> On March 31, 2015, 1:36 a.m., Ben Mahler wrote: > > Great work Cody! For posterity, how did you pinpoint flags.hpp?
Manual experimentation a while ago with Joris. We were searching for general compile time improvements. A conversation recently reminded me of this and that we never pushed a patch for it. Checked again, and saw really we should try to land it. General process was: Find a couple files that take a long time to compile (Where does the command printing of make sit for a long time), comment out the body (#if 0 #endif), see if it still takes a while. If so, binary search through the headers and figure out which ones are worst. Go into those headers, repeat until you find the root file code. - Cody ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/32558/#review78303 ----------------------------------------------------------- On March 27, 2015, 1:21 a.m., Cody Maloney wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/32558/ > ----------------------------------------------------------- > > (Updated March 27, 2015, 1:21 a.m.) > > > Review request for mesos, Joris Van Remoortere and Michael Park. > > > Bugs: MESOS-292 > https://issues.apache.org/jira/browse/MESOS-292 > > > Repository: mesos > > > Description > ------- > > Split the mesos master, slave flags into header + source file to > improve compile time significantly. Should be no functional changes. > > Largely copy-paste, with a little reworking to remove unnecessary > headers from the .hpp, and order headers a little more reliably > (as well as remove duplicate includes) in the .cpp. > > # Impact of these changes > `time make check -j8` > Before: > make check 2732.93s user 103.89s system 514% cpu 9:11.63 total > > After: > make check 2421.18s user 96.60s system 506% cpu 8:16.67 total > > 4 core machine, 8 hypter threads enabled. SSD, 16 GB RAM, > Intel i7-4790K. > > The numbers aren't incredibly stable (Other software running, > overclocking enabled, etc). They are a good general measure though of > speedup. > > I do a `make check` rather than just `make` because that is what devs > do in a day-to-day flow, and if runtime is significantly impacted, it > will show up. > > Test steps: > ``` > # Warm cache > ../configure --disable-python --disable-java > make check > # Timed build > rm -rf * > ../configure --disable-python --disable-java > time make check > ``` > > Note: Similar, likely greater improvements in compile time would happen > if stout were made to not be header only. Specifically <stout/os.hpp>. > > > Diffs > ----- > > src/Makefile.am 7a06c7028eca8164b1f5fdea6a7ecd37ee6826bb > src/logging/flags.hpp 4facb33201cee5a82451a13ca05607c875574752 > src/logging/flags.cpp PRE-CREATION > src/master/allocator/allocator.hpp b67b8fddbd7a3fffc6fe24d5e77cd1db8cb6f69b > src/master/flags.hpp 85ce3285731c11b464aca6eaaa0a9165c73afebd > src/master/flags.cpp PRE-CREATION > src/slave/flags.hpp 3da71afad38ae41adefab979dbed2ae0b10a98ef > src/slave/flags.cpp PRE-CREATION > > Diff: https://reviews.apache.org/r/32558/diff/ > > > Testing > ------- > > make check on ArchLinux with GCC 4.9.2 > make distcheck CentOS 7 > > > Thanks, > > Cody Maloney > >
