The URL /p/xbmc/xbmc/ci/master/tree/xbmc/ (don't bother going to it) always times out. We don't have instrumentation results from it (yet). The repo is git://git.code.sf.net/p/xbmc/xbmc/ which is 1.4G, 18 top level dirs, 20 top level files, 1902 dirs total, 19278 files total. It consumes CPU, RAM, and time.
[#6695] for short-term safety guards.
Related
Tickets:
#6695allura:cj/6686
Changed Git to generate last_commit_ids indexlessly instead of making a ton of round-trips to mongo. Seems to be around 10x faster in my testing. This isn't a fully indexless browsing implementation because there's still a lot of model use in controller dispatch and getting the commit info. The latter could be had from git at the same time as the last_commit_ids for no extra cost, but would require more refactoring to surface it, and for the moment it's only a single mongo query so shouldn't too bad of a hit.
NB: This mostly only affects building LCDs on demand.
no_merges=True
in the git log call? Previously we did show merges as last commits.I'm pretty sure people complained that showing merges as the last commit was a bug. I don't think it's very standard. That said, the
no_merges=True
may not even be necessary; I only added it because I remembered having to fix (or possibly not having gotten to it yet) merge commits showing up as the LCD for a file.Ok, that's fine. Mostly wanted to make sure it was intentional.
[#5734] might be the ticket you were thinking of.
We also had [#6629] which was a problem with merge requests having different output from git that had to be parsed differently.
Related
Tickets: #5734
Tickets:
#6629Rebased and force-pushed to:
allura:cj/6686