One year of frame pointers - how do people feel?

At the time there were some very strong opinions about frame pointers, both for and against. There were concerns that it would have an unacceptable impact on performance, and claims about the potential benefits of easy profiling.

Certainly some of those claims materialized (several upstream packages such as Python required fixes to address the performance degradation, but also several core packages have received notable optimizations as a result of frame-pointer-enabled profiling).

I’m mostly interested in how opinions have changed now that the impact is concrete rather than speculative.

If you opposed them at the time, do you still oppose them now? Does the Fedora toolchain team still oppose them, and does that opposition still extend to e.g. CentOS Stream or RHEL? How many packages actually suffered performance regressions that weren’t trivially resolved, or disabled frame pointers due to observed performance challenges?

1 Like