[Infrastructure] Previewing pull requests

Ryan Sleevi sleevi at google.com
Tue May 12 07:52:57 MST 2020


On Tue, May 12, 2020 at 7:57 AM Dimitris Zacharopoulos (HARICA) <
dzacharo at harica.gr> wrote:

> Thanks Ryan for the detailed description and suggestions for moving
> forward. Please see below.
>
> On 2020-05-07 9:48 μ.μ., Ryan Sleevi wrote:
>
> On our last call, we talked a little about the workmode for previewing
> documents / edits.
>
> After talking around to a few of our other standards folks, it looks like
> the order I described is roughly correct.
>
> That is, our "shovel ready" solution is to do so via "trusted" branches in
> the main cabforum/documents repository, which we've already got Travis
> set-up to automatically build and deploy artifacts from. They're "trusted"
> because anyone with a branch in cabforum/documents is implicitly trusted to
> just modify all the build system anyways :)
>
> However, to have artifacts for pull requests, we need to split out the
> build/deploy steps. This is discussed a little at
> https://docs.travis-ci.com/user/pull-requests#pull-requests-and-security-restrictions .
> From looking around at other SDOs, the common approach seems to be setting
> up a web hook for inbound PRs, doing an isolated build of the artifacts,
> and then doing the deploy step from there.
>
> The "isolated build" can be done a number of ways. The easiest way seems
> to be just to only fetch the .md files from arbitrary ("untrusted") pull
> requests and run those through the cabforum/document's build scripts, and
> then deploy those generated artifacts. If we want to change the build
> scripts / templates themselves, those would go through the trusted branches
> above.
>
> Practically speaking, this means the order of operations is:
>
>    - Jos sends a PR for the Makefile invocation to generate our artifacts
>    - Jos and Ryan work through getting the dependencies right for both
>    our dockerfile (for local editing) and the Travis bits (for the automated
>    building)
>
>
> This enables the following workflow:
>
>    1. Dimitris (or a chair/delegate, implicit throughout this remainder),
>    when getting ready to merge a change, changes the base branch
>    <https://github.blog/2016-08-15-change-the-base-branch-of-a-pull-request/>
>    from being cabforum/documents:master into cabforum/documents:[something
>    temporary Dimitris will use)
>    2. Dimitris squash&merges the pull request into that branch and apply
>    the necessary corrections/edits
>    3. Travis is generating artifacts for every commit into this branch,
>    which can be previewed for correctness
>    4. Dimitris creates a pull request from cabforum/documents:[something
>    temporary] into cabforum/documents:[master]
>    5. Dimitris squash&merges into master (editing commit message as
>    necessary)
>    6. Travis will again kick off new artifacts, this time for the commit
>    to master
>    7. Dimitris deleges the cabforum/documents:[something temporary]
>    branch, as it's no longer used
>    8. [Optional] Dimitris tags cabforum/documents:master @ that commit as
>    being the appropriate document number (e.g. BR 1.6.7), creating multiple
>    tags if multiple documents were updated (e.g. the same commit is tagged BR
>    1.9.1 and EVG 1.8.1, if both documents were updated)
>
>
> This works for me and I assume the same applies to Wayne.
>
>
> Longer term, the following steps would be necessary
>
>    - Setup a webhook to fetch the .md files from a PR and run through the
>    make/deploy setup
>    - Wire the webhook to cabforum/master to fire when PRs are opened by
>    anyone
>
>
> And then the workflow just becomes:
>
>    1. Someone opens up a PR, enabling cabforum/documents maintainers to
>    make edits
>    2. Voting happens
>    3. Dimitris makes edits directly into the PR's source branch
>    4. Previews are generated of those edits
>    5. Squash & Merged into cabforum/master
>    6. Documents published
>    7. [Optional] Tags created
>
>
> How much additional effort do you think it would take to develop the
> second solution? If it's not feasible for the next couple of weeks, we can
> go with the first workflow and come back to it later.
>

More time than I have in the next few weeks :)


>
> It would also help if we made some progress with cleaning up the official
> repo from the various old branches, as we agreed in Bratislava
> https://cabforum.org/pipermail/infrastructure/2020-February/000173.html
>
>
> Thanks,
> Dimitris.
> _______________________________________________
> Infrastructure mailing list
> Infrastructure at cabforum.org
> http://cabforum.org/mailman/listinfo/infrastructure
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cabforum.org/pipermail/infrastructure/attachments/20200512/e2db8e96/attachment.html>


More information about the Infrastructure mailing list