We have not exactly defined this so far, but it is a good point to do so.

I personally find it good to have changes associated with an issue, because
it allows you to trace back why the change was done.
To make sure we do not overdo this and impose totally unnecessary overhead,
I would suggest the following:

*No issue is required for*

  - Small fixes like typos, simple warnings, adding/improving a comment

  - Adding and improving existing pages of the documentation

  - Simple improvements of style / elegance / efficiency (simple rewriting
a loop / condition / method interaction) if no behavior is changed

==> Basically anything that does not change or add functionality


*An issue is required for*

Everything else, in particular:

  - Anything that changes functionality or behavior relevant to users

  - Anything that changes functionality or behavior relevant to other
components

  - Anything that adds a feature


I would vote to allow coarse issues and have multiple commits that
reference it

[FLINK-1234] [runtime] Runtime support some cool new thing
[FLINK-1234] [java api] Add hook for cool thing to java api
[FLINK-1234] [scala api] Add hook for that thing to scala api
[FLINK-1234] [optimizer] Make optimizer aware that it can exploit this thing


-------------------------

The guide lines for pull-requests for committers are as follows:

*A pull request with comments/additional signoff is required for anything
that*

  - breaks the public APIs

  - adds methods to the public APIs (that will need to be kept stable from
them on)

  - alters user-facing behavior (e.g., mutability of types, null value
handling, window semantics, ...)

  - adds user-facing knobs (switches, config parameters, execution option
on the execution environment)

  - adds additional maven dependencies

  - changes the way components interact

  - touches highly sensitive and performance critical parts, such memory
management or network stack

==> Changes that come with a pull request should have one or more issues
associated with them.


Anyone that wants to have comments or some additional pairs of eyes in the
code should make a pull request as well.

-------------------------

*Naming scheme for commits*

[issue] [component] Message

For fixes without an issue, the issue can be dropped.




What do you think? Should we put this into the Wiki?


Greetings,
Stephan




On Wed, Jan 7, 2015 at 11:48 AM, Aljoscha Krettek <aljos...@apache.org>
wrote:

> Hi,
> I feel we never really talked about this. So, should we open Jira issues
> even for very small fixes and then add the ticket number to the commit? Or
> should we just commit those small fixes. Right now, I have two small fixes
> (one is 4 lines, the other one is two lines) for the ValueTypeInfo and
> TextValueInputFormat. Very obscure stuff, I know. :D
>
> Cheers,
> Aljoscha
>

Reply via email to