On Mon, Apr 5, 2021 at 6:00 AM Stefan Fuhrmann <stef...@apache.org> wrote:
> Hi all, > > Way back in 2014, I started work on an SVN equivalent to apr_thread_pool > and came back to it recently. The key features are output and callbacks > happen in the same order as in sequential code, same for any svn_error_t > raised, and a low-overhead single-threaded code path. > > Now, there is a working prototype and I used it in > svn_wc__internal_walk_status. > Integration was simple enough and scaling is excellent, both for cold > and hot > file caches. I'm on staycation this week, so I would like to bring the > code > to trunk (with proper documentation and all). But beyond that, I don't > know > how responsive I can be in case things go wrong. So, I would very much > like > somebody dedicate time to reviewing it this week, give feedback etc. > > Any takers? > > -- Stefan^2. Now is actually a good time to introduce this because we have other heavy duty optimizations on trunk, and it would be a good thing if we could get back on track with regular feature releases between LTS releases, with something "marketable" in them, which I think this qualifies as. 1.15 could be summed up as a "faster performance" release. The only reason I can think of why anyone might object is if the changes add new dependencies. If you could post a patch of what you have so far, even if incomplete, it will help us see how big (or small) the proposed changes are and what they entail. I'm willing to give the code an extra pair of eyeballs, build and test, and if it gets the blessing of the current test suite I'm willing to build trunk with the new code and try it as my daily driver to put it through its paces in real world (client side) usage. If you're concerned that you won't be able to work on it after this week, you could opt to put it on a branch, but I understand that it might languish there. Your call... Cheers, Nathan
