On Tue, 12 Aug 2025 10:49:19 GMT, Magnus Ihse Bursie <i...@openjdk.org> wrote:
> Your updated result seemed more reliable and probable, but they also do not > differ that much from the baseline. I still saw a few improvements and no > regression, so it might still be worth pursuing, but maybe your suggested > change does not bring as much improvement as was first hinted at. I'd say individual improvements are a bit harder to observe in the kind of measurements I'm doing now, and the likelihood of measuring noise is higher. Also, the minimum inclusion threshold I'm using now is the same for all features, but that's unlikely to be optimal. Maybe we could revert to the original approach (count inclusion for a typical build)? That proved to improve compile time, and I could hide non-standard headers behind an `#if`. ------------- PR Comment: https://git.openjdk.org/jdk/pull/26681#issuecomment-3179118681
