Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

classic Classic list List threaded Threaded
13 messages Options
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Douglas Gregor

On Feb 12, 2010, at 8:25 AM, David Greene wrote:

On Thursday 11 February 2010 18:17:33 Tanya Lattner wrote:
Just a reminder that the 2.7 code freeze is on Feb 21st.

All major changes should be committed approximately 1 week before the code
freeze to ensure adequate testing. Please do your part to keep the tree
stable in the days leading up to the code freeze.

Since the metadata stuff just settled recently, I like to ask for some time to
get the non-temporal stuff in.  This is really critical for our work here and
it would be nice to get this into 2.7.

Generally, I favor timed releases over feature releases, but... the Clang team would also like a little more time to prepare for the 2.7 release. Specifically, we propose to push back by 2 weeks, with the revised schedule being:

3/7 - Code Freeze (9PM PST)
3/13 - Pre-release1 released & community testing begins
3/20 - Pre-release1 testing ends
3/27 - Pre-release2 released & community testing begins 
4/3 - Pre-release2 testing ends
4/5 - Release

Why now?

Clang's C++ support is at an important transitional point: we can self-host a Debug build, and are starting to build significant C++ open source projects such as CMake, Firefox, Qt, and even parts of Boost. LLVM 2.7 is the perfect opportunity to enable Clang C++ support by default and announce to the open-source community that we now have something worth looking at. However, we have several known semantic analysis bugs and miscompiles that prevent self-hosting with optimization enabled, cause Firefox to crash on startup, etc. To advertise Clang C++ widely as part of 2.7 while these bugs remain would be embarrassing, but we feel that we can address the major problems with only a two-week slip in the schedule. 

Clang C++ only gets one big coming-out party (ever); a little more time will make a big difference.

- Doug

_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
dag
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

dag
On Friday 12 February 2010 17:02:41 Douglas Gregor wrote:

> Clang C++ only gets one big coming-out party (ever); a little more time
> will make a big difference.

Hooray!  Getting this into 2.7 would also be a great thing for OS
distributions as they could package 2.7 as a real development package.
 I know it would help another project I'm working on.

                                            -Dave

_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Daniel Dunbar
In reply to this post by Douglas Gregor
I am definitely in favor of this if it is ok with Tanya.

I hope to spend some time in the next few weeks on tracking down
miscompiles, and it would be great to get Clang to the
"early-but-usable-beta" stage so it makes sense to roll binaries for
2.7.

Tanya, I can also do the x86-32-pc-linux release testing if no one
else steps up.

 - Daniel

On Fri, Feb 12, 2010 at 3:02 PM, Douglas Gregor <[hidden email]> wrote:

>
> On Feb 12, 2010, at 8:25 AM, David Greene wrote:
>
> On Thursday 11 February 2010 18:17:33 Tanya Lattner wrote:
>
> Just a reminder that the 2.7 code freeze is on Feb 21st.
>
> All major changes should be committed approximately 1 week before the code
>
> freeze to ensure adequate testing. Please do your part to keep the tree
>
> stable in the days leading up to the code freeze.
>
> Since the metadata stuff just settled recently, I like to ask for some time
> to
> get the non-temporal stuff in.  This is really critical for our work here
> and
> it would be nice to get this into 2.7.
>
> Generally, I favor timed releases over feature releases, but... the Clang
> team would also like a little more time to prepare for the 2.7 release.
> Specifically, we propose to push back by 2 weeks, with the revised schedule
> being:
> 3/7 - Code Freeze (9PM PST)
> 3/13 - Pre-release1 released & community testing begins
> 3/20 - Pre-release1 testing ends
> 3/27 - Pre-release2 released & community testing begins
> 4/3 - Pre-release2 testing ends
> 4/5 - Release
> Why now?
> Clang's C++ support is at an important transitional point: we can self-host
> a Debug build, and are starting to build significant C++ open source
> projects such as CMake, Firefox, Qt, and even parts of Boost. LLVM 2.7 is
> the perfect opportunity to enable Clang C++ support by default and announce
> to the open-source community that we now have something worth looking
> at. However, we have several known semantic analysis bugs and miscompiles
> that prevent self-hosting with optimization enabled, cause Firefox to crash
> on startup, etc. To advertise Clang C++ widely as part of 2.7 while these
> bugs remain would be embarrassing, but we feel that we can address the major
> problems with only a two-week slip in the schedule.
> Clang C++ only gets one big coming-out party (ever); a little more time will
> make a big difference.
> - Doug
> _______________________________________________
> cfe-dev mailing list
> [hidden email]
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>
>

_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Chandler Carruth
On Fri, Feb 12, 2010 at 5:48 PM, Daniel Dunbar <[hidden email]> wrote:
> I am definitely in favor of this if it is ok with Tanya.
>
> I hope to spend some time in the next few weeks on tracking down
> miscompiles, and it would be great to get Clang to the
> "early-but-usable-beta" stage so it makes sense to roll binaries for
> 2.7.
>
> Tanya, I can also do the x86-32-pc-linux release testing if no one
> else steps up.

I've access to several machines for pc-linux testing, 32bit and 64
bit, amd and intel, as well as several distributions.

>
>  - Daniel
>
> On Fri, Feb 12, 2010 at 3:02 PM, Douglas Gregor <[hidden email]> wrote:
>>
>> On Feb 12, 2010, at 8:25 AM, David Greene wrote:
>>
>> On Thursday 11 February 2010 18:17:33 Tanya Lattner wrote:
>>
>> Just a reminder that the 2.7 code freeze is on Feb 21st.
>>
>> All major changes should be committed approximately 1 week before the code
>>
>> freeze to ensure adequate testing. Please do your part to keep the tree
>>
>> stable in the days leading up to the code freeze.
>>
>> Since the metadata stuff just settled recently, I like to ask for some time
>> to
>> get the non-temporal stuff in.  This is really critical for our work here
>> and
>> it would be nice to get this into 2.7.
>>
>> Generally, I favor timed releases over feature releases, but... the Clang
>> team would also like a little more time to prepare for the 2.7 release.
>> Specifically, we propose to push back by 2 weeks, with the revised schedule
>> being:
>> 3/7 - Code Freeze (9PM PST)
>> 3/13 - Pre-release1 released & community testing begins
>> 3/20 - Pre-release1 testing ends
>> 3/27 - Pre-release2 released & community testing begins
>> 4/3 - Pre-release2 testing ends
>> 4/5 - Release
>> Why now?
>> Clang's C++ support is at an important transitional point: we can self-host
>> a Debug build, and are starting to build significant C++ open source
>> projects such as CMake, Firefox, Qt, and even parts of Boost. LLVM 2.7 is
>> the perfect opportunity to enable Clang C++ support by default and announce
>> to the open-source community that we now have something worth looking
>> at. However, we have several known semantic analysis bugs and miscompiles
>> that prevent self-hosting with optimization enabled, cause Firefox to crash
>> on startup, etc. To advertise Clang C++ widely as part of 2.7 while these
>> bugs remain would be embarrassing, but we feel that we can address the major
>> problems with only a two-week slip in the schedule.
>> Clang C++ only gets one big coming-out party (ever); a little more time will
>> make a big difference.
>> - Doug
>> _______________________________________________
>> cfe-dev mailing list
>> [hidden email]
>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>
>>
>
> _______________________________________________
> LLVM Developers mailing list
> [hidden email]         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>

_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Daniel Dunbar
On Fri, Feb 12, 2010 at 5:52 PM, Chandler Carruth <[hidden email]> wrote:

> On Fri, Feb 12, 2010 at 5:48 PM, Daniel Dunbar <[hidden email]> wrote:
>> I am definitely in favor of this if it is ok with Tanya.
>>
>> I hope to spend some time in the next few weeks on tracking down
>> miscompiles, and it would be great to get Clang to the
>> "early-but-usable-beta" stage so it makes sense to roll binaries for
>> 2.7.
>>
>> Tanya, I can also do the x86-32-pc-linux release testing if no one
>> else steps up.
>
> I've access to several machines for pc-linux testing, 32bit and 64
> bit, amd and intel, as well as several distributions.

Great! I devolunteer for x86-32-pc-linux testing then! :)

Instead, I'll try and write some decent examples of using clang as a library.

 - Daniel

>>
>>  - Daniel
>>
>> On Fri, Feb 12, 2010 at 3:02 PM, Douglas Gregor <[hidden email]> wrote:
>>>
>>> On Feb 12, 2010, at 8:25 AM, David Greene wrote:
>>>
>>> On Thursday 11 February 2010 18:17:33 Tanya Lattner wrote:
>>>
>>> Just a reminder that the 2.7 code freeze is on Feb 21st.
>>>
>>> All major changes should be committed approximately 1 week before the code
>>>
>>> freeze to ensure adequate testing. Please do your part to keep the tree
>>>
>>> stable in the days leading up to the code freeze.
>>>
>>> Since the metadata stuff just settled recently, I like to ask for some time
>>> to
>>> get the non-temporal stuff in.  This is really critical for our work here
>>> and
>>> it would be nice to get this into 2.7.
>>>
>>> Generally, I favor timed releases over feature releases, but... the Clang
>>> team would also like a little more time to prepare for the 2.7 release.
>>> Specifically, we propose to push back by 2 weeks, with the revised schedule
>>> being:
>>> 3/7 - Code Freeze (9PM PST)
>>> 3/13 - Pre-release1 released & community testing begins
>>> 3/20 - Pre-release1 testing ends
>>> 3/27 - Pre-release2 released & community testing begins
>>> 4/3 - Pre-release2 testing ends
>>> 4/5 - Release
>>> Why now?
>>> Clang's C++ support is at an important transitional point: we can self-host
>>> a Debug build, and are starting to build significant C++ open source
>>> projects such as CMake, Firefox, Qt, and even parts of Boost. LLVM 2.7 is
>>> the perfect opportunity to enable Clang C++ support by default and announce
>>> to the open-source community that we now have something worth looking
>>> at. However, we have several known semantic analysis bugs and miscompiles
>>> that prevent self-hosting with optimization enabled, cause Firefox to crash
>>> on startup, etc. To advertise Clang C++ widely as part of 2.7 while these
>>> bugs remain would be embarrassing, but we feel that we can address the major
>>> problems with only a two-week slip in the schedule.
>>> Clang C++ only gets one big coming-out party (ever); a little more time will
>>> make a big difference.
>>> - Doug
>>> _______________________________________________
>>> cfe-dev mailing list
>>> [hidden email]
>>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>>
>>>
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> [hidden email]         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>>
>

_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Holger Schurig
In reply to this post by Douglas Gregor
> Clang C++ only gets one big coming-out party (ever); a little
> more time will make a big difference.

Hmm, wouldn't that speak against jamming this into 2.7 ?  Let 2.7
go out and don't bark at it's C++ status. That gives you a full
release-cycle to get C++ into some *REALLY* good state. For
example, a state where you can not just say "Oh, Firefox doesn't
crash anymore", but where are benchmarks for compilation time,
compilation memory consumption and execution speeds exists.

So, if there's only one coming-out party, make it a real party,
not some "me too" event!   :-)


--
http://www.holgerschurig.de
_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Maurice

On Mon, Feb 15, 2010 at 9:30 AM, Holger Schurig <[hidden email]> wrote:
> Clang C++ only gets one big coming-out party (ever); a little
> more time will make a big difference.

Hmm, wouldn't that speak against jamming this into 2.7 ?  Let 2.7
go out and don't bark at it's C++ status. That gives you a full
release-cycle to get C++ into some *REALLY* good state. For
example, a state where you can not just say "Oh, Firefox doesn't
crash anymore", but where are benchmarks for compilation time,
compilation memory consumption and execution speeds exists.

So, if there's only one coming-out party, make it a real party,
not some "me too" event!   :-)


I may have misunderstood, but I figured that Doug is suggesting to
get known issues fixed so that clang++ might be ready for
broader testing. 
More test coverage will shake out a lot of the remaining issues
at a quicker rate.

If the community contributes by increasing the test coverage
for clang++ then the core clang team could concentrate on fixing the
real the community would report.

I think Doug is suggesting that a bit more time will enable more
community participation in shaking out remaining issues during
the next cycle.

Would clang++ not be ready sooner in this way?

Cheers,
Maurice



_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Óscar Fuentes
Maurice Gittens
<[hidden email]> writes:

> I may have misunderstood, but I figured that Doug is suggesting to get
> known issues fixed so that clang++ might be ready for broader testing.
> More test coverage will shake out a lot of the remaining issues at a
> quicker rate.
>
> If the community contributes by increasing the test coverage
> for clang++ then the core clang team could concentrate on fixing the
> real the community would report.
>
> I think Doug is suggesting that a bit more time will enable more
> community participation in shaking out remaining issues during
> the next cycle.
>
> Would clang++ not be ready sooner in this way?

This reminds me of the KDE 4.x debacle. If clang++ is to be released in
"almost working" status, expecting that the users will play the role of
QA engineers, please advertise it as such. Prominently.

Signed: a KDE user.

_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Holger Schurig
In reply to this post by Maurice
> I may have misunderstood, but I figured that Doug is suggesting
> to get known issues fixed so that clang++ might be ready for
> broader testing.

I read that clang-c++ can "barely" compile CMake, Firefox, Qt,
and even parts of Boost. Just with a debug compile (where the
debug info isn't the essential info, but the lack of
optimizations).

Now this two week delay would be used to get the optimizations
hopefully into a state where they don't suck anymore, e.g. where
Firefox doesn't crash when compiled with -O2. While this is
certainly nice and a pre-requisite for the "outcoming party",
it's the minimum that is necessary.

What I propose is to continue with LLVM 2.7 release as planned,
without c++ support. This way c++ get's some months for
development time. Those months can then be used to make
clang-c++ really nice, not just "barely usable, three apps
compiled and didn't crash immediately".

When LLVM 2.8 comes out, it's C++ support can be really nice, and
*THAT* will give a nice PR effect. An outcome party that is
worth visiting. Or more like a baptism where both the people in
the church and the angles in the heaves rejoice.

--
http://www.holgerschurig.de
_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Douglas Gregor
In reply to this post by Holger Schurig

On Feb 15, 2010, at 12:30 AM, Holger Schurig wrote:

>> Clang C++ only gets one big coming-out party (ever); a little
>> more time will make a big difference.
>
> Hmm, wouldn't that speak against jamming this into 2.7 ?  Let 2.7
> go out and don't bark at it's C++ status. That gives you a full
> release-cycle to get C++ into some *REALLY* good state.

Where's the excitement in that? Another release cycle is 6 months away, and if past progress is any indication, we'll have gotten Clang C++ to the point where it "just works" for the vast majority of C++ programs [1]. At that point, it will also be 9 months after our first "self-host", which is an eternity in open-source time.

No, 2.7 is the right time-frame for the Clang C++ announcement: we're not done, but we have enough to show to gain attention and draw in contributors, and we have some momentum now from having just announced a successful self-host.

> For
> example, a state where you can not just say "Oh, Firefox doesn't
> crash anymore", but where are benchmarks for compilation time,
> compilation memory consumption and execution speeds exists.

Those benchmarks are trivial to produce, and while we won't have time to further optimize for any of them, we will be able to establish a baseline [2] in 2.7 and improve from there.

        - Doug

[2] Compilation memory consumption and compilation time already look better than GCC, along with clearer diagnostics and Clang's other benefits, so we'll have something to say already.
_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Ted Kremenek
On Feb 15, 2010, at 7:49 AM, Douglas Gregor wrote:

>> Hmm, wouldn't that speak against jamming this into 2.7 ?  Let 2.7
>> go out and don't bark at it's C++ status. That gives you a full
>> release-cycle to get C++ into some *REALLY* good state.
>
> Where's the excitement in that? Another release cycle is 6 months away, and if past progress is any indication, we'll have gotten Clang C++ to the point where it "just works" for the vast majority of C++ programs [1]. At that point, it will also be 9 months after our first "self-host", which is an eternity in open-source time.
>
> No, 2.7 is the right time-frame for the Clang C++ announcement: we're not done, but we have enough to show to gain attention and draw in contributors, and we have some momentum now from having just announced a successful self-host.

I agree with Doug.  Clang C++ support has reached the maturity where it is very interesting to the open source community.  That's not the same as advocating that people switch over to using it as their default C++ compiler.  I agree that we should make the distinction very clear.

Having general testing by early adopters (in a 2.7 release) will shake out a lot of the issues in Clang C++, making the 2.8 release all the more better.  Beyond correctness issues, we know there will be performance issues to address as well, and real users will help identify those as well.

In the 2.8 time frame, if I were a casual user considering using Clang for compiling my C++ projects, I'd have a lot more confidence in that release knowing that it had been tested by others.  Further, people who use LLVM+Clang to do interesting stuff will care very much about Clang C++, even though it is still a WIP.
_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Tanya Lattner
In reply to this post by Douglas Gregor

On Feb 12, 2010, at 3:02 PM, Douglas Gregor wrote:


On Feb 12, 2010, at 8:25 AM, David Greene wrote:

On Thursday 11 February 2010 18:17:33 Tanya Lattner wrote:
Just a reminder that the 2.7 code freeze is on Feb 21st.

All major changes should be committed approximately 1 week before the code
freeze to ensure adequate testing. Please do your part to keep the tree
stable in the days leading up to the code freeze.

Since the metadata stuff just settled recently, I like to ask for some time to
get the non-temporal stuff in.  This is really critical for our work here and
it would be nice to get this into 2.7.

Generally, I favor timed releases over feature releases, but... the Clang team would also like a little more time to prepare for the 2.7 release. Specifically, we propose to push back by 2 weeks, with the revised schedule being:

3/7 - Code Freeze (9PM PST)
3/13 - Pre-release1 released & community testing begins
3/20 - Pre-release1 testing ends
3/27 - Pre-release2 released & community testing begins 
4/3 - Pre-release2 testing ends
4/5 - Release

After reading through the responses, I think that it makes sense to postpone this for a couple weeks. I'll update the schedule.

I'd like to reiterate that we do not normally push back releases for new features. However, Clang is on the verge of hitting a major milestone and I'll make an exception in this case. I do not see any negatives to waiting a couple extra weeks.

Thanks,
Tanya

_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
Reply | Threaded
Open this post in threaded view
|

Re: [LLVMdev] Reminder: 2.7 code freeze in 1.5 weeks

Douglas Gregor

On Feb 15, 2010, at 11:19 AM, Tanya Lattner wrote:


On Feb 12, 2010, at 3:02 PM, Douglas Gregor wrote:


On Feb 12, 2010, at 8:25 AM, David Greene wrote:

On Thursday 11 February 2010 18:17:33 Tanya Lattner wrote:
Just a reminder that the 2.7 code freeze is on Feb 21st.

All major changes should be committed approximately 1 week before the code
freeze to ensure adequate testing. Please do your part to keep the tree
stable in the days leading up to the code freeze.

Since the metadata stuff just settled recently, I like to ask for some time to
get the non-temporal stuff in.  This is really critical for our work here and
it would be nice to get this into 2.7.

Generally, I favor timed releases over feature releases, but... the Clang team would also like a little more time to prepare for the 2.7 release. Specifically, we propose to push back by 2 weeks, with the revised schedule being:

3/7 - Code Freeze (9PM PST)
3/13 - Pre-release1 released & community testing begins
3/20 - Pre-release1 testing ends
3/27 - Pre-release2 released & community testing begins 
4/3 - Pre-release2 testing ends
4/5 - Release

After reading through the responses, I think that it makes sense to postpone this for a couple weeks. I'll update the schedule.

I'd like to reiterate that we do not normally push back releases for new features. However, Clang is on the verge of hitting a major milestone and I'll make an exception in this case. I do not see any negatives to waiting a couple extra weeks.

Thank you!

- Doug

_______________________________________________
cfe-dev mailing list
[hidden email]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev