.

masklinn• 12 years ago

See the bug report linked by e40[0] for explanations. Interestingly, the bug actually exists since 4.5.0, it just doesn't trigger because of different optimisations being applied elsewhere.

[0] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61904

andrey-p• 12 years ago

I really enjoyed reading this even I don't understand any of what the bug is actually about. There's something about people who have taken swearing to a virtuosic level.

jonahx• 12 years ago

Whatever you think of his language and attitude, the guy deserves some love for working a Goonies reference into a complaint about the GCC compiler.

kazinator• 12 years ago

I've noticed it also has idiotic warnings. I have some code with macrology that generates lexically scoped typedefs, which are sometimes not used. I do not wish to be warned about these, damn it! Who cares about a typedef at lexical scope that is not used? It's compile-time only. What's next; a warning about every file-scope typedef that has not been used? "Hey you included <stddef.h> but never used ptrdiff_t!"

userbinator2• 12 years ago

As a long-time Asm programmer who has read a lot of compiler-generated code (usually for RE purposes), I completely sympathise with Linus - despite all the research and improvements, compilers are still very stupid, and will routinely generate code that no sane human would. The widespread notion that compilers can almost always generate better (faster or smaller) code than a human is probably closer to myth than reality. The one thing that is certainly true is they can compile orders of magnitude faster than a human can, and that's usually the main reason for using them.

I think the x86-64 ABI is also partly to blame for this mess - IMHO it's far too complex (e.g. the "red zone") and makes it easy for these bizarre edge-case-bugs to occur.

"Spilling a constant" is something I've never seen before and I'd probably WTF the first time I saw it, but thinking about it more carefully, it might actually be a wise choice in some cases - x86-64 unfortunately allows very few instructions with 64-bit immediate values, and those few instructions are very long (opcode + 8-byte immediate), so it could make sense to store some constant in memory, on the stack, for a while and put it in a register when it's needed with some moves - 64-bit r/m move with 8-bit displacement is 3/4 bytes. Of course that's not what GCC is doing here; that last move instruction is 6 bytes alone (mov + mod/rm + disp32.)

GFK_of_xmaspast• 12 years ago

This is deep into 'dog bites man' territory; Torvalds NOT being an asshole is the newsworthy story.

GCC 4.9 is doing “some seriously crazy shit” according to Linus Torvalds

Related Stories