Open nico opened 8 years ago
I looked at one other bot (maybe https://build.chromium.org/p/chromium.fyi/stats/ClangToTWin(dll) ?) and combined the regression ranges from that bot with the linux one. The upper bound is from the linux bot, the lower one from the other bot.
Are you sure the Window is right? Based on the linked bot, these are the two builds bracketing the regression: https://build.chromium.org/p/chromium.fyi/builders/ClangToTLinux/builds/5130 https://build.chromium.org/p/chromium.fyi/builders/ClangToTLinux/builds/5129
The clang window for that is (r268787, r268820].
r268806 doesn't seem like a plausible culprit given that we don't run tests.
I don't see a good culprit on the LLVM side (git log 3a963d121a2e690698987c175ca9760007e87935...9c0ef542c413c898248f2d682851c6b56314673a).
We don't run LLVM tests on these bots.
Here's a graph: https://build.chromium.org/p/chromium.fyi/stats/ClangToTLinux (scroll all the way to the bottom to "runhooks")
Extended Description
We have a bunch of bots that basically do:
svn up
llvm, run cmake, runninja
to build all of llvmThe time of the first step increased by about 20% on all our bots. The regression range is r268806:268820 (both ends inclusive). Maybe r268806 makes things slow for some reason?