This is without my patch to change banned directory checking in issue 706466, right?

I'm synced to #464540 and using the clang that comes with that bundled.

I think part of this might be the fact that we're visiting vardecls, even in banned namespaces. I'm experimenting with a local CL that refactors how we do banned directory/namespace checking that I'm hoping will resolve this.

(see also issue 672115)

Yeah, I saw the other bug. Sadly, my experiments show that this probably doesn't have a major effect (or even has the opposite effect).

Is there a good way to profile clang to see where the time is going?

Any profiler should work. (Instruments on Mac is good, I guess people use perf on Linux, and ETW on Windows.) Make sure to attach to the cc1 subprocess.

This is with no changes:
dcheng@nausicaa:~/src/chrome/src$ ninja -C out/Release -t clean                                                                                                                                                              
ninja: error: remove(gen/chrome/browser/resources/safe_browsing/all): Directory not empty                                                                                                                    
ninja: error: remove(gen/chrome/browser/resources/settings/settings_resources.unpak): Directory notempty                                                                                    
Cleaning... 37908 files.                                                                                                                                                                        
dcheng@nausicaa:~/src/chrome/src$ time ninja -C out/Release chrome                                                                                                                        
ninja: Entering directory `out/Release'                                                                                                                                                                      
[31295/31295] LINK ./chrome                                                                                                                                                                                                   
                                                                                                                                                                                                                
real    48m33.806s                                                                                                                                                               
user    1807m34.716s                                                                                                                                                                                         
sys     72m17.232s     

And with the changes in https://chromium-review.googlesource.com/c/478907/ which were supposed to "help":
dcheng@nausicaa:~/src/chrome/src$ ninja -C out/Release -t clean                                                                                                                                                 
ninja: error: remove(gen/chrome/browser/resources/safe_browsing/all): Directory not empty                                                                                                           
ninja: error: remove(gen/chrome/browser/resources/settings/settings_resources.unpak): Directory notempty                                                                                                     
Cleaning... 37908 files.                                                                                                                                                                          
dcheng@nausicaa:~/src/chrome/src$ time ninja -C out/Release chrome                                                                                                                                             
ninja: Entering directory `out/Release'                                                                                                                                                                   
[31295/31295] LINK ./chrome                                                                                                                                                                                     
                                                                                                                                                                                     
real    49m8.400s                                                                                                                                                                                  
user    1803m14.484s                                                                                                                                                                                                                     
sys     97m18.408s 


I've tried profiling as well: with perf record + perf report off the binary generated by update.py:
- A lot of things aren't symbolized
- unmangling doesn't seem to work

Looking at the report, it reports 0.17% of build time is spent in libFindBadConstructs.so, which seems very little--so I'm surprised that the overall affect is about 0.5% when building Chrome. Maybe I'm interpreting the numbers incorrectly (I haven't done a baseline run with no plugin changes yet).

You can do a local RelWithDebInfo cmake build of clang to get more symbols, or add -g to your build flags similar to the cxxflags thingy in https://codereview.chromium.org/2709613004/diff/140001/tools/clang/scripts/update.py .

Issue has not been modified or commented on in the last 365 days, please re-open or file a new bug if this is still an issue.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot