[libcxx-commits] [libcxx] Commit mostly working single wrapped module map (PR #87402)

[David Blaikie via libcxx-commits]

dwblaikie wrote:

> The less fortunate news is that it appears the current approach, without the top level std module, is something like 25% slower to build llvm-tblgen (for example) and produces a module cache about 1.8x the size (205M vs 123M). than the approach with the top level std module.

Generally one needs to modularize from the leaves/bottom-up - anything less than that will produce a lot of duplication in the intermediate modules due to the leaves not being modularized. (or creating something more like a PCH - a single module that includes a bunch of non-modular code, but then all modular code that needs any of the non-modular code always goes through that single module (essentially any situation where significant amounts of (especially C++) non-modular code is depended on by more than one module is a bad thing, and should be avoided))

https://github.com/llvm/llvm-project/pull/87402