On 10.09.2020 15:51, Andrew Cooper wrote: > On 10/09/2020 13:09, Jan Beulich wrote: >> While looking at what it would take to move around libelf/ >> in the hypervisor subtree, I've run into this rule, which I >> think can do with a few improvements and some simplification. > > I realise this might be a controversial move, but can we *please* > address this by removing our use of symlinking, rather than kicking the > problem down the road. > > For header files in particular, there is no need to symlink at all. > Some properly formed -I runes for the compiler will do the right thing, > and avoid all intermediate regeneration issues.
With some further work to separate headers in e.g. Xen's acpi/ into ones to be exposed and ones not to be exposed, this would likely be an option. It's not clear to me though how you mean to deal with libelf.h and elfstructs.h. Nor is it clear how we'd deal with x86's cpuid-autogen.h, which needs to have distinct instances in the two subtrees. And of course the present full exposure of asm-x86 rather wants limiting than setting in stone by using -I to point into the hypervisor tree. Installing of the headers into dist/ will also need re-working then. Taking things together - no, I don't think I'm up to doing this, yet I think the series presented is an improvement. Jan
