About an IRC discussion on using pull requests for hg development

[Augie Fackler]

On May 9, 2012, at 10:54 AM, Levi Bard wrote:

>> <mpm>
>> Since review must happen on the list, and everything should be
>> reviewed, code submission needs to happen on the list. And raising the
>> bar for starting a review above hitting reply is a big step backward.
>> 
>> (Also remember that it's handy for me to be able to cherrypick
>> submissions and send pieces back for reworking.)
> 
> This conclusion also follows the flawed logic from the previous snippet.
> A ML is a terrible forum for code review, and an even worse forum for
> code submission. This is a workflow that requires every involved party
> to have an ideally configured mail client for viewing and reviewing
> code precisely the way the mercurial project does it and to have
> adapted to the workflow of interacting with reviews of series of
> patches in this way.
> A web review system allows all the configuration to happen on the
> server, and everyone gets the same interface to and experience with
> the review.
> I also agree that one should just be able to reply via mail to a
> notification from the system, and have that aggregated into the web
> view of the review. And if anything, it should be _easier_ to
> cherrypick changesets from a repo and get the benefits of merge
> machinery, etc., than to manually apply patches from a mail.
> 
> Benefits of a hosted review system over a ML:
> * Code and diff highlighting (for everyone)
> * Ability to transform the view of the changes being reviewed
> (per-changeset, aggregate diff, etc.)
> * Ability to review non-text changes (largefiles?)
> * Ability to navigate to different changesets/source files/etc. via
> hyperlinks directly from the review context
> * Familiarity for free software developers who have interacted with
> Bitbucket, Launchpad, Github, and friends
> * The entire process is public and transparent from end to end:
>   * Current process: Write patch => patchbomb to list => review =>
> "queued" announcement => ??? => patch lands in hg repo, unless
> something goes wrong
>   * Web review process: Write patch => push => review =>
> merged/rebased to hg repo

For those of us that read unified diff well, the existing publicly available code review tools are just depressing. I find myself wishing for emailed patchbombs instead of Github's pull requests almost every time I encounter one. If you restrict your set of review tools to the ones that are open source, it's even more dismal. Nothing that I'm aware of in the open source world makes it possible to group a set of changesets as a review yet still supports looking at the individual diffs.