| Commit-Queue | +1 |
Andrew GrieveSame concern as the other CL: does this GN arg have any expected impact on compile/link times? I'm unfamiliar with the arg, but I imagine it would lead to smaller binaries -> faster links?
Sorry, didn't realize at first that changing it for CI would cause it to change the CQ bots.
But... I think this shouldn't have much impact, as strip_debug_info applies at the link step, but with symbol_level=0, there isn't any debug info to strip in the first place.
This should only impact bots that set symbol_level != 0
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Andrew GrieveSame concern as the other CL: does this GN arg have any expected impact on compile/link times? I'm unfamiliar with the arg, but I imagine it would lead to smaller binaries -> faster links?
Sorry, didn't realize at first that changing it for CI would cause it to change the CQ bots.
But... I think this shouldn't have much impact, as strip_debug_info applies at the link step, but with symbol_level=0, there isn't any debug info to strip in the first place.
This should only impact bots that set symbol_level != 0
Rebased my other CL on this one since they conflict and my guess is that this is less likely to cause issues
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
| Auto-Submit | +1 |
ping - okay to go ahead with this one?
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Remove strip_debug_info from Android bots that run tests
It causes tests to not show line numbers in stack traces
Before: https://ci.chromium.org/ui/p/chromium/builders/try/android-x64-rel/892829/overview
After: https://ci.chromium.org/ui/p/chromium/builders/try/android-x64-rel/893135/overview
Note that many bots still set symbol_level=0, which makes
strip_debug_info effectively a no-op anyways.
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Actually this is not a no-op change. The total size of compiled artifacts is almost 3x now:
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Actually this is not a no-op change. The total size of compiled artifacts is almost 3x now:
- Before in https://ci.chromium.org/ui/p/chromium/builders/ci/android-10-x86-rel/7730/infra: is about 26GB, took 18 mins (See the "stdout" of the step "isolate tests")
- After in https://ci.chromium.org/ui/p/chromium/builders/ci/android-10-x86-rel/7732/infra: is about 88GB, took 40 mins.
That bot sets symbol_level=1, so yeah, it would affect it. It should only not have an effect for bots that set symbol_level=0.
We should really never use strip_debug_info on Android, and I'm not really sure why the flag exists tbh. Setting symbol_level=1 adds overhead to every compile to create debug info, and then strip_debug_info strips the debug info when linking.
E.g. If we're going to throw away the debug info, it's much more efficient to set symbol_level=0 so that the debug info is not generated in the first place.
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Andrew GrieveActually this is not a no-op change. The total size of compiled artifacts is almost 3x now:
- Before in https://ci.chromium.org/ui/p/chromium/builders/ci/android-10-x86-rel/7730/infra: is about 26GB, took 18 mins (See the "stdout" of the step "isolate tests")
- After in https://ci.chromium.org/ui/p/chromium/builders/ci/android-10-x86-rel/7732/infra: is about 88GB, took 40 mins.
That bot sets symbol_level=1, so yeah, it would affect it. It should only not have an effect for bots that set symbol_level=0.
We should really never use strip_debug_info on Android, and I'm not really sure why the flag exists tbh. Setting symbol_level=1 adds overhead to every compile to create debug info, and then strip_debug_info strips the debug info when linking.
E.g. If we're going to throw away the debug info, it's much more efficient to set symbol_level=0 so that the debug info is not generated in the first place.
I checked before this that all of the Android trybots set symbol_level=0, except for android-binary-size. I think the symbol_level=0 setting is in the "release_trybot" config, so probably non-trybots have symbol_level=1.
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Andrew GrieveActually this is not a no-op change. The total size of compiled artifacts is almost 3x now:
- Before in https://ci.chromium.org/ui/p/chromium/builders/ci/android-10-x86-rel/7730/infra: is about 26GB, took 18 mins (See the "stdout" of the step "isolate tests")
- After in https://ci.chromium.org/ui/p/chromium/builders/ci/android-10-x86-rel/7732/infra: is about 88GB, took 40 mins.
Andrew GrieveThat bot sets symbol_level=1, so yeah, it would affect it. It should only not have an effect for bots that set symbol_level=0.
We should really never use strip_debug_info on Android, and I'm not really sure why the flag exists tbh. Setting symbol_level=1 adds overhead to every compile to create debug info, and then strip_debug_info strips the debug info when linking.
E.g. If we're going to throw away the debug info, it's much more efficient to set symbol_level=0 so that the debug info is not generated in the first place.
I checked before this that all of the Android trybots set symbol_level=0, except for android-binary-size. I think the symbol_level=0 setting is in the "release_trybot" config, so probably non-trybots have symbol_level=1.
I see. Thanks. I missed the symbol_level flag.
Setting symbol_level=1 adds overhead to every compile to create debug info, and then strip_debug_info strips the debug info when linking.
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |