[PATCH] D99305: [docs] Document our norms around reverts

[Joseph Tremoulet via Phabricator via llvm-commits]

JosephTremoulet added inline comments.


================
Comment at: llvm/docs/DeveloperPolicy.rst:306
+
+As a community, we strongly value having the tip of tree in a good state.  As
+such, we tend to make much heavier use of reverts than some other open source
----------------
FWIW, coming from a culture of gated check-ins where only the automation account has commit permissions and there's often complex infrastructure to manage the queue of pending changes, the thing I had to learn about LLVM and have to explain to newcomers is actually kind of the opposite of this first sentence -- that the LLVM community values the ability to freely commit above keeping inner-loop tests green at all times.  And that this has some great benefits like facilitating breaking reviews/changes into small incremental pieces, fostering a culture of trust and accountability, etc.  And that the commit rate in LLVM is quite high (looking at git stats for the past year, there were an average of 95 commits per day, with a peak of 172 commits in one day) to facilitate that, and adding commit barriers would impede that.

Then after all that is when we get to "how does testing work?" and that often it's only mostly green and there are breaks and reverts every day and it's not considered a big deal but rather just considered a fact of a healthily active codebase.

I don't know if and how much of that is worth including here, and apologies if I'm speaking out of turn and mischaracterizing the community, but it struck me that this seems to be written for the audience asking "how dare you revert my commit?!" and wanted to note there's another school out there more likely to ask "how dare you push straight to master?!" :)


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D99305/new/

https://reviews.llvm.org/D99305