[Lldb-commits] [PATCH] D47235: Move ModuleList's dependency on clangDriver into Host

[Pavel Labath via Phabricator via lldb-commits]

labath added a comment.

In a way, this makes sense, but in a lot of other ways, it actually makes things worse :)

My long-term goal is to be able to build lldb-server without even having clang checked out (because it doesn't really need anything clang-related), and this would certainly make that impossible. Having Core depend on clang does not worry me much because it depends on so many things already, so the way I was thinking of resolving that is to move low-level things out of it (my tentative list includes things like Event/Listener/Broadcaster, State, Communication). That would leave Core with containing only the high-level stuff for which a clang dependency is not that surprising (although I agree that a ModuleList is not the best place for such a dependency).

I guess it would be nice to encapsulate this in some sort of a plugin (since the setting is used from the clang expression parser plugin, I guess this would be the natural home for it) , but I haven't looked in detail at could that work. What I do know is that we already have the ability to inject settings from within a plugin (see SymbolFileDWARF::DebuggerInitialize). Maybe that would work here too?

PS: This patch would make the clang modules path similar to the python directory computation, but this is also something that I've been meaning to change somehow (lldb-server does not need python either).



================
Comment at: source/Core/ModuleList.cpp:88
 
-  llvm::SmallString<128> path;
-  clang::driver::Driver::getDefaultModuleCachePath(path);
-  SetClangModulesCachePath(path);
+  SetClangModulesCachePath(HostInfoBase::GetDefaultClangModuleCachePath());
 }
----------------
You should reference this through the `HostInfo` typedef to get the "static polymorphism" to work.


https://reviews.llvm.org/D47235