You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@couchdb.apache.org by Benoit Chesneau <bc...@gmail.com> on 2014/01/20 13:07:31 UTC
we need more reviews
Hi All,
Over the last few months, we have had a pretty good track in commits --
which is great! However I think we can do better working as a team on
this. I am sure others also feel the same way. For example:
- Some code is still landing in master with little or absence of review
in JIRA nor the mailing list
- There is insufficient discussion, sometimes a short word on IRC or out
of sight of the mailing list. Some of the discussions feel more like a
monologue, or don't reach a satisfactory conclusion or clear decision.
For example, the last discussion about metadata never reached a
conclusion and as a result, no changes have been made.
I really want to find a way to fix that. It is our responsibility to fix
that.
The project should count as a real team work.
A team is not a collection of individuals working on their side
where`git commit` or `jira close` are the only meeting points. That also
opens the doors to more conflicts, since people have to react against a
thing that have been done. We should discuss and review more the code
(and a review should be answered). Actually I do think that no code
should land in master without at least one or more (depending how
sensitive the patch is) public review. But we may be less strict. At
least nothing should come without a context. It’s also true for the
commits messages. There may be a discussion on IRC that explain that but
IRC is not a reliable source. People should understand the why and the
what for without having to read the code and should understand the
direction and eventually discuss it before it is taken.
All of these only require some guidelines. But we should start to
enforce them and make them public. Developers should act together
again.
In term of action I think we could:
- have review system with votes on the patch before inclusion in the
master (as in the apache https project). Many projects are working
like this and this is part of their success. We should of course explain
why a review is important and make sure that reviews are not transformed
in a fight between some egos.
- have some guidelines on how to discuss features
- make it clear that IRC is not the place where decisions are took and
enforce the ML usage
- make sure that people are writing good commit messages. Some projects
(erlang. linux are coming to my mind) make sure to fix them before
they land in maint or master branches.
Thoughts?
- benoit
NB: this a repost on public @dev mailing-list from the PMC mailing-list.
Re: we need more reviews
Posted by Andy Wenk <an...@nms.de>.
On 20 January 2014 14:11, Benoit Chesneau <bc...@gmail.com> wrote:
> On Mon, Jan 20, 2014 at 1:46 PM, Andy Wenk <an...@nms.de> wrote:
>
>> awesome to start this discussion. It was just some hours ago, that Hank
>> Knight exactly asked these questions. I am +1 on defining guidelines to
>> describe the process. It would not be the biggest problem to install
>> gerrit
>> (https://code.google.com/p/gerrit/) as a review system somewhere but I
>> would like to ask if it is also ok to use the PR queue like Dale wrote
>> above. I think the process is well known and learned. The only big
>> difference is that it is still possible to push to master without review
>> and PR's. But I don't see a problem to establish a review process with
>> PR's
>> when we define easy guidelines.
>>
>
> I looked recently at Gerrit for the refuge project and really liked what I
> see, with the github bridge it would offer a cool solution. Also gerrit is
> already in use in some apache projects I think.
>
> A word to git commit messages. If you want to read really informative
>> commit messages as an example, juts look at the PostgreSQL project.
>> Especially Tom Lane is writing awesome messages. So I am +100 to
>> informative and meaningful commit messages.
>>
>
> Do you have some example that you could link? I wonder if for the
> guideline we couldn't collect some examples from different projects.
>
yes sure. Here is the summary of the latest PostgreSQL commits:
http://git.postgresql.org/gitweb/?p=postgresql.git;a=summary
and here one (just out of many) from Tom:
http://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=115f414124e71749d2d8f512e469ca63bc2166e5
I don't mean that we should write a novel in the commit message but when I
speak about commit messages, I always bring the example of Tom (and others
at PG) and state, that one could write an awesome documentation simply out
of the commit messages :)
A good way between short messages and novels would be great I think ...
Cheers
Andy
--
Andy Wenk
Hamburg - Germany
RockIt!
http://www.couchdb-buch.de
http://www.pg-praxisbuch.de
GPG fingerprint: C044 8322 9E12 1483 4FEC 9452 B65D 6BE3 9ED3 9588
https://people.apache.org/keys/committer/andywenk.asc
Re: we need more reviews
Posted by Benoit Chesneau <bc...@gmail.com>.
On Mon, Jan 20, 2014 at 1:46 PM, Andy Wenk <an...@nms.de> wrote:
>
>
> awesome to start this discussion. It was just some hours ago, that Hank
> Knight exactly asked these questions. I am +1 on defining guidelines to
> describe the process. It would not be the biggest problem to install gerrit
> (https://code.google.com/p/gerrit/) as a review system somewhere but I
> would like to ask if it is also ok to use the PR queue like Dale wrote
> above. I think the process is well known and learned. The only big
> difference is that it is still possible to push to master without review
> and PR's. But I don't see a problem to establish a review process with PR's
> when we define easy guidelines.
>
I looked recently at Gerrit for the refuge project and really liked what I
see, with the github bridge it would offer a cool solution. Also gerrit is
already in use in some apache projects I think.
>
> A word to git commit messages. If you want to read really informative
> commit messages as an example, juts look at the PostgreSQL project.
> Especially Tom Lane is writing awesome messages. So I am +100 to
> informative and meaningful commit messages.
>
Do you have some example that you could link? I wonder if for the guideline
we couldn't collect some examples from different projects.
- benoit
Re: we need more reviews
Posted by Andy Wenk <an...@nms.de>.
On 20 January 2014 13:26, Dale Harvey <da...@arandomurl.com> wrote:
> I fully agree, its something I mentioned at the couchdb conf in vancouver,
> a review first system encourages contributions and has multiple benefits
>
> * At least 2 people look at the code, less likely to push silly mistakes
> * Can codify and practice review rules
> * Its much easier to view the current activity in the project
> * Can bring up points of discussion before its too late
>
> But I think the most important thing is that it removes the burden of
> responsibility from the committer to the project as a whole, also hugely
> beneficial is that it makes it particularly obvious when a patch has reach
> a stalemate and forces someone to make the call.
>
> For reference on PouchDB every committer is trusted to push code, nobody
> (including myself) pushes their own code to master, it goes in the PR queue
> and gets a +1 from any other committer (who will usually push it), thats
> essentially the same process we use at mozilla and coming to think of it
> any other project I have worked on, any commiter has the ability to -1 a
> patch at which point they give a reason and usually some solution gets
> agreed on
>
>
> On 20 January 2014 12:07, Benoit Chesneau <bc...@gmail.com> wrote:
>
> > Hi All,
> >
> > Over the last few months, we have had a pretty good track in commits --
> > which is great! However I think we can do better working as a team on
> > this. I am sure others also feel the same way. For example:
> >
> > - Some code is still landing in master with little or absence of review
> > in JIRA nor the mailing list
> >
> > - There is insufficient discussion, sometimes a short word on IRC or out
> > of sight of the mailing list. Some of the discussions feel more like a
> > monologue, or don't reach a satisfactory conclusion or clear decision.
> > For example, the last discussion about metadata never reached a
> > conclusion and as a result, no changes have been made.
> >
> > I really want to find a way to fix that. It is our responsibility to fix
> > that.
> > The project should count as a real team work.
> >
> > A team is not a collection of individuals working on their side
> > where`git commit` or `jira close` are the only meeting points. That also
> > opens the doors to more conflicts, since people have to react against a
> > thing that have been done. We should discuss and review more the code
> > (and a review should be answered). Actually I do think that no code
> > should land in master without at least one or more (depending how
> > sensitive the patch is) public review. But we may be less strict. At
> > least nothing should come without a context. It’s also true for the
> > commits messages. There may be a discussion on IRC that explain that but
> > IRC is not a reliable source. People should understand the why and the
> > what for without having to read the code and should understand the
> > direction and eventually discuss it before it is taken.
> >
> > All of these only require some guidelines. But we should start to
> > enforce them and make them public. Developers should act together
> > again.
> >
> > In term of action I think we could:
> >
> > - have review system with votes on the patch before inclusion in the
> > master (as in the apache https project). Many projects are working
> > like this and this is part of their success. We should of course explain
> > why a review is important and make sure that reviews are not transformed
> > in a fight between some egos.
> > - have some guidelines on how to discuss features
> > - make it clear that IRC is not the place where decisions are took and
> > enforce the ML usage
> > - make sure that people are writing good commit messages. Some projects
> > (erlang. linux are coming to my mind) make sure to fix them before
> > they land in maint or master branches.
> >
> > Thoughts?
> >
> >
> > - benoit
> >
> > NB: this a repost on public @dev mailing-list from the PMC mailing-list.
> >
>
awesome to start this discussion. It was just some hours ago, that Hank
Knight exactly asked these questions. I am +1 on defining guidelines to
describe the process. It would not be the biggest problem to install gerrit
(https://code.google.com/p/gerrit/) as a review system somewhere but I
would like to ask if it is also ok to use the PR queue like Dale wrote
above. I think the process is well known and learned. The only big
difference is that it is still possible to push to master without review
and PR's. But I don't see a problem to establish a review process with PR's
when we define easy guidelines.
A word to git commit messages. If you want to read really informative
commit messages as an example, juts look at the PostgreSQL project.
Especially Tom Lane is writing awesome messages. So I am +100 to
informative and meaningful commit messages.
Cheers
Andy
--
Andy Wenk
Hamburg - Germany
RockIt!
http://www.couchdb-buch.de
http://www.pg-praxisbuch.de
GPG fingerprint: C044 8322 9E12 1483 4FEC 9452 B65D 6BE3 9ED3 9588
https://people.apache.org/keys/committer/andywenk.asc
Re: we need more reviews
Posted by Benoit Chesneau <bc...@gmail.com>.
On Mon, Jan 20, 2014 at 1:26 PM, Dale Harvey <da...@arandomurl.com> wrote:
>
>
> But I think the most important thing is that it removes the burden of
> responsibility from the committer to the project as a whole, also hugely
> beneficial is that it makes it particularly obvious when a patch has reach
> a stalemate and forces someone to make the call.
>
> For reference on PouchDB every committer is trusted to push code, nobody
> (including myself) pushes their own code to master, it goes in the PR queue
> and gets a +1 from any other committer (who will usually push it), thats
> essentially the same process we use at mozilla and coming to think of it
> any other project I have worked on, any commiter has the ability to -1 a
> patch at which point they give a reason and usually some solution gets
> agreed on
>
>
>
+1
Re: we need more reviews
Posted by Adam Kocoloski <ko...@apache.org>.
On Jan 22, 2014, at 6:47 AM, Noah Slater <ns...@apache.org> wrote:
> My first comment is: if we want more reviews, let's have more committers.
>
> We double our committer base in 2013, and the results look like this:
>
> https://www.ohloh.net/p/couchdb/analyses/latest/languages_summary
>
> And I see comments like this on StackOverflow:
>
> "I've recently noticed that Couch DB is back in heavy development."
>
> So I think we should continue to aggressively recruit more committers
> to the project in 2014. Excelsior!
>
> However, as for the way we do reviews...
>
> Infra ticket about Gerrit
> https://issues.apache.org/jira/browse/INFRA-2205
>
> Effort seems to be mothballed. But I'm sure we could restart it if we
> were serious.
>
> However, we could use this:
>
> https://reviews.apache.org/groups/hive/
>
> It is well integrated with Apache infrastructure already, sends mails
> to the mailing list, and so on.
>
> Happy to request an instance and we can experiment with it if we like.
> If it doesn't work out, we stop using it. No harm. Could make it
> entirely voluntary until we figure out a workflow that we all like.
>
> Should I do that?
+1 to going with the flow at Apache on this one, and making it entirely voluntary at first.
Adam
Re: we need more reviews
Posted by Noah Slater <ns...@apache.org>.
Done: https://issues.apache.org/jira/browse/INFRA-7254
On 23 January 2014 14:12, Noah Slater <ns...@apache.org> wrote:
> Okay, I'll leave this a few days and then request a Review Board
> instance so we can experiment with it.
>
> Will be entirely voluntary! If it starts to become useful, we can
> codify how we expect it to be used. Until that point, we can just send
> our patches to Review Board and ask for a review on the mailing list.
>
> On 23 January 2014 03:30, Mutton, James <jm...@akamai.com> wrote:
>> On 1/22/14 3:07 PM, "Russell Branca" <ch...@apache.org> wrote:
>>
>>
>>>Huge +1 to more review.
>>>
>>>Let's also setup some code guidelines for Erlang/Javascript/C/HTML so we
>>>have an authoritative list of rules to follow to ensure code consistency,
>>>and similarly, let's get some guidelines for git commits messages as well,
>>>Tim Pope's article comes to mind:
>>>http://tbaggery.com/2008/04/19/a-note-about-git-commit-messages.html.
>>>
>>>Whichever review tool we decide to use, we should use it universally. We
>>>should figure out a way to do github integration so that PR's show up in
>>>the tool, and then we update the PR with a link to the relevant review
>>>page. We also need a good way to initiate the analogue of a PR in whatever
>>>tool we use. One of the complications with the review process today, is
>>>that we're all accustomed to doing review through github PR's, but we can
>>>use github PR's for branches who exist solely on the ASF origin. I imagine
>>>this won't be an issue with Hive or Gerrit, but we should verify that
>>>we're
>>>not required to initiate code review from third party origins.
>>>
>>>On a side note, one of my personal nits with github PR's, that I hope
>>>might
>>>be resolved with one of these tools, is losing comments when the code
>>>changes. For instance, if I do a review and make 10 comments that each
>>>represent a line item that needs to be addressed, I would like to see a
>>>task list on the review indicating there are unfinished items to be
>>>addressed, basically a way to make a set of checkboxes necessary to
>>>complete the review. I find it odd that github doesn't do this, even more
>>>so by the fact that changing code to address comments will often mess with
>>>the commenting and hide the messages you posted, making it challenging to
>>>decipher whether the review items have been addressed. This is by no means
>>>a requirement to using any approach, just hoping that someone knows of a
>>>good way to approach this problem in whatever tool we use.
>>>
>>>
>>>-Russell
>>>
>>>
>>>On Wed, Jan 22, 2014 at 6:06 AM, Benoit Chesneau
>>><bc...@gmail.com>wrote:
>>>
>>>> On Wed, Jan 22, 2014 at 12:47 PM, Noah Slater <ns...@apache.org>
>>>>wrote:
>>>>
>>>> > My first comment is: if we want more reviews, let's have more
>>>>committers.
>>>> >
>>>> > We double our committer base in 2013, and the results look like this:
>>>> >
>>>> > https://www.ohloh.net/p/couchdb/analyses/latest/languages_summary
>>>> >
>>>> > And I see comments like this on StackOverflow:
>>>> >
>>>> > "I've recently noticed that Couch DB is back in heavy
>>>>development."
>>>>
>>>>
>>>> > So I think we should continue to aggressively recruit more committers
>>>> > to the project in 2014. Excelsior!
>>>> >
>>>>
>>>> Welcoming new committers in the project is certainly a good thing but I
>>>> think it's orthogonal to the need of more review and team work. We
>>>>should
>>>> find a good workflow now while we are still a limited number of
>>>>committers
>>>> because It will be harder to enforce anything on a larger team. We
>>>>should
>>>> settle on some guidelines now. It will also help us to attract more
>>>>people
>>>> IMO.
>>>>
>>>>
>>>> >
>>>> > However, as for the way we do reviews...
>>>> >
>>>> > Infra ticket about Gerrit
>>>> > https://issues.apache.org/jira/browse/INFRA-2205
>>>> >
>>>> > Effort seems to be mothballed. But I'm sure we could restart it if we
>>>> > were serious.
>>>> >
>>>> > However, we could use this:
>>>> >
>>>> > https://reviews.apache.org/groups/hive/
>>>> >
>>>> > It is well integrated with Apache infrastructure already, sends mails
>>>> > to the mailing list, and so on.
>>>> >
>>>> > Happy to request an instance and we can experiment with it if we like.
>>>> > If it doesn't work out, we stop using it. No harm. Could make it
>>>> > entirely voluntary until we figure out a workflow that we all like.
>>>> >
>>>> > Should I do that?
>>>> >
>>>>
>>>> I thought gerrit was already integrated but any tool that could help us
>>>>to
>>>> make the review is OK for me. How hive compares to gerrit? Could we also
>>>> plug it with github like gerrit [1] ?
>>>>
>>>> - benoît
>>>>
>>>> [1] https://gerrit.googlesource.com/plugins/github/ (and some others)
>>>>
>>>> >
>>>> > On 20 January 2014 15:30, Nick North <no...@gmail.com> wrote:
>>>> > > On 20 January 2014 12:26, Dale Harvey <da...@arandomurl.com> wrote:
>>>> > >
>>>> > >> I fully agree, its something I mentioned at the couchdb conf in
>>>> > vancouver,
>>>> > >> a review first system encourages contributions and has multiple
>>>> benefits
>>>> > >>
>>>> > >> * At least 2 people look at the code, less likely to push silly
>>>> > mistakes
>>>> > >> * Can codify and practice review rules
>>>> > >> * Its much easier to view the current activity in the project
>>>> > >> * Can bring up points of discussion before its too late
>>>> > >>
>>>> > >> But I think the most important thing is that it removes the burden
>>>>of
>>>> > >> responsibility from the committer to the project as a whole, also
>>>> hugely
>>>> > >> beneficial is that it makes it particularly obvious when a patch
>>>>has
>>>> > reach
>>>> > >> a stalemate and forces someone to make the call.
>>>> > >>
>>>> > >> For reference on PouchDB every committer is trusted to push code,
>>>> nobody
>>>> > >> (including myself) pushes their own code to master, it goes in the
>>>>PR
>>>> > queue
>>>> > >> and gets a +1 from any other committer (who will usually push it),
>>>> thats
>>>> > >> essentially the same process we use at mozilla and coming to think
>>>>of
>>>> it
>>>> > >> any other project I have worked on, any commiter has the ability to
>>>> -1 a
>>>> > >> patch at which point they give a reason and usually some solution
>>>>gets
>>>> > >> agreed on
>>>> > >>
>>>> > >
>>>> > > I like this system, with one small quibble about responsibility. I
>>>> don't
>>>> > > think it should be seen as removing the burden of responsibility
>>>>from
>>>> the
>>>> > > committer to the project as a whole. It becomes easy then for
>>>>everyone,
>>>> > > including the committer, to think someone else will find bugs, and
>>>> no-one
>>>> > > puts in enough effort. I'd say it is still primarily the
>>>>responsibility
>>>> > of
>>>> > > the committer to ensure that code is error free, but that at least
>>>>one
>>>> > > person who knows that area of the code base should sign off on it.
>>>>Some
>>>> > > spreading of responsibility is good, but too much can actually lead
>>>>to
>>>> a
>>>> > > decline in quality.
>>>> > >
>>>> > >
>>>> > > Nick
>>>> >
>>>> >
>>>> >
>>>> > --
>>>> > Noah Slater
>>>> > https://twitter.com/nslater
>>>> >
>>>>
>>
>>
>> +1 to reviews and getting code guidelines. Nice to see that Apache has a
>> reviewboard instance available (https://reviews.apache.org/groups/ hive
>> link). I've spent a lot of time using reviewboard, it's got a good UI and
>> API. Reviews are submitted using a separate command line tool
>> (post-review) to actually post the review, but it lets you make multiple
>> updates from feedback and allows others to see the diffs on the same
>> review so you can see how a change evolves. You can leave comments or put
>> a "ship-it" tag to +1 the commit, then from a distance it's possible to
>> see all the reviews outstanding and how many votes, if any, they've
>> received. It's friendly for non-blocking reviews too, since the code
>> review process is separate. It also lets you can work on reviews a little
>> at a time and save them in between instead of having to devote a block of
>> time to code reviews (I do this sometimes on big changes).
>>
>> Garrit wants to gate the source-tree if I remember correctly so is
>> possibly less friendly with remote repositories.
>>
>> </JamesM>
>>
>
>
>
> --
> Noah Slater
> https://twitter.com/nslater
--
Noah Slater
https://twitter.com/nslater
Re: we need more reviews
Posted by Noah Slater <ns...@apache.org>.
Okay, I'll leave this a few days and then request a Review Board
instance so we can experiment with it.
Will be entirely voluntary! If it starts to become useful, we can
codify how we expect it to be used. Until that point, we can just send
our patches to Review Board and ask for a review on the mailing list.
On 23 January 2014 03:30, Mutton, James <jm...@akamai.com> wrote:
> On 1/22/14 3:07 PM, "Russell Branca" <ch...@apache.org> wrote:
>
>
>>Huge +1 to more review.
>>
>>Let's also setup some code guidelines for Erlang/Javascript/C/HTML so we
>>have an authoritative list of rules to follow to ensure code consistency,
>>and similarly, let's get some guidelines for git commits messages as well,
>>Tim Pope's article comes to mind:
>>http://tbaggery.com/2008/04/19/a-note-about-git-commit-messages.html.
>>
>>Whichever review tool we decide to use, we should use it universally. We
>>should figure out a way to do github integration so that PR's show up in
>>the tool, and then we update the PR with a link to the relevant review
>>page. We also need a good way to initiate the analogue of a PR in whatever
>>tool we use. One of the complications with the review process today, is
>>that we're all accustomed to doing review through github PR's, but we can
>>use github PR's for branches who exist solely on the ASF origin. I imagine
>>this won't be an issue with Hive or Gerrit, but we should verify that
>>we're
>>not required to initiate code review from third party origins.
>>
>>On a side note, one of my personal nits with github PR's, that I hope
>>might
>>be resolved with one of these tools, is losing comments when the code
>>changes. For instance, if I do a review and make 10 comments that each
>>represent a line item that needs to be addressed, I would like to see a
>>task list on the review indicating there are unfinished items to be
>>addressed, basically a way to make a set of checkboxes necessary to
>>complete the review. I find it odd that github doesn't do this, even more
>>so by the fact that changing code to address comments will often mess with
>>the commenting and hide the messages you posted, making it challenging to
>>decipher whether the review items have been addressed. This is by no means
>>a requirement to using any approach, just hoping that someone knows of a
>>good way to approach this problem in whatever tool we use.
>>
>>
>>-Russell
>>
>>
>>On Wed, Jan 22, 2014 at 6:06 AM, Benoit Chesneau
>><bc...@gmail.com>wrote:
>>
>>> On Wed, Jan 22, 2014 at 12:47 PM, Noah Slater <ns...@apache.org>
>>>wrote:
>>>
>>> > My first comment is: if we want more reviews, let's have more
>>>committers.
>>> >
>>> > We double our committer base in 2013, and the results look like this:
>>> >
>>> > https://www.ohloh.net/p/couchdb/analyses/latest/languages_summary
>>> >
>>> > And I see comments like this on StackOverflow:
>>> >
>>> > "I've recently noticed that Couch DB is back in heavy
>>>development."
>>>
>>>
>>> > So I think we should continue to aggressively recruit more committers
>>> > to the project in 2014. Excelsior!
>>> >
>>>
>>> Welcoming new committers in the project is certainly a good thing but I
>>> think it's orthogonal to the need of more review and team work. We
>>>should
>>> find a good workflow now while we are still a limited number of
>>>committers
>>> because It will be harder to enforce anything on a larger team. We
>>>should
>>> settle on some guidelines now. It will also help us to attract more
>>>people
>>> IMO.
>>>
>>>
>>> >
>>> > However, as for the way we do reviews...
>>> >
>>> > Infra ticket about Gerrit
>>> > https://issues.apache.org/jira/browse/INFRA-2205
>>> >
>>> > Effort seems to be mothballed. But I'm sure we could restart it if we
>>> > were serious.
>>> >
>>> > However, we could use this:
>>> >
>>> > https://reviews.apache.org/groups/hive/
>>> >
>>> > It is well integrated with Apache infrastructure already, sends mails
>>> > to the mailing list, and so on.
>>> >
>>> > Happy to request an instance and we can experiment with it if we like.
>>> > If it doesn't work out, we stop using it. No harm. Could make it
>>> > entirely voluntary until we figure out a workflow that we all like.
>>> >
>>> > Should I do that?
>>> >
>>>
>>> I thought gerrit was already integrated but any tool that could help us
>>>to
>>> make the review is OK for me. How hive compares to gerrit? Could we also
>>> plug it with github like gerrit [1] ?
>>>
>>> - benoît
>>>
>>> [1] https://gerrit.googlesource.com/plugins/github/ (and some others)
>>>
>>> >
>>> > On 20 January 2014 15:30, Nick North <no...@gmail.com> wrote:
>>> > > On 20 January 2014 12:26, Dale Harvey <da...@arandomurl.com> wrote:
>>> > >
>>> > >> I fully agree, its something I mentioned at the couchdb conf in
>>> > vancouver,
>>> > >> a review first system encourages contributions and has multiple
>>> benefits
>>> > >>
>>> > >> * At least 2 people look at the code, less likely to push silly
>>> > mistakes
>>> > >> * Can codify and practice review rules
>>> > >> * Its much easier to view the current activity in the project
>>> > >> * Can bring up points of discussion before its too late
>>> > >>
>>> > >> But I think the most important thing is that it removes the burden
>>>of
>>> > >> responsibility from the committer to the project as a whole, also
>>> hugely
>>> > >> beneficial is that it makes it particularly obvious when a patch
>>>has
>>> > reach
>>> > >> a stalemate and forces someone to make the call.
>>> > >>
>>> > >> For reference on PouchDB every committer is trusted to push code,
>>> nobody
>>> > >> (including myself) pushes their own code to master, it goes in the
>>>PR
>>> > queue
>>> > >> and gets a +1 from any other committer (who will usually push it),
>>> thats
>>> > >> essentially the same process we use at mozilla and coming to think
>>>of
>>> it
>>> > >> any other project I have worked on, any commiter has the ability to
>>> -1 a
>>> > >> patch at which point they give a reason and usually some solution
>>>gets
>>> > >> agreed on
>>> > >>
>>> > >
>>> > > I like this system, with one small quibble about responsibility. I
>>> don't
>>> > > think it should be seen as removing the burden of responsibility
>>>from
>>> the
>>> > > committer to the project as a whole. It becomes easy then for
>>>everyone,
>>> > > including the committer, to think someone else will find bugs, and
>>> no-one
>>> > > puts in enough effort. I'd say it is still primarily the
>>>responsibility
>>> > of
>>> > > the committer to ensure that code is error free, but that at least
>>>one
>>> > > person who knows that area of the code base should sign off on it.
>>>Some
>>> > > spreading of responsibility is good, but too much can actually lead
>>>to
>>> a
>>> > > decline in quality.
>>> > >
>>> > >
>>> > > Nick
>>> >
>>> >
>>> >
>>> > --
>>> > Noah Slater
>>> > https://twitter.com/nslater
>>> >
>>>
>
>
> +1 to reviews and getting code guidelines. Nice to see that Apache has a
> reviewboard instance available (https://reviews.apache.org/groups/ hive
> link). I've spent a lot of time using reviewboard, it's got a good UI and
> API. Reviews are submitted using a separate command line tool
> (post-review) to actually post the review, but it lets you make multiple
> updates from feedback and allows others to see the diffs on the same
> review so you can see how a change evolves. You can leave comments or put
> a "ship-it" tag to +1 the commit, then from a distance it's possible to
> see all the reviews outstanding and how many votes, if any, they've
> received. It's friendly for non-blocking reviews too, since the code
> review process is separate. It also lets you can work on reviews a little
> at a time and save them in between instead of having to devote a block of
> time to code reviews (I do this sometimes on big changes).
>
> Garrit wants to gate the source-tree if I remember correctly so is
> possibly less friendly with remote repositories.
>
> </JamesM>
>
--
Noah Slater
https://twitter.com/nslater
Re: we need more reviews
Posted by "Mutton, James" <jm...@akamai.com>.
On 1/22/14 3:07 PM, "Russell Branca" <ch...@apache.org> wrote:
>Huge +1 to more review.
>
>Let's also setup some code guidelines for Erlang/Javascript/C/HTML so we
>have an authoritative list of rules to follow to ensure code consistency,
>and similarly, let's get some guidelines for git commits messages as well,
>Tim Pope's article comes to mind:
>http://tbaggery.com/2008/04/19/a-note-about-git-commit-messages.html.
>
>Whichever review tool we decide to use, we should use it universally. We
>should figure out a way to do github integration so that PR's show up in
>the tool, and then we update the PR with a link to the relevant review
>page. We also need a good way to initiate the analogue of a PR in whatever
>tool we use. One of the complications with the review process today, is
>that we're all accustomed to doing review through github PR's, but we can
>use github PR's for branches who exist solely on the ASF origin. I imagine
>this won't be an issue with Hive or Gerrit, but we should verify that
>we're
>not required to initiate code review from third party origins.
>
>On a side note, one of my personal nits with github PR's, that I hope
>might
>be resolved with one of these tools, is losing comments when the code
>changes. For instance, if I do a review and make 10 comments that each
>represent a line item that needs to be addressed, I would like to see a
>task list on the review indicating there are unfinished items to be
>addressed, basically a way to make a set of checkboxes necessary to
>complete the review. I find it odd that github doesn't do this, even more
>so by the fact that changing code to address comments will often mess with
>the commenting and hide the messages you posted, making it challenging to
>decipher whether the review items have been addressed. This is by no means
>a requirement to using any approach, just hoping that someone knows of a
>good way to approach this problem in whatever tool we use.
>
>
>-Russell
>
>
>On Wed, Jan 22, 2014 at 6:06 AM, Benoit Chesneau
><bc...@gmail.com>wrote:
>
>> On Wed, Jan 22, 2014 at 12:47 PM, Noah Slater <ns...@apache.org>
>>wrote:
>>
>> > My first comment is: if we want more reviews, let's have more
>>committers.
>> >
>> > We double our committer base in 2013, and the results look like this:
>> >
>> > https://www.ohloh.net/p/couchdb/analyses/latest/languages_summary
>> >
>> > And I see comments like this on StackOverflow:
>> >
>> > "I've recently noticed that Couch DB is back in heavy
>>development."
>>
>>
>> > So I think we should continue to aggressively recruit more committers
>> > to the project in 2014. Excelsior!
>> >
>>
>> Welcoming new committers in the project is certainly a good thing but I
>> think it's orthogonal to the need of more review and team work. We
>>should
>> find a good workflow now while we are still a limited number of
>>committers
>> because It will be harder to enforce anything on a larger team. We
>>should
>> settle on some guidelines now. It will also help us to attract more
>>people
>> IMO.
>>
>>
>> >
>> > However, as for the way we do reviews...
>> >
>> > Infra ticket about Gerrit
>> > https://issues.apache.org/jira/browse/INFRA-2205
>> >
>> > Effort seems to be mothballed. But I'm sure we could restart it if we
>> > were serious.
>> >
>> > However, we could use this:
>> >
>> > https://reviews.apache.org/groups/hive/
>> >
>> > It is well integrated with Apache infrastructure already, sends mails
>> > to the mailing list, and so on.
>> >
>> > Happy to request an instance and we can experiment with it if we like.
>> > If it doesn't work out, we stop using it. No harm. Could make it
>> > entirely voluntary until we figure out a workflow that we all like.
>> >
>> > Should I do that?
>> >
>>
>> I thought gerrit was already integrated but any tool that could help us
>>to
>> make the review is OK for me. How hive compares to gerrit? Could we also
>> plug it with github like gerrit [1] ?
>>
>> - benoît
>>
>> [1] https://gerrit.googlesource.com/plugins/github/ (and some others)
>>
>> >
>> > On 20 January 2014 15:30, Nick North <no...@gmail.com> wrote:
>> > > On 20 January 2014 12:26, Dale Harvey <da...@arandomurl.com> wrote:
>> > >
>> > >> I fully agree, its something I mentioned at the couchdb conf in
>> > vancouver,
>> > >> a review first system encourages contributions and has multiple
>> benefits
>> > >>
>> > >> * At least 2 people look at the code, less likely to push silly
>> > mistakes
>> > >> * Can codify and practice review rules
>> > >> * Its much easier to view the current activity in the project
>> > >> * Can bring up points of discussion before its too late
>> > >>
>> > >> But I think the most important thing is that it removes the burden
>>of
>> > >> responsibility from the committer to the project as a whole, also
>> hugely
>> > >> beneficial is that it makes it particularly obvious when a patch
>>has
>> > reach
>> > >> a stalemate and forces someone to make the call.
>> > >>
>> > >> For reference on PouchDB every committer is trusted to push code,
>> nobody
>> > >> (including myself) pushes their own code to master, it goes in the
>>PR
>> > queue
>> > >> and gets a +1 from any other committer (who will usually push it),
>> thats
>> > >> essentially the same process we use at mozilla and coming to think
>>of
>> it
>> > >> any other project I have worked on, any commiter has the ability to
>> -1 a
>> > >> patch at which point they give a reason and usually some solution
>>gets
>> > >> agreed on
>> > >>
>> > >
>> > > I like this system, with one small quibble about responsibility. I
>> don't
>> > > think it should be seen as removing the burden of responsibility
>>from
>> the
>> > > committer to the project as a whole. It becomes easy then for
>>everyone,
>> > > including the committer, to think someone else will find bugs, and
>> no-one
>> > > puts in enough effort. I'd say it is still primarily the
>>responsibility
>> > of
>> > > the committer to ensure that code is error free, but that at least
>>one
>> > > person who knows that area of the code base should sign off on it.
>>Some
>> > > spreading of responsibility is good, but too much can actually lead
>>to
>> a
>> > > decline in quality.
>> > >
>> > >
>> > > Nick
>> >
>> >
>> >
>> > --
>> > Noah Slater
>> > https://twitter.com/nslater
>> >
>>
+1 to reviews and getting code guidelines. Nice to see that Apache has a
reviewboard instance available (https://reviews.apache.org/groups/ hive
link). I've spent a lot of time using reviewboard, it's got a good UI and
API. Reviews are submitted using a separate command line tool
(post-review) to actually post the review, but it lets you make multiple
updates from feedback and allows others to see the diffs on the same
review so you can see how a change evolves. You can leave comments or put
a "ship-it" tag to +1 the commit, then from a distance it's possible to
see all the reviews outstanding and how many votes, if any, they've
received. It's friendly for non-blocking reviews too, since the code
review process is separate. It also lets you can work on reviews a little
at a time and save them in between instead of having to devote a block of
time to code reviews (I do this sometimes on big changes).
Garrit wants to gate the source-tree if I remember correctly so is
possibly less friendly with remote repositories.
</JamesM>
Re: we need more reviews
Posted by Russell Branca <ch...@apache.org>.
Huge +1 to more review.
Let's also setup some code guidelines for Erlang/Javascript/C/HTML so we
have an authoritative list of rules to follow to ensure code consistency,
and similarly, let's get some guidelines for git commits messages as well,
Tim Pope's article comes to mind:
http://tbaggery.com/2008/04/19/a-note-about-git-commit-messages.html.
Whichever review tool we decide to use, we should use it universally. We
should figure out a way to do github integration so that PR's show up in
the tool, and then we update the PR with a link to the relevant review
page. We also need a good way to initiate the analogue of a PR in whatever
tool we use. One of the complications with the review process today, is
that we're all accustomed to doing review through github PR's, but we can
use github PR's for branches who exist solely on the ASF origin. I imagine
this won't be an issue with Hive or Gerrit, but we should verify that we're
not required to initiate code review from third party origins.
On a side note, one of my personal nits with github PR's, that I hope might
be resolved with one of these tools, is losing comments when the code
changes. For instance, if I do a review and make 10 comments that each
represent a line item that needs to be addressed, I would like to see a
task list on the review indicating there are unfinished items to be
addressed, basically a way to make a set of checkboxes necessary to
complete the review. I find it odd that github doesn't do this, even more
so by the fact that changing code to address comments will often mess with
the commenting and hide the messages you posted, making it challenging to
decipher whether the review items have been addressed. This is by no means
a requirement to using any approach, just hoping that someone knows of a
good way to approach this problem in whatever tool we use.
-Russell
On Wed, Jan 22, 2014 at 6:06 AM, Benoit Chesneau <bc...@gmail.com>wrote:
> On Wed, Jan 22, 2014 at 12:47 PM, Noah Slater <ns...@apache.org> wrote:
>
> > My first comment is: if we want more reviews, let's have more committers.
> >
> > We double our committer base in 2013, and the results look like this:
> >
> > https://www.ohloh.net/p/couchdb/analyses/latest/languages_summary
> >
> > And I see comments like this on StackOverflow:
> >
> > "I've recently noticed that Couch DB is back in heavy development."
>
>
> > So I think we should continue to aggressively recruit more committers
> > to the project in 2014. Excelsior!
> >
>
> Welcoming new committers in the project is certainly a good thing but I
> think it's orthogonal to the need of more review and team work. We should
> find a good workflow now while we are still a limited number of committers
> because It will be harder to enforce anything on a larger team. We should
> settle on some guidelines now. It will also help us to attract more people
> IMO.
>
>
> >
> > However, as for the way we do reviews...
> >
> > Infra ticket about Gerrit
> > https://issues.apache.org/jira/browse/INFRA-2205
> >
> > Effort seems to be mothballed. But I'm sure we could restart it if we
> > were serious.
> >
> > However, we could use this:
> >
> > https://reviews.apache.org/groups/hive/
> >
> > It is well integrated with Apache infrastructure already, sends mails
> > to the mailing list, and so on.
> >
> > Happy to request an instance and we can experiment with it if we like.
> > If it doesn't work out, we stop using it. No harm. Could make it
> > entirely voluntary until we figure out a workflow that we all like.
> >
> > Should I do that?
> >
>
> I thought gerrit was already integrated but any tool that could help us to
> make the review is OK for me. How hive compares to gerrit? Could we also
> plug it with github like gerrit [1] ?
>
> - benoît
>
> [1] https://gerrit.googlesource.com/plugins/github/ (and some others)
>
> >
> > On 20 January 2014 15:30, Nick North <no...@gmail.com> wrote:
> > > On 20 January 2014 12:26, Dale Harvey <da...@arandomurl.com> wrote:
> > >
> > >> I fully agree, its something I mentioned at the couchdb conf in
> > vancouver,
> > >> a review first system encourages contributions and has multiple
> benefits
> > >>
> > >> * At least 2 people look at the code, less likely to push silly
> > mistakes
> > >> * Can codify and practice review rules
> > >> * Its much easier to view the current activity in the project
> > >> * Can bring up points of discussion before its too late
> > >>
> > >> But I think the most important thing is that it removes the burden of
> > >> responsibility from the committer to the project as a whole, also
> hugely
> > >> beneficial is that it makes it particularly obvious when a patch has
> > reach
> > >> a stalemate and forces someone to make the call.
> > >>
> > >> For reference on PouchDB every committer is trusted to push code,
> nobody
> > >> (including myself) pushes their own code to master, it goes in the PR
> > queue
> > >> and gets a +1 from any other committer (who will usually push it),
> thats
> > >> essentially the same process we use at mozilla and coming to think of
> it
> > >> any other project I have worked on, any commiter has the ability to
> -1 a
> > >> patch at which point they give a reason and usually some solution gets
> > >> agreed on
> > >>
> > >
> > > I like this system, with one small quibble about responsibility. I
> don't
> > > think it should be seen as removing the burden of responsibility from
> the
> > > committer to the project as a whole. It becomes easy then for everyone,
> > > including the committer, to think someone else will find bugs, and
> no-one
> > > puts in enough effort. I'd say it is still primarily the responsibility
> > of
> > > the committer to ensure that code is error free, but that at least one
> > > person who knows that area of the code base should sign off on it. Some
> > > spreading of responsibility is good, but too much can actually lead to
> a
> > > decline in quality.
> > >
> > >
> > > Nick
> >
> >
> >
> > --
> > Noah Slater
> > https://twitter.com/nslater
> >
>
Re: we need more reviews
Posted by Benoit Chesneau <bc...@gmail.com>.
On Wed, Jan 22, 2014 at 12:47 PM, Noah Slater <ns...@apache.org> wrote:
> My first comment is: if we want more reviews, let's have more committers.
>
> We double our committer base in 2013, and the results look like this:
>
> https://www.ohloh.net/p/couchdb/analyses/latest/languages_summary
>
> And I see comments like this on StackOverflow:
>
> "I've recently noticed that Couch DB is back in heavy development."
> So I think we should continue to aggressively recruit more committers
> to the project in 2014. Excelsior!
>
Welcoming new committers in the project is certainly a good thing but I
think it's orthogonal to the need of more review and team work. We should
find a good workflow now while we are still a limited number of committers
because It will be harder to enforce anything on a larger team. We should
settle on some guidelines now. It will also help us to attract more people
IMO.
>
> However, as for the way we do reviews...
>
> Infra ticket about Gerrit
> https://issues.apache.org/jira/browse/INFRA-2205
>
> Effort seems to be mothballed. But I'm sure we could restart it if we
> were serious.
>
> However, we could use this:
>
> https://reviews.apache.org/groups/hive/
>
> It is well integrated with Apache infrastructure already, sends mails
> to the mailing list, and so on.
>
> Happy to request an instance and we can experiment with it if we like.
> If it doesn't work out, we stop using it. No harm. Could make it
> entirely voluntary until we figure out a workflow that we all like.
>
> Should I do that?
>
I thought gerrit was already integrated but any tool that could help us to
make the review is OK for me. How hive compares to gerrit? Could we also
plug it with github like gerrit [1] ?
- benoît
[1] https://gerrit.googlesource.com/plugins/github/ (and some others)
>
> On 20 January 2014 15:30, Nick North <no...@gmail.com> wrote:
> > On 20 January 2014 12:26, Dale Harvey <da...@arandomurl.com> wrote:
> >
> >> I fully agree, its something I mentioned at the couchdb conf in
> vancouver,
> >> a review first system encourages contributions and has multiple benefits
> >>
> >> * At least 2 people look at the code, less likely to push silly
> mistakes
> >> * Can codify and practice review rules
> >> * Its much easier to view the current activity in the project
> >> * Can bring up points of discussion before its too late
> >>
> >> But I think the most important thing is that it removes the burden of
> >> responsibility from the committer to the project as a whole, also hugely
> >> beneficial is that it makes it particularly obvious when a patch has
> reach
> >> a stalemate and forces someone to make the call.
> >>
> >> For reference on PouchDB every committer is trusted to push code, nobody
> >> (including myself) pushes their own code to master, it goes in the PR
> queue
> >> and gets a +1 from any other committer (who will usually push it), thats
> >> essentially the same process we use at mozilla and coming to think of it
> >> any other project I have worked on, any commiter has the ability to -1 a
> >> patch at which point they give a reason and usually some solution gets
> >> agreed on
> >>
> >
> > I like this system, with one small quibble about responsibility. I don't
> > think it should be seen as removing the burden of responsibility from the
> > committer to the project as a whole. It becomes easy then for everyone,
> > including the committer, to think someone else will find bugs, and no-one
> > puts in enough effort. I'd say it is still primarily the responsibility
> of
> > the committer to ensure that code is error free, but that at least one
> > person who knows that area of the code base should sign off on it. Some
> > spreading of responsibility is good, but too much can actually lead to a
> > decline in quality.
> >
> >
> > Nick
>
>
>
> --
> Noah Slater
> https://twitter.com/nslater
>
Re: we need more reviews
Posted by Dirkjan Ochtman <di...@ochtman.nl>.
On Wed, Jan 22, 2014 at 2:29 PM, Adam Kocoloski <ko...@apache.org> wrote:
> What really excites me is that all of the areas where we saw major activity in 2013 are complementary to improvements that we can offer in the core. It feels like the project is more than ready to capitalize on the new features and enhancements that we can introduce.
Yup, it feels like 2014 has the potential to be quite awesome in this regard.
Cheers,
Dirkjan
Re: we need more reviews
Posted by Adam Kocoloski <ko...@apache.org>.
Heya Dirkjan, that's a fair assessment. The unfortunate reality is that many of us who have experience with the Erlang core _have_ continued to make improvements (often open source) to the codebase -- we've just done it outside of Apache. We're committed to changing that going forward; it's better for the project and better for our commercial ventures if our open source work happens here.
What really excites me is that all of the areas where we saw major activity in 2013 are complementary to improvements that we can offer in the core. It feels like the project is more than ready to capitalize on the new features and enhancements that we can introduce. Cheers,
Adam
On Jan 22, 2014, at 8:19 AM, Noah Slater <ns...@apache.org> wrote:
> Yes, most of the commits were on Fauxton, but almost without fail,
> handing out a commit bit results in an upsurge of activity. Sometimes
> a huge upsurge. And I think that we need to keep enabling people to
> more easily contribute to the project.
>
> As for there being no candidates, I disagree. There are several people
> in this community who do not have a commit bit, who I personally have
> shortlisted for election. Sometimes, I'm just waiting to see a little
> uptick in patches, mailing list posts, or what have you.
>
> Also, activity begets activity. The work you are doing on the releases
> is hugely influential in communicating to the outside world that
> CouchDB is alive and kicking, and that in turn draws in more users,
> and more contributors.
>
> On 22 January 2014 13:16, Dirkjan Ochtman <di...@ochtman.nl> wrote:
>> On Wed, Jan 22, 2014 at 12:47 PM, Noah Slater <ns...@apache.org> wrote:
>>> My first comment is: if we want more reviews, let's have more committers.
>>>
>>> We double our committer base in 2013, and the results look like this:
>>>
>>> https://www.ohloh.net/p/couchdb/analyses/latest/languages_summary
>>
>> I don't think this is a panacea.
>>
>> Yes, adding more committers is good. But it seems to me that *most* of
>> the effort in 2013 has gone into Fauxton, which AFAICT is almost
>> exclusively the work of people who are paid to work on it by a single
>> vendor. This is great, but I think we're currently sorely lacking in
>> more coreish committer man-hours, and I don't think we have good
>> candidates to add as committers here (although I do hope the bigcouch
>> + rcouch merge will result in a resurgence of core work).
>>
>> Cheers,
>>
>> Dirkjan
>
>
>
> --
> Noah Slater
> https://twitter.com/nslater
Re: we need more reviews
Posted by Noah Slater <ns...@apache.org>.
Yes, most of the commits were on Fauxton, but almost without fail,
handing out a commit bit results in an upsurge of activity. Sometimes
a huge upsurge. And I think that we need to keep enabling people to
more easily contribute to the project.
As for there being no candidates, I disagree. There are several people
in this community who do not have a commit bit, who I personally have
shortlisted for election. Sometimes, I'm just waiting to see a little
uptick in patches, mailing list posts, or what have you.
Also, activity begets activity. The work you are doing on the releases
is hugely influential in communicating to the outside world that
CouchDB is alive and kicking, and that in turn draws in more users,
and more contributors.
On 22 January 2014 13:16, Dirkjan Ochtman <di...@ochtman.nl> wrote:
> On Wed, Jan 22, 2014 at 12:47 PM, Noah Slater <ns...@apache.org> wrote:
>> My first comment is: if we want more reviews, let's have more committers.
>>
>> We double our committer base in 2013, and the results look like this:
>>
>> https://www.ohloh.net/p/couchdb/analyses/latest/languages_summary
>
> I don't think this is a panacea.
>
> Yes, adding more committers is good. But it seems to me that *most* of
> the effort in 2013 has gone into Fauxton, which AFAICT is almost
> exclusively the work of people who are paid to work on it by a single
> vendor. This is great, but I think we're currently sorely lacking in
> more coreish committer man-hours, and I don't think we have good
> candidates to add as committers here (although I do hope the bigcouch
> + rcouch merge will result in a resurgence of core work).
>
> Cheers,
>
> Dirkjan
--
Noah Slater
https://twitter.com/nslater
Re: we need more reviews
Posted by Dirkjan Ochtman <di...@ochtman.nl>.
On Wed, Jan 22, 2014 at 12:47 PM, Noah Slater <ns...@apache.org> wrote:
> My first comment is: if we want more reviews, let's have more committers.
>
> We double our committer base in 2013, and the results look like this:
>
> https://www.ohloh.net/p/couchdb/analyses/latest/languages_summary
I don't think this is a panacea.
Yes, adding more committers is good. But it seems to me that *most* of
the effort in 2013 has gone into Fauxton, which AFAICT is almost
exclusively the work of people who are paid to work on it by a single
vendor. This is great, but I think we're currently sorely lacking in
more coreish committer man-hours, and I don't think we have good
candidates to add as committers here (although I do hope the bigcouch
+ rcouch merge will result in a resurgence of core work).
Cheers,
Dirkjan
Re: we need more reviews
Posted by Noah Slater <ns...@apache.org>.
My first comment is: if we want more reviews, let's have more committers.
We double our committer base in 2013, and the results look like this:
https://www.ohloh.net/p/couchdb/analyses/latest/languages_summary
And I see comments like this on StackOverflow:
"I've recently noticed that Couch DB is back in heavy development."
So I think we should continue to aggressively recruit more committers
to the project in 2014. Excelsior!
However, as for the way we do reviews...
Infra ticket about Gerrit
https://issues.apache.org/jira/browse/INFRA-2205
Effort seems to be mothballed. But I'm sure we could restart it if we
were serious.
However, we could use this:
https://reviews.apache.org/groups/hive/
It is well integrated with Apache infrastructure already, sends mails
to the mailing list, and so on.
Happy to request an instance and we can experiment with it if we like.
If it doesn't work out, we stop using it. No harm. Could make it
entirely voluntary until we figure out a workflow that we all like.
Should I do that?
On 20 January 2014 15:30, Nick North <no...@gmail.com> wrote:
> On 20 January 2014 12:26, Dale Harvey <da...@arandomurl.com> wrote:
>
>> I fully agree, its something I mentioned at the couchdb conf in vancouver,
>> a review first system encourages contributions and has multiple benefits
>>
>> * At least 2 people look at the code, less likely to push silly mistakes
>> * Can codify and practice review rules
>> * Its much easier to view the current activity in the project
>> * Can bring up points of discussion before its too late
>>
>> But I think the most important thing is that it removes the burden of
>> responsibility from the committer to the project as a whole, also hugely
>> beneficial is that it makes it particularly obvious when a patch has reach
>> a stalemate and forces someone to make the call.
>>
>> For reference on PouchDB every committer is trusted to push code, nobody
>> (including myself) pushes their own code to master, it goes in the PR queue
>> and gets a +1 from any other committer (who will usually push it), thats
>> essentially the same process we use at mozilla and coming to think of it
>> any other project I have worked on, any commiter has the ability to -1 a
>> patch at which point they give a reason and usually some solution gets
>> agreed on
>>
>
> I like this system, with one small quibble about responsibility. I don't
> think it should be seen as removing the burden of responsibility from the
> committer to the project as a whole. It becomes easy then for everyone,
> including the committer, to think someone else will find bugs, and no-one
> puts in enough effort. I'd say it is still primarily the responsibility of
> the committer to ensure that code is error free, but that at least one
> person who knows that area of the code base should sign off on it. Some
> spreading of responsibility is good, but too much can actually lead to a
> decline in quality.
>
>
> Nick
--
Noah Slater
https://twitter.com/nslater
Re: we need more reviews
Posted by Nick North <no...@gmail.com>.
On 20 January 2014 12:26, Dale Harvey <da...@arandomurl.com> wrote:
> I fully agree, its something I mentioned at the couchdb conf in vancouver,
> a review first system encourages contributions and has multiple benefits
>
> * At least 2 people look at the code, less likely to push silly mistakes
> * Can codify and practice review rules
> * Its much easier to view the current activity in the project
> * Can bring up points of discussion before its too late
>
> But I think the most important thing is that it removes the burden of
> responsibility from the committer to the project as a whole, also hugely
> beneficial is that it makes it particularly obvious when a patch has reach
> a stalemate and forces someone to make the call.
>
> For reference on PouchDB every committer is trusted to push code, nobody
> (including myself) pushes their own code to master, it goes in the PR queue
> and gets a +1 from any other committer (who will usually push it), thats
> essentially the same process we use at mozilla and coming to think of it
> any other project I have worked on, any commiter has the ability to -1 a
> patch at which point they give a reason and usually some solution gets
> agreed on
>
I like this system, with one small quibble about responsibility. I don't
think it should be seen as removing the burden of responsibility from the
committer to the project as a whole. It becomes easy then for everyone,
including the committer, to think someone else will find bugs, and no-one
puts in enough effort. I'd say it is still primarily the responsibility of
the committer to ensure that code is error free, but that at least one
person who knows that area of the code base should sign off on it. Some
spreading of responsibility is good, but too much can actually lead to a
decline in quality.
Nick
Re: we need more reviews
Posted by Dale Harvey <da...@arandomurl.com>.
I fully agree, its something I mentioned at the couchdb conf in vancouver,
a review first system encourages contributions and has multiple benefits
* At least 2 people look at the code, less likely to push silly mistakes
* Can codify and practice review rules
* Its much easier to view the current activity in the project
* Can bring up points of discussion before its too late
But I think the most important thing is that it removes the burden of
responsibility from the committer to the project as a whole, also hugely
beneficial is that it makes it particularly obvious when a patch has reach
a stalemate and forces someone to make the call.
For reference on PouchDB every committer is trusted to push code, nobody
(including myself) pushes their own code to master, it goes in the PR queue
and gets a +1 from any other committer (who will usually push it), thats
essentially the same process we use at mozilla and coming to think of it
any other project I have worked on, any commiter has the ability to -1 a
patch at which point they give a reason and usually some solution gets
agreed on
On 20 January 2014 12:07, Benoit Chesneau <bc...@gmail.com> wrote:
> Hi All,
>
> Over the last few months, we have had a pretty good track in commits --
> which is great! However I think we can do better working as a team on
> this. I am sure others also feel the same way. For example:
>
> - Some code is still landing in master with little or absence of review
> in JIRA nor the mailing list
>
> - There is insufficient discussion, sometimes a short word on IRC or out
> of sight of the mailing list. Some of the discussions feel more like a
> monologue, or don't reach a satisfactory conclusion or clear decision.
> For example, the last discussion about metadata never reached a
> conclusion and as a result, no changes have been made.
>
> I really want to find a way to fix that. It is our responsibility to fix
> that.
> The project should count as a real team work.
>
> A team is not a collection of individuals working on their side
> where`git commit` or `jira close` are the only meeting points. That also
> opens the doors to more conflicts, since people have to react against a
> thing that have been done. We should discuss and review more the code
> (and a review should be answered). Actually I do think that no code
> should land in master without at least one or more (depending how
> sensitive the patch is) public review. But we may be less strict. At
> least nothing should come without a context. It’s also true for the
> commits messages. There may be a discussion on IRC that explain that but
> IRC is not a reliable source. People should understand the why and the
> what for without having to read the code and should understand the
> direction and eventually discuss it before it is taken.
>
> All of these only require some guidelines. But we should start to
> enforce them and make them public. Developers should act together
> again.
>
> In term of action I think we could:
>
> - have review system with votes on the patch before inclusion in the
> master (as in the apache https project). Many projects are working
> like this and this is part of their success. We should of course explain
> why a review is important and make sure that reviews are not transformed
> in a fight between some egos.
> - have some guidelines on how to discuss features
> - make it clear that IRC is not the place where decisions are took and
> enforce the ML usage
> - make sure that people are writing good commit messages. Some projects
> (erlang. linux are coming to my mind) make sure to fix them before
> they land in maint or master branches.
>
> Thoughts?
>
>
> - benoit
>
> NB: this a repost on public @dev mailing-list from the PMC mailing-list.
>