[PATCH v2] X86: disambiguate unqualified btr, bts

Jim Grosbach wrote:

No. The above rule is absolutely the wrong thing to do, as has been
previously noted.

I don't give a shit about whether you think it is "absolutely wrong"
or not; I did what hpa and the Intel manual outlined. If you have
some _reason_ not to do that, bring it up.

I reported four bugs a few days ago, and the community has shown ZERO
(if not NEGATIVE) interest in fixing them. I got Linus and hpa to
comment on the issue, and help the community figure out what needs to
be done. I posted not one, but MULTIPLE patches demonstrating
desirable behavior, despite having ZERO prior experience with compiler
engineering. Nobody else has posted a single patch, or helped me
write one; instead, they have been sitting around being fabulously
counter-productive, and stalling all progress.

Can you remind me why I'm still trying to help LLVM, and don't just
throw it out the window? (Hint: It's sheer persistence; anyone else
would've given up a long time ago)

Do you value contributors at all? (That's a rhetorical, because I
already know the answer from the way you've been treating me: no)

Do you care about getting LLVM to work with real-world codebases?
(Again a rhetorical, because I already know the answer: no)

In case you have missed this, this is not LKML. Please keep your abusive
language at home. Linus and hpa are no almighty authorities here and
this is not the Linux kernel community.

Joerg

Joerg Sonnenberger wrote:

Linus and hpa are no almighty authorities here and
this is not the Linux kernel community.

Who said anything about almighty authorities, and who mentioned Linus
or the kernel community now? Their emails are on the LLVMDev list for
everyone to read: I picked up what made sense to me.

But whatever.

who mentioned Linus
or the kernel community now?

You did:

  I got Linus and hpa to
comment on the issue,

Linus' comments were also confrontational and impolite, and he then proceeded to continue Linux-specific discussions that were completely off-topic for this list while keeping LLVMDev on the cc: list, wasting the time (and bandwidth) of all of the subscribers to this list who are not Linux developers.

The Linux kernel community has been in the tech news in the last couple of days defending abusive behaviour on mailing lists. This is something that we do not accept in the LLVM community, nor in any of the other open source communities that I am a member of. For example:

I don't give a shit about whether you think it is "absolutely wrong"
or not

This is completely inappropriate for any public mailing list. Apparently the LKML puts up with this kind of things, but most community driven projects do not.

Do you value contributors at all? (That's a rhetorical, because I
already know the answer from the way you've been treating me: no)

Yes, contributors are valuable, however contributors are members of a community and are expected to behave in a way that reflects this. There are a lot of people on this mailing list that I have argued with on technical matters, but none has ever felt the need to resort to profanity or personal attacks.

Do you care about getting LLVM to work with real-world codebases?
(Again a rhetorical, because I already know the answer: no)

You mean like iOS, OS X, or FreeBSD, which all use Clang/LLVM as their system compiler? Or perhaps the Android NDK, which ships Clang 3.1 in the latest release and has Clang 3.3 in trunk? Or Debian, which now tests building all packages with Clang and has a repository where they can be downloaded?

Yes, we're quite interested in getting LLVM to work with real-world codebases. In many cases, we encounter a question of whether to support poor design choices in GCC that are used by a small number of packages, or impose something stricter, which benefits everyone in the long run in terms of code cleanups.

You view Linux as a very important package. This list has some Google people who might agree with you, but it also has a lot of Apple employees, and people like Joerg (who led the integration of Clang and NetBSD) and myself (a member of the FreeBSD Core Team), for whom it's just some third-party codebase with a reputation for relying on GCC-specific behaviour and containing code that relies on undefined behaviour and a vocal community that complains to GCC whenever they change their interpretation of undefined behaviour. If we can support it, that's great. If supporting it comes at a cost to other users of LLVM (either in terms of worse code, or less error reporting), then that's something that will need to be evaluated carefully on its merits, which is what the discussion that was already in progress was doing.

David

Linus is an expert on the x86 instruction set, so his advice on what to do with bts and btc should be taken seriously. So i thing that although Linus is not an LLVM authority, he should be considered an x86 authority.

James.

David Chisnall wrote:

  I got Linus and hpa to
comment on the issue,

Linus' comments were also confrontational and impolite, and he then proceeded to continue Linux-specific discussions that were completely off-topic for this list while keeping LLVMDev on the cc: list, wasting the time (and bandwidth) of all of the subscribers to this list who are not Linux developers.

I'm sorry you didn't find their inputs valuable, and discriminate
against them on the basis of tone. I personally found them _very_
insightful.

The Linux kernel community has been in the tech news in the last couple of days defending abusive behaviour on mailing lists. This is something that we do not accept in the LLVM community, nor in any of the other open source communities that I am a member of. For example:

They run the world's largest open source project; whether you think
their behavior is "abusive" or not is inconsequential. That said, you
do have every right to choose how you want your community to be run.

I don't give a shit about whether you think it is "absolutely wrong"
or not

This is completely inappropriate for any public mailing list. Apparently the LKML puts up with this kind of things, but most community driven projects do not.

Making transcendental statements without evidence is getting us
nowhere: I can't speak for communities other than the ones I've
personally been involved with, and neither should you. It just so
happens that I've been involved with the git community for a long
time.

Irrespective of whether I personally think my message was
"appropriate" or not, I apologize because the LLVM community seems to
be offended by it.

Do you value contributors at all? (That's a rhetorical, because I
already know the answer from the way you've been treating me: no)

Yes, contributors are valuable, however contributors are members of a community and are expected to behave in a way that reflects this. There are a lot of people on this mailing list that I have argued with on technical matters, but none has ever felt the need to resort to profanity or personal attacks.

You've conveniently ignored the facts, and focused all your attention
on criticizing the tone of my last email. Want to write an essay
about it?

Do you care about getting LLVM to work with real-world codebases?
(Again a rhetorical, because I already know the answer: no)

You mean like iOS, OS X, or FreeBSD, which all use Clang/LLVM as their system compiler? Or perhaps the Android NDK, which ships Clang 3.1 in the latest release and has Clang 3.3 in trunk? Or Debian, which now tests building all packages with Clang and has a repository where they can be downloaded?

I'm well aware of what LLVM powers, thank you. I've seen it up close
in Rust and Rubinius.

Yes, we're quite interested in getting LLVM to work with real-world codebases. In many cases, we encounter a question of whether to support poor design choices in GCC that are used by a small number of packages, or impose something stricter, which benefits everyone in the long run in terms of code cleanups.

Don't talk hypotheticals and make vague generalizations. I posted a
patch, which nobody bothered to look at [1]. Those are the facts.

You view Linux as a very important package. This list has some Google people who might agree with you, but it also has a lot of Apple employees, and people like Joerg (who led the integration of Clang and NetBSD) and myself (a member of the FreeBSD Core Team), for whom it's just some third-party codebase with a reputation for relying on GCC-specific behaviour and containing code that relies on undefined behaviour and a vocal community that complains to GCC whenever they change their interpretation of undefined behaviour. If we can support it, that's great. If supporting it comes at a cost to other users of LLVM (either in terms of worse code, or less error reporting), then that's something that will need to be evaluated carefully on its merits, which is what the discussion that was already in progress was doing.

More vague generalizations. I couldn't care less about who your
employers are. I just want LLVM to get more users; period. Am I
being overtly unreasonable, selfish, or demanding?

[1]: http://lists.cs.uiuc.edu/pipermail/llvmdev/2013-July/063799.html

There are many people with many years of experience with x86 on this mailing list, including compiler and assembler developers, kernel programmers, assembly-level performance experts, and instruction set architects. In addition to being familiar with LLVM’s design and philosophy, they also represent dozens of interested parties, both open source and corporate.

Nobody will deny that Linus is an x86 expert, but this list is full of them, and it is clear from the discussion that has been going on that not all of them are in agreement with him.

–Owen