Closed steveri closed 6 years ago
Because there are changes to the flow, the following should happen roughly at the same time:
-Do one final commit in this CGRAFlow branch changing the names of the P&R/CGRA branches back to 'master'. -Merge this pull request. -Merge P&R and CGRA branches into their own master. -Verify that master for the flow did not break.
Re: "the following should happen roughly at the same time"
Actually I was thinking we could preserve a 'never-broken' master by doing the following:
check in this master, which is working, and which points to 'nbdev2' and 'io_tile' branches.
At their leisure (but presumably soon) Steve merges CGRAGenerator/nbdev2 to CGRAGenerator/master and Caleb or Makai merges SMT/io_tiles to SMT/master such that now
To that end, I will go ahead and merge this pull to get the ball rolling...
SR
I would vote for rewording at their leisure to ASAP. I believe we already saw a potential issue related to stale master branches (issue with taeyoung's base branch). In terms of methodology, I think it's best practice to always keep master fresh. This way, a team member can branch off master with confidence that it's up to date, and they don't have to hunt down the right branch to base off of.
I would also vote that we don't merge into CGRAFlow master with the travis script pointing to non-master branches (again to promote the idea that individual repositories are master is fresh)
I agree that branches should be merged ASAP. PNR io_tiles branch is ready to be merged. In keeping with other good practices, I'm having @cdonovick review and commit my changes. I think he's busy today, but I anticipate that will be merged by EOD Tuesday. Regarding CGRAFlow master pointing to non-master branches, I agree it's probably better not to. I don't think it's worth reverting though.
Sorry, I guess I jumped the gun. FWIW at this point I merged CGRAGenerator/nbdev2 and CGRAGenerator/master, and CGRAFlow is currently pointing to CGRAGenerator/master and passing (build 1209). Looks like SMT branches will be sorted out soon, and maybe we can move forward from there.
In keeping with the will of the crowd, I will try and remember to order it differently next time around... :)
There's a third option, I guess, that may or may not make everyone happy. It preserves the never-break rule for master and also the (new?) master-only-points-to-master-sub-repositories rule. Next time we could do the following, unless people decide it's too complicated and/or prefer a different path:
PS On the previous comment, in case it wasn't obvious, in our current situation it would have been the case that 'staging' == 'nbdev2'. Which I guess is what Ross was proposing all along, wasn't it :)
That seems like a reasonable path. Should be possible to write a script that uses the GitHub API to perform those sequence of merges in order once the original (step 1) staging branch is green. I'd be willing to take that on once I get some time
P&R is merged.
Okay I changed CGRAFlow/master to point to ONLY MASTER BRANCHES, maybe now the world will be right again :)
master build passed, maybe we're finally done with this thread?
This is phase 1 of the master plan, see email for details. These changes should suffice to make all our existing tests (8x8 pointwise, conv12, conv21, conv31, convbw) pass in travis using SMT-PNR along with the latest version of the CGRA generator (nbdev2).