Re: [webkit-dev] Git meeting notes

[Philippe Normand]

On Mon, 2010-04-12 at 13:59 -0700, Benjamin wrote:
> Quick and dirty notes from the Git discussion.
> 
> - April 12th 2010 / Garage 2 /  10:15 - 11:15
> 
> Who was there: Rim, Nokia, Qualcomm, Sony, Apple, Google, Gtk guys
> (sorry I forgot which company you were from)
> 

Igalia :-)

Philippe


signature.asc
Description: This is a digitally signed message part
___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Eric Seidel]

COMMIT~1 or COMMIT^ both are one commit prior to COMMIT.
COMMIT~2 or COMMIT^^ are both 2 commits prior.

On Tue, Apr 13, 2010 at 8:35 AM, Geoffrey Garen  wrote:
>> If Mark would add tags for the nightlies one could type "git tag --
>> contains=COMMIT" to figure out which tags include the version.
>
> This still doesn't scale to archived builds of individual revisions.
>
> On a related note, is there a git syntax for "one revision prior to X?" -- I 
> use that in SVN all the time.
>
> Geoff
> ___
> webkit-dev mailing list
> webkit-dev@lists.webkit.org
> http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
>
___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Geoffrey Garen]

> If Mark would add tags for the nightlies one could type "git tag --
> contains=COMMIT" to figure out which tags include the version.

This still doesn't scale to archived builds of individual revisions.

On a related note, is there a git syntax for "one revision prior to X?" -- I 
use that in SVN all the time.

Geoff
___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[David Levin]

On Tue, Apr 13, 2010 at 8:14 AM, Alexey Proskuryakov  wrote:

>
> 13.04.2010, в 00:10, Benjamin Meyer написал(а):
>
> >> One thing that wasn't mentioned at the meeting is that git doesn't seem
> to have the same monotonously increasing revision numbers as svn does. It
> will be a problem to replace REGRESSION(r12345) with
> REGRESSION(96c3b0300ccf16b64efc260c21c85ba9030f2e3a).
> >
> > Can you clarify what you mean?  Do you mean we have tools that expect a
> number and not a sha or something else?
>
>
> No, I don't think there are any tools that depend on this. It will just
> take away precious space in bug title, obscuring the actual problem
> description.
>

I don't see a problem with this. In fact, I would appreciate if you all now
called me dc86ed7c29395f2b0863967f5350dfca4151be9d which is a sha1 hash of
my full name plus my birth date and place.

Sincerely,
dc86ed7c29395f2b0863967f5350dfca4151be9d


>
> - WBR, Alexey Proskuryakov
>
> ___
> webkit-dev mailing list
> webkit-dev@lists.webkit.org
> http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
>
___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Alexey Proskuryakov]

13.04.2010, в 00:10, Benjamin Meyer написал(а):

>> One thing that wasn't mentioned at the meeting is that git doesn't seem to 
>> have the same monotonously increasing revision numbers as svn does. It will 
>> be a problem to replace REGRESSION(r12345) with 
>> REGRESSION(96c3b0300ccf16b64efc260c21c85ba9030f2e3a).
> 
> Can you clarify what you mean?  Do you mean we have tools that expect a 
> number and not a sha or something else?


No, I don't think there are any tools that depend on this. It will just take 
away precious space in bug title, obscuring the actual problem description.

- WBR, Alexey Proskuryakov

___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Maciej Stachowiak]

On Apr 12, 2010, at 9:44 PM, Maciej Stachowiak wrote:



On Apr 12, 2010, at 8:15 PM, Timothy Hatcher wrote:

Only the first 5-7 characters are needed to identify a single  
commit (enough of the hash prefix to be unique). So REGRESSION 
(96c3b0) vs REGRESSION(r12345).


How long will 7 characters last before you hit the birthday paradox?


And to answer my own question (because I'm a nerd), a 7 hex digit hash  
prefix has a 50% chance of at least one pair of revisions colliding  
when there are at least 19290 revs. So we're already past the birthday  
paradox threshold to be using only 7 digits (assuming the hashes are  
uniformly distributed).


This answer brought to you by Wikipedia and Wolfram|Alpha, favorites  
of pedantic nerds everywhere:


http://en.wikipedia.org/wiki/Birthday_problem#Cast_as_a_collision_problem
http://www.wolframalpha.com/input/?i=sqrt%282*2%5E%284*7%29*ln%281%2F%281-.5%29%29%29

Regards,
Maciej

___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Benjamin Meyer]

On Apr 12, 2010, at 6:28 PM, Alexey Proskuryakov wrote:

> One thing that wasn't mentioned at the meeting is that git doesn't seem to 
> have the same monotonously increasing revision numbers as svn does. It will 
> be a problem to replace REGRESSION(r12345) with 
> REGRESSION(96c3b0300ccf16b64efc260c21c85ba9030f2e3a).

Can you clarify what you mean?  Do you mean we have tools that expect a number 
and not a sha or something else?

> Similarly, it would be hard to tell if a given nightly build includes changes 
> made in a certain commit.

Each nightly would have a sha so you could list all of the sha's between the 
two and grep for the one you are looking for

git rev-list yesterday_sha..today_sha | grep certain_sha

Or more similar to svn would be to count the commits in the sha to get the 
increasing revision number.  Each daily could include this number and so you 
could count the number that certain_sha would be to see if it was between 
yesterday and today.

git rev-list sha | wc -l


-Benjamin Meyer
___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Benjamin Meyer]

On Apr 12, 2010, at 2:19 PM, William Siegrist wrote:

> 
> On Apr 12, 2010, at 1:59 PM, Benjamin wrote:
> 
>> 
>> - Server side there is a trac plugin we can use and sounded like there wasn't
>> 
> 
> I'm not sure what the last half of the above was supposed to say, but yes, 
> there is git support for Trac. You can see it on Buildbot's trac:
> 
> http://buildbot.net/trac/timeline
> 

Woops yah that was a half written sentence.  

Server side there is a trac plugin we can use.  From the discussions it sounded 
like there wasn't any server side major issues.

-Benjamin Meyer
___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Holger Freyther]

On Tuesday 13 April 2010 06:32:24 Alexey Proskuryakov wrote:
> I can't tell from documentation if git-bisect requires rebuilding each
> revision. This is something we don't currently have to do, given that
> there are nightlies to test with.
 
> But my question was somewhat different (Dave Kilzer told me about git
> bisect earlier today). How can I tell if a given nightly build includes
> changes made in a certain commit?

If Mark would add tags for the nightlies one could type "git tag --
contains=COMMIT" to figure out which tags include the version.

Regarding the monotonic number there are two things that I have used at 
different projects.

One is to use "git describe" which contains a mix of the commit-id, number of 
commits since the last tag. The example output from a different project would 
be "on-waves/0.3.96-8-ga4ebb46". It tells me the tag name, that I've 8 commits 
more than in the tag and the commit id of the eights commit.

The other option is, when assuming to never rebase the master, we can type git 
log --pretty=oneline | wc -l and we will end up with an increasing number.






___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Maciej Stachowiak]

On Apr 12, 2010, at 8:15 PM, Timothy Hatcher wrote:

Only the first 5-7 characters are needed to identify a single commit  
(enough of the hash prefix to be unique). So REGRESSION(96c3b0) vs  
REGRESSION(r12345).


How long will 7 characters last before you hit the birthday paradox?

Regards,
Maciej

___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[David Kilzer]

Sure, but you have to rebuild WebKit for each test during the git-bisect.  
Being able to use pre-built binaries is faster, especially if they're already 
downloaded.

Dave


From: Daniel Cheng 
>To: David Kilzer 
>Cc: Timothy Hatcher ; Alexey Proskuryakov 
>; WebKit Development 
>Sent: Mon, April 12, 2010 9:07:13 PM
>Subject: Re: [webkit-dev] Git meeting notes
>
>git-bisect can be used for a binary regression search.
>
>
>Daniel
>
>On Mon, Apr 12, 2010 at 9:05 PM, Daniel Cheng  wrote:
>
>git-bisect can be used for a binary regression search.
>>
>>
>>Daniel
>>
>>
>>
>>On Mon, Apr 12, 2010 at 8:45 PM, David Kilzer  wrote:
>>
>>>>>
>>>It's not immediately clear how to sort a list of commit hashes into a 
>>>sequential list, e.g., if using a list of nightly builds for a binary 
>>>regression search.
>>>
>>>>>>Perhaps there is a git command that sorts them for you given an existing 
>>>>>>repository?
>>>
>>>>>>Dave
>>>
>>>
>>>
>>>
>>>>>>On Mon, April 12, 2010 at 8:15:40 PM, Timothy Hatcher wrote:
>>>
>>>>>>> Only the first 5-7 characters are needed to identify a single commit 
>>>>>>> (enough of
>>>>>>> the hash prefix to be unique). So REGRESSION(96c3b0) vs
>>>>>>> REGRESSION(r12345).
>>>
>>>>>>On Apr 12, 2010, at 6:28 PM, Alexey Proskuryakov
>>>>>>> wrote:
>>>
>>>>>>> One thing that wasn't mentioned at the meeting is that git
>>>>>>> doesn't seem to have the same monotonously increasing revision numbers 
>>>>>>> as svn
>>>>>>> does. It will be a problem to replace REGRESSION(r12345) with
>>>>>>> REGRESSION(96c3b0300ccf16b64efc260c21c85ba9030f2e3a).
>>>
>>>>>>— Timothy
>>>>>>> Hatcher
>>>
>>>
>>>>>>___
>>>>>>webkit-dev
>>>>>>> mailing list
>>>
>>>> href="mailto:webkit-dev@lists.webkit.org";>webkit-dev@lists.webkit.org
>>>
>>>http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
>>>>>>___
>>>>>>webkit-dev mailing list
>>>webkit-dev@lists.webkit.org
>>>http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
>>>
>>
>___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Alexey Proskuryakov]

I can't tell from documentation if git-bisect requires rebuilding each 
revision. This is something we don't currently have to do, given that there are 
nightlies to test with.

But my question was somewhat different (Dave Kilzer told me about git bisect 
earlier today). How can I tell if a given nightly build includes changes made 
in a certain commit?

- WBR, Alexey Proskuryakov

12.04.2010, в 21:07, Daniel Cheng написал(а):

> (sent from the wrong email, sorry)
> git-bisect can be used for a binary regression search.
> 
> Daniel
> 
> On Mon, Apr 12, 2010 at 9:05 PM, Daniel Cheng  wrote:
> git-bisect can be used for a binary regression search.
> 
> Daniel
> 
> On Mon, Apr 12, 2010 at 8:45 PM, David Kilzer  wrote:
> It's not immediately clear how to sort a list of commit hashes into a 
> sequential list, e.g., if using a list of nightly builds for a binary 
> regression search.
> 
> Perhaps there is a git command that sorts them for you given an existing 
> repository?
> 
> Dave
> 
> 
> 
> On Mon, April 12, 2010 at 8:15:40 PM, Timothy Hatcher wrote:
> 
> > Only the first 5-7 characters are needed to identify a single commit 
> > (enough of
> > the hash prefix to be unique). So REGRESSION(96c3b0) vs
> > REGRESSION(r12345).
> 
> On Apr 12, 2010, at 6:28 PM, Alexey Proskuryakov
> > wrote:
> 
> > One thing that wasn't mentioned at the meeting is that git
> > doesn't seem to have the same monotonously increasing revision numbers as 
> > svn
> > does. It will be a problem to replace REGRESSION(r12345) with
> > REGRESSION(96c3b0300ccf16b64efc260c21c85ba9030f2e3a).
> 
> — Timothy
> > Hatcher
> 
> 
> ___
> webkit-dev
> > mailing list
> 
> > href="mailto:webkit-dev@lists.webkit.org";>webkit-dev@lists.webkit.org
> http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
> ___
> webkit-dev mailing list
> webkit-dev@lists.webkit.org
> http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
> 
> 


___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Daniel Cheng]

(sent from the wrong email, sorry)
git-bisect can be used for a binary regression search.

Daniel

On Mon, Apr 12, 2010 at 9:05 PM, Daniel Cheng  wrote:

> git-bisect can be used for a binary regression search.
>
> Daniel
>
> On Mon, Apr 12, 2010 at 8:45 PM, David Kilzer  wrote:
>
>> It's not immediately clear how to sort a list of commit hashes into a
>> sequential list, e.g., if using a list of nightly builds for a binary
>> regression search.
>>
>> Perhaps there is a git command that sorts them for you given an existing
>> repository?
>>
>> Dave
>>
>>
>>
>> On Mon, April 12, 2010 at 8:15:40 PM, Timothy Hatcher wrote:
>>
>> > Only the first 5-7 characters are needed to identify a single commit
>> (enough of
>> > the hash prefix to be unique). So REGRESSION(96c3b0) vs
>> > REGRESSION(r12345).
>>
>> On Apr 12, 2010, at 6:28 PM, Alexey Proskuryakov
>> > wrote:
>>
>> > One thing that wasn't mentioned at the meeting is that git
>> > doesn't seem to have the same monotonously increasing revision numbers
>> as svn
>> > does. It will be a problem to replace REGRESSION(r12345) with
>> > REGRESSION(96c3b0300ccf16b64efc260c21c85ba9030f2e3a).
>>
>> — Timothy
>> > Hatcher
>>
>>
>> ___
>> webkit-dev
>> > mailing list
>>
>> > href="mailto:webkit-dev@lists.webkit.org";>webkit-dev@lists.webkit.org
>> http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
>> ___
>> webkit-dev mailing list
>> webkit-dev@lists.webkit.org
>> http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
>>
>
>
___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[David Kilzer]

It's not immediately clear how to sort a list of commit hashes into a 
sequential list, e.g., if using a list of nightly builds for a binary 
regression search.

Perhaps there is a git command that sorts them for you given an existing 
repository?

Dave



On Mon, April 12, 2010 at 8:15:40 PM, Timothy Hatcher wrote:

> Only the first 5-7 characters are needed to identify a single commit (enough 
> of 
> the hash prefix to be unique). So REGRESSION(96c3b0) vs 
> REGRESSION(r12345).

On Apr 12, 2010, at 6:28 PM, Alexey Proskuryakov 
> wrote:

> One thing that wasn't mentioned at the meeting is that git 
> doesn't seem to have the same monotonously increasing revision numbers as svn 
> does. It will be a problem to replace REGRESSION(r12345) with 
> REGRESSION(96c3b0300ccf16b64efc260c21c85ba9030f2e3a).

— Timothy 
> Hatcher


___
webkit-dev 
> mailing list

> href="mailto:webkit-dev@lists.webkit.org";>webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Timothy Hatcher]

Only the first 5-7 characters are needed to identify a single commit (enough of 
the hash prefix to be unique). So REGRESSION(96c3b0) vs REGRESSION(r12345).

On Apr 12, 2010, at 6:28 PM, Alexey Proskuryakov wrote:

> One thing that wasn't mentioned at the meeting is that git doesn't seem to 
> have the same monotonously increasing revision numbers as svn does. It will 
> be a problem to replace REGRESSION(r12345) with 
> REGRESSION(96c3b0300ccf16b64efc260c21c85ba9030f2e3a).

— Timothy Hatcher


___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Alexey Proskuryakov]

12.04.2010, в 13:59, Benjamin написал(а):

> * Ran overtime on this meeting


One thing that wasn't mentioned at the meeting is that git doesn't seem to have 
the same monotonously increasing revision numbers as svn does. It will be a 
problem to replace REGRESSION(r12345) with 
REGRESSION(96c3b0300ccf16b64efc260c21c85ba9030f2e3a).

Similarly, it would be hard to tell if a given nightly build includes changes 
made in a certain commit.

Is there a good solution for these issues?

- WBR, Alexey Proskuryakov

___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Marc-Antoine Ruel]

2010/4/12 Peter Kasting 

> On Mon, Apr 12, 2010 at 1:59 PM, Benjamin  wrote:
>
>> - Google/Chrome has converted from svn to Git for some of chrome
>>
>
> This sounds like more than it actually is.  Chrome server-side is svn.
>  Client side there are some team members who are using git-svn and other git
> scripts.  I believe, but am not sure, that the number of team members using
> svn significantly outnumbers those using git-svn and similar.
>

To be clear, I said Chromium OS is using git. Some people are using Chromium
with git-svn  the same way
some people are using git-svn with
WebKit
.

M-A
___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[William Siegrist]

On Apr 12, 2010, at 1:59 PM, Benjamin wrote:

> 
> - Server side there is a trac plugin we can use and sounded like there wasn't
> 

I'm not sure what the last half of the above was supposed to say, but yes, 
there is git support for Trac. You can see it on Buildbot's trac:

http://buildbot.net/trac/timeline


-Bill

___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

Re: [webkit-dev] Git meeting notes

[Peter Kasting]

On Mon, Apr 12, 2010 at 1:59 PM, Benjamin  wrote:

> - Google/Chrome has converted from svn to Git for some of chrome
>

This sounds like more than it actually is.  Chrome server-side is svn.
 Client side there are some team members who are using git-svn and other git
scripts.  I believe, but am not sure, that the number of team members using
svn significantly outnumbers those using git-svn and similar.

PK
___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

[webkit-dev] Git meeting notes

[Benjamin]

Quick and dirty notes from the Git discussion.

- April 12th 2010 / Garage 2 /  10:15 - 11:15

Who was there: Rim, Nokia, Qualcomm, Sony, Apple, Google, Gtk guys
(sorry I forgot which company you were from)

Everyone that is present is using Git internally to some degree.
- Apple ~20% are using git-svn the rest on svn
- Google/Chrome has converted from svn to Git for some of chrome
- A bunch only on Git and push to a git repository
- Some git-svn users
- A few svn users
- Everyone else ~100% git-svn

The idea of putting up a survey or something to get a better statistic
was brought up.

One person mentioned that one person was using mercurial, but the
discussion was very much not "If we move to a distributed revision
control which one", but more "Our community is using Git to a large
degree do we want to move to it"

There was some preaching to the choir about how various groups were
using Git to be more efficient etc

It sounded like a bunch of people have tried Git in the past when it
was a less mature project and got a bad taste.  In particular on
Windows using the cygwin git which was big, awkward and slow and back
when the docs were very poor.

Git on Windows is slower then git on other platforms.

We should put together various docs
- Go through and cleanup and update the Git development wiki page
(update to point to msysgit on windows for example)
- Create a basic workflow page showing the ~10 commands that are
needed to use git for those who want to try out git
- Create a overview of workflows showing off what we do today and a
simple git workflow to show off the advantages
- Include in the document reasons why moving to the git v.s.
continuing only on git-svn has advantages (merges, sha's, distributed
nature)

- Server side there is a trac plugin we can use and sounded like there wasn't

- Access the svn server from the other side of the world can be slow,
git solves that, also mention that this could be solved via a svn
mirror

* Ran overtime on this meeting

Post meeting notes from a few final discussions
Server side Apple's IT  wont allow ssh connections so we would have to
use http push.  An alternative would be host the git server outside of
apple.

Conclusion: Not move to Git until more of our community is using it.
Improve docs in particular and revisit this at a future date.

Please correct or clarify any mistakes, thanks

-Benjamin Meyer (icefox)
___
webkit-dev mailing list
webkit-dev@lists.webkit.org
http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev