diff mbox series

[bug#59513] doc: contributing: Tweak the Commit Policy.

Message ID 20221123104946.29480-1-mail@cbaines.net
State New
Headers show
Series [bug#59513] doc: contributing: Tweak the Commit Policy. | expand

Commit Message

Christopher Baines Nov. 23, 2022, 10:49 a.m. UTC
Add more examples of when it can be appropriate to push changes without
review, as I think this can be appropriate in the case of trivial changes (as
mentioned before), but also non-trivial fixes.

No longer suggest pushing simple new packages or package upgrades (that don't
cause lots of rebuilds) without sending to guix-patches. Now there's some
automation for testing changes sent to guix-patches, sending changes there
before pushing can mean that more rigerious testing takes place and help speed
up substitutes becoming available. This is true, even if no human review takes
place.

Only suggest waiting one week for review for simpler changes, wait two weeks
for more significant changes.

Also, reorder some of the information in this section so it's grouped together
better.

* doc/contributing.texi (Commit Policy): Tweak.
---
 doc/contributing.texi | 38 ++++++++++++++++----------------------
 1 file changed, 16 insertions(+), 22 deletions(-)

Comments

zimoun Nov. 23, 2022, 8:27 p.m. UTC | #1
Hi Chris,

On Wed, 23 Nov 2022 at 10:49, Christopher Baines <mail@cbaines.net> wrote:

> +For a minority of changes, it can be appropriate to push them directly
> +without sending them for review.  This includes both trivial changes
> +(e.g. fixing typos) but also reverting problomatic changes and
                                              -^
> +addressing regressions.


> -For patches that just add a new package, and a simple one, it's OK to
> -commit, if you're confident (which means you successfully built it in a
> -chroot setup, and have done a reasonable copyright and license
> -auditing).  Likewise for package upgrades, except upgrades that trigger
> -a lot of rebuilds (for example, upgrading GnuTLS or GLib).  We have a
> -mailing list for commit notifications (@email{guix-commits@@gnu.org}),
> -so people can notice.  Before pushing your changes, make sure to run
> -@code{git pull --rebase}.
> +In general though, all changes should be posted to
> +@email{guix-patches@@gnu.org}.  This mailing list fills the
> +patch-tracking database (@pxref{Tracking Bugs and Patches}).  Leave time
> +for a review, without committing anything (@pxref{Submitting Patches}).
> +If you didn’t receive any reply after one week (two weeks for more
> +significant changes), and if you're confident, it's OK to commit.

I would write:

        … changes), and if you're confident (which means you
        successfully built it in a chroot setup, and have done a
        reasonable copyright and license auditing), it’s OK to commit.

and I would keep the «two weeks» instead of the «one week except».

I think it is also useful to provide the information about commit
notifications (guix-commits mailing list).

For what it is worth, I find clearer the structure,

    For patches that …
    For anything else, …

or

    For a minority of changes, …
    For anything else, …

than «In general though, all changes …».

Cheers,
simon
Christopher Baines Nov. 24, 2022, 8:40 a.m. UTC | #2
zimoun <zimon.toutoune@gmail.com> writes:

> Hi Chris,

Hey! Thanks for taking a look.

> On Wed, 23 Nov 2022 at 10:49, Christopher Baines <mail@cbaines.net> wrote:
>
>> +For a minority of changes, it can be appropriate to push them directly
>> +without sending them for review.  This includes both trivial changes
>> +(e.g. fixing typos) but also reverting problomatic changes and
>                                               -^
>> +addressing regressions.
>
>
>> -For patches that just add a new package, and a simple one, it's OK to
>> -commit, if you're confident (which means you successfully built it in a
>> -chroot setup, and have done a reasonable copyright and license
>> -auditing).  Likewise for package upgrades, except upgrades that trigger
>> -a lot of rebuilds (for example, upgrading GnuTLS or GLib).  We have a
>> -mailing list for commit notifications (@email{guix-commits@@gnu.org}),
>> -so people can notice.  Before pushing your changes, make sure to run
>> -@code{git pull --rebase}.
>> +In general though, all changes should be posted to
>> +@email{guix-patches@@gnu.org}.  This mailing list fills the
>> +patch-tracking database (@pxref{Tracking Bugs and Patches}).  Leave time
>> +for a review, without committing anything (@pxref{Submitting Patches}).
>> +If you didn’t receive any reply after one week (two weeks for more
>> +significant changes), and if you're confident, it's OK to commit.
>
> I would write:
>
>         … changes), and if you're confident (which means you
>         successfully built it in a chroot setup, and have done a
>         reasonable copyright and license auditing), it’s OK to commit.

chroot setup doesn't really make sense to me, I'm not sure why that
needs specifying (like do we not want things for the Hurd pushing, since
the guix-daemon doesn't support build isolation there yet)?

Also, this guidance is very general, and I think it should be applicable
to all changes. We already trust people with commit access to know what
needs doing, I see this documentation as more about how, so I'd prefer
not to try and put a list here.

> and I would keep the «two weeks» instead of the «one week except».

My reason for changing this is that I think waiting two weeks after
sending a simple patch is unreasonable. The value from the automated
testing will come after one to two days, I just put a week to avoid
changing it too much, but maybe the lower bound should be two days.

> I think it is also useful to provide the information about commit
> notifications (guix-commits mailing list).

Why though? What do we expect people with commit access to do when they
read about that here?

> For what it is worth, I find clearer the structure,
>
>     For patches that …
>     For anything else, …
>
> or
>
>     For a minority of changes, …
>     For anything else, …
>
> than «In general though, all changes …».

That seems fine to me, I think "everything" maybe carries more weight
than "anything" though.

Thanks,

Chris
zimoun Nov. 24, 2022, 11:59 a.m. UTC | #3
Hi,

On Thu, 24 Nov 2022 at 08:40, Christopher Baines <mail@cbaines.net> wrote:

>> On Wed, 23 Nov 2022 at 10:49, Christopher Baines <mail@cbaines.net> wrote:
>>
>>> +For a minority of changes, it can be appropriate to push them directly
>>> +without sending them for review.  This includes both trivial changes
>>> +(e.g. fixing typos) but also reverting problomatic changes and
>>                                               -^
>>> +addressing regressions.

To be sure you have not missed the typo here. :-)

s/problomatic/problematic


>>         … changes), and if you're confident (which means you
>>         successfully built it in a chroot setup, and have done a
>>         reasonable copyright and license auditing), it’s OK to commit.
>
> chroot setup doesn't really make sense to me, I'm not sure why that
> needs specifying (like do we not want things for the Hurd pushing, since
> the guix-daemon doesn't support build isolation there yet)?

Good point about chroot. :-)


> Also, this guidance is very general, and I think it should be applicable
> to all changes. We already trust people with commit access to know what
> needs doing, I see this documentation as more about how, so I'd prefer
> not to try and put a list here.

Yes, we trust people.  But a public and explicit policy reinforces the
trust, IMHO.  It also documents what commit access means.  It is not
because people with commit access already know what they need doing that
all people know, I guess.


>> and I would keep the «two weeks» instead of the «one week except».
>
> My reason for changing this is that I think waiting two weeks after
> sending a simple patch is unreasonable. The value from the automated
> testing will come after one to two days, I just put a week to avoid
> changing it too much, but maybe the lower bound should be two days.

Who is verifying the impact of a change? :-) Just a recent example to
fix the ideas.  The same situation is happening more than often but not
that often neither. :-)

(It is an example and no blame here.  Or blame on me only, for not
taking enough care of Julia packages.)

Patch#58644 [1] submitted on 19 Oct and pushed on 8 Nov; which is 22
days.  Unfortunately, this patch breaks julia-documenter [2], so it
means many Julia packages are currently broken; since 17 days.

Commit 83ede5a02e1fc531d912eb92eb0a22a4b897997c,

        gnu: git: Update to 2.38.1.

        Fixes CVE-2022-39253 and CVE-2022-39260.

        * gnu/packages/version-control.scm (git): Update to 2.38.1.

        Co-authored-by: Ludovic Courtès <ludo@gnu.org>

        1 file changed, 3 insertions(+), 3 deletions(-)
        gnu/packages/version-control.scm | 6 +++---

from v2.38.0 to v2.38.1 seems inoffensive. :-)  But,

--8<---------------cut here---------------start------------->8---
$ ag 'inherit git'
gnu/packages/version-control.scm
613:    (inherit git)
676:  (package/inherit git-minimal

$ guix refresh -l git git-minimal  | cut -f1 -d':'
Building the following 292 packages would ensure 658 dependent packages are rebuilt
--8<---------------cut here---------------end--------------->8---

(The one at line 676 is not impacted by the change, IIRC.)


Who does check these 292 packages?

For instance, this patch has an impact on Julia packages,

--8<---------------cut here---------------start------------->8---
$ guix refresh -l git git-minimal  | cut -f2 -d':'  | sed 's/ /\n/g' | grep julia
julia-geometrybasics@0.4.1
julia-configurations@0.16.4
julia-pyplot@2.10.0
julia-recipespipeline@0.3.4
julia-quadmath@0.5.5
julia-plotthemes@2.0.1
julia-infinity@0.2.4
julia-testimages@1.5.0
julia-optim@1.6.0
julia-referencetests@0.9.7
julia-imagemagick@1.2.1
--8<---------------cut here---------------end--------------->8---

As part of the Julia team, maybe I could have a look.  Well, I had not
had the time in these 2 weeks to fix it yet.  For two reasons:

 1. Because I noticed the failure just a couple of days ago.
 2. Because I was busy elsewhere.

About #1, I can follow a RSS feed by Cuirass.  But somehow, it is too
late then either I am working in a rush to minimize the breakage or
either the package is broken… as today.

I have not yet carefully looked at the new QA (neat!).  Is it possible
to follow some notifications?

About #2, two weeks let the time to check the impact of a change.  And
even, depending on my schedule, sometime it is short. :-)  But hey, I
agree that the things need to move forward. :-)

My point is: Considering leaf packages, yeah once submitted, the review
can be fast (couple of days) especially with the new QA.  Considering
all the other packages, who is checking the impact of a change?

Otherwise, we have again and again some broken packages.  For sure, the
QA is helping *a lot* for improving!  Well, on one hand, I understand
the willing to merge faster and, even I am not convinced that from two
weeks to one week would be detrimental.  On the other hand, using Guix,
I replaced the pressure when running “apt-get upgrade” by an eternal
annoyance of broken packages popping here or there.

Somehow, <https://ci.guix.gnu.org/eval/4095/dashboard> should be always
all green and faster the Git tree moves, harder it is to achieve, IMHO.

1: https://issues.guix.gnu.org/58644
2: <https://data.guix.gnu.org/repository/1/branch/master/package/julia-documenter/output-history>


>> I think it is also useful to provide the information about commit
>> notifications (guix-commits mailing list).
>
> Why though? What do we expect people with commit access to do when they
> read about that here?

Maybe it is a niche, I used it a couple of time.  For instance, to
comment a change already merged, see [3] I guess.

3: <https://lists.gnu.org/archive/html/guix-devel/2022-10/msg00099.html>


Cheers,
simon
Christopher Baines Nov. 28, 2022, 11:46 a.m. UTC | #4
zimoun <zimon.toutoune@gmail.com> writes:

> On Thu, 24 Nov 2022 at 08:40, Christopher Baines <mail@cbaines.net> wrote:
>
>>> On Wed, 23 Nov 2022 at 10:49, Christopher Baines <mail@cbaines.net> wrote:
>>>
>>>> +For a minority of changes, it can be appropriate to push them directly
>>>> +without sending them for review.  This includes both trivial changes
>>>> +(e.g. fixing typos) but also reverting problomatic changes and
>>>                                               -^
>>>> +addressing regressions.
>
> To be sure you have not missed the typo here. :-)
>
> s/problomatic/problematic

Thanks, I've fixed that locally now.

>> Also, this guidance is very general, and I think it should be applicable
>> to all changes. We already trust people with commit access to know what
>> needs doing, I see this documentation as more about how, so I'd prefer
>> not to try and put a list here.
>
> Yes, we trust people.  But a public and explicit policy reinforces the
> trust, IMHO.  It also documents what commit access means.  It is not
> because people with commit access already know what they need doing that
> all people know, I guess.

I don't disagree that we should make the expectations about
functionality and testing explicit, but I want to see that separate from
the commit policy.

>>> and I would keep the «two weeks» instead of the «one week except».
>>
>> My reason for changing this is that I think waiting two weeks after
>> sending a simple patch is unreasonable. The value from the automated
>> testing will come after one to two days, I just put a week to avoid
>> changing it too much, but maybe the lower bound should be two days.
>
> Who is verifying the impact of a change? :-) Just a recent example to
> fix the ideas.  The same situation is happening more than often but not
> that often neither. :-)

...

> My point is: Considering leaf packages, yeah once submitted, the review
> can be fast (couple of days) especially with the new QA.  Considering
> all the other packages, who is checking the impact of a change?
>
> Otherwise, we have again and again some broken packages.  For sure, the
> QA is helping *a lot* for improving!  Well, on one hand, I understand
> the willing to merge faster and, even I am not convinced that from two
> weeks to one week would be detrimental.  On the other hand, using Guix,
> I replaced the pressure when running “apt-get upgrade” by an eternal
> annoyance of broken packages popping here or there.

This is going a bit off topic I think.

In general, the direction I'm trying to move the policy in here is one
where more changes get sent to guix-patches rather than getting pushed
straight to the repository.

Checking the impact of changes is important, but you can't do that with
a policy on committing. If however people send changes to guix-patches
prior to pushing, then there's at least a chance that some automatic
"verifying/checking" can take place.
Ludovic Courtès Dec. 1, 2022, 9:44 p.m. UTC | #5
Hi!

Christopher Baines <mail@cbaines.net> skribis:

> Add more examples of when it can be appropriate to push changes without
> review, as I think this can be appropriate in the case of trivial changes (as
> mentioned before), but also non-trivial fixes.
>
> No longer suggest pushing simple new packages or package upgrades (that don't
> cause lots of rebuilds) without sending to guix-patches. Now there's some
> automation for testing changes sent to guix-patches, sending changes there
> before pushing can mean that more rigerious testing takes place and help speed
> up substitutes becoming available. This is true, even if no human review takes
> place.
>
> Only suggest waiting one week for review for simpler changes, wait two weeks
> for more significant changes.
>
> Also, reorder some of the information in this section so it's grouped together
> better.
>
> * doc/contributing.texi (Commit Policy): Tweak.

FWIW I like the spirit of these changes.

Now to the letter…  :-)

>  @subsection Commit Policy
>  
> -If you get commit access, please make sure to follow
> -the policy below (discussions of the policy can take place on
> +If you get commit access, please make sure to follow the policy below
> +(discussions of the policy can take place on
>  @email{guix-devel@@gnu.org}).
>  
> -Non-trivial patches should always be posted to
> -@email{guix-patches@@gnu.org} (trivial patches include fixing typos,
> -etc.).  This mailing list fills the patch-tracking database
> -(@pxref{Tracking Bugs and Patches}).
> +For a minority of changes, it can be appropriate to push them directly
> +without sending them for review.  This includes both trivial changes
> +(e.g. fixing typos) but also reverting problomatic changes and
> +addressing regressions.
>  
> -For patches that just add a new package, and a simple one, it's OK to

Similar to zimoun’s first comment I think, I would like the beginning of
the sentence to clearly tell you whether it’s the situation you’re
interested in.  “For a minority of changes” doesn’t fit the bill in my
view.

So I would suggest something along the lines of:

  Changes should be posted to @email{guix-patches@@gnu.org}.  This
  mailing list […].  It also allows patches to be picked up and tested
  by the quality assurance robot; the result of that testing eventually
  shows up on the dashboard at
  @indicateurl{https://qa.guix.gnu.org/issue/@var{number}}, where
  @var{number} is the number assigned by the issue tracker.  Leave time
  […] it’s OK to commit.

  As an exception, some changes considered consensual and ``trivial'' or
  ``obvious'' may instead be pushed directly.  These include: fixing
  typos, and reverting commits that caused immediate problems.

That way we state the general rule first, and the exception next.  That
also explicitly mentions how that relates to qa.guix.

Regarding the list of exceptions, I feel that these two exceptions
listed here may be less than what we may except on a day-to-day basis;
perhaps there are other things to add there, but I’m not sure what.

Would be nice to get feedback from a maintainer too.

Thanks for working on this!

Ludo’.
Ludovic Courtès Dec. 2, 2022, 9:45 a.m. UTC | #6
Hi,

Christopher Baines <mail@cbaines.net> skribis:

>> I would write:
>>
>>         … changes), and if you're confident (which means you
>>         successfully built it in a chroot setup, and have done a
>>         reasonable copyright and license auditing), it’s OK to commit.
>
> chroot setup doesn't really make sense to me, I'm not sure why that
> needs specifying

The chroot requirement is specified because a non-isolated build is
worth nothing: it might work on some machine and fail on another one for
hard-to-find reasons.

We could stop mentioning it if we think everyone keeps chroot enabled
(that’s probably the case), but it doesn’t hurt to keep it.

> (like do we not want things for the Hurd pushing, since the
> guix-daemon doesn't support build isolation there yet)?

Well yes, that’s the thing.  For i586-gnu we’re currently stuck because
we haven’t yet defined how to normalize the build environment:

  https://guix.gnu.org/en/blog/2020/childhurds-and-substitutes/
  https://issues.guix.gnu.org/43857

And it’s not theoretical: some util-linux tests may or may not work
depending on whether the Hurd box has /proc set up, the Texinfo test
suite would pass if there happens to be a usable ‘pt_chown’ program
around (see <https://issues.guix.gnu.org/59616>), etc.

So I think the current situation is that we can build base packages for
i586-gnu, that’s fine, but don’t have a solid foundation like we do on
GNU/Linux, so there’s no point in going too far.

Now, I don’t think the i586-gnu situation is an important consideration
for the commit policy.

Ludo’.
diff mbox series

Patch

diff --git a/doc/contributing.texi b/doc/contributing.texi
index 40ae33ecac..6f772961ea 100644
--- a/doc/contributing.texi
+++ b/doc/contributing.texi
@@ -1819,23 +1819,25 @@  It additionally calls @code{make check-channel-news} to be sure
 
 @subsection Commit Policy
 
-If you get commit access, please make sure to follow
-the policy below (discussions of the policy can take place on
+If you get commit access, please make sure to follow the policy below
+(discussions of the policy can take place on
 @email{guix-devel@@gnu.org}).
 
-Non-trivial patches should always be posted to
-@email{guix-patches@@gnu.org} (trivial patches include fixing typos,
-etc.).  This mailing list fills the patch-tracking database
-(@pxref{Tracking Bugs and Patches}).
+For a minority of changes, it can be appropriate to push them directly
+without sending them for review.  This includes both trivial changes
+(e.g. fixing typos) but also reverting problomatic changes and
+addressing regressions.
 
-For patches that just add a new package, and a simple one, it's OK to
-commit, if you're confident (which means you successfully built it in a
-chroot setup, and have done a reasonable copyright and license
-auditing).  Likewise for package upgrades, except upgrades that trigger
-a lot of rebuilds (for example, upgrading GnuTLS or GLib).  We have a
-mailing list for commit notifications (@email{guix-commits@@gnu.org}),
-so people can notice.  Before pushing your changes, make sure to run
-@code{git pull --rebase}.
+In general though, all changes should be posted to
+@email{guix-patches@@gnu.org}.  This mailing list fills the
+patch-tracking database (@pxref{Tracking Bugs and Patches}).  Leave time
+for a review, without committing anything (@pxref{Submitting Patches}).
+If you didn’t receive any reply after one week (two weeks for more
+significant changes), and if you're confident, it's OK to commit.
+
+That last part is subject to being adjusted, allowing individuals to
+commit directly on non-controversial changes on parts they’re familiar
+with.
 
 When pushing a commit on behalf of somebody else, please add a
 @code{Signed-off-by} line at the end of the commit log message---e.g.,
@@ -1850,14 +1852,6 @@  right before pushing:
 make check-channel-news
 @end example
 
-For anything else, please post to @email{guix-patches@@gnu.org} and
-leave time for a review, without committing anything (@pxref{Submitting
-Patches}).  If you didn’t receive any reply after two weeks, and if
-you're confident, it's OK to commit.
-
-That last part is subject to being adjusted, allowing individuals to commit
-directly on non-controversial changes on parts they’re familiar with.
-
 @subsection Addressing Issues
 
 Peer review (@pxref{Submitting Patches}) and tools such as