[PATCH] D19096: ThinLTO/ModuleLinker: add a flag to not always pull-in linkonce when performing importing

[Mehdi AMINI via llvm-commits]

joker.eph added inline comments.

================
Comment at: lib/Transforms/IPO/FunctionImport.cpp:424
@@ +423,3 @@
+    // Instruct the linker that the client will take care of linkonce resolution
+    unsigned Flags = Linker::Flags::DontForceLinkLinkonceODR;
+
----------------
tejohnson wrote:
> This should be passed down from the client, not hardwired here. E.g. this resolution is not yet done in gold. E.g. pass down to importFunctions() or to FunctionImporter constructor.
Will update

================
Comment at: test/Linker/funcimport2.ll:6
@@ -5,3 +5,3 @@
 
-; CHECK: define linkonce_odr hidden void @foo() {
+; CHECK: define available_externally hidden void @foo() {
 define available_externally hidden void @foo() {
----------------
tejohnson wrote:
> Looks like this change should be part of D19308 instead?
I don't think so: it is testing that we don't import foo (we're still seeing the original definition). So it is not about the resolution in D19308 

================
Comment at: test/ThinLTO/X86/Inputs/alias_resolution.ll:5
@@ +4,3 @@
+
+ at globalfuncAlias = alias void (...), bitcast (void ()* @globalfunc to void (...)*)
+ at globalfuncWeakAlias = weak alias void (...), bitcast (void ()* @globalfunc to void (...)*)
----------------
tejohnson wrote:
> Looks like this was accidentally left in this patch (test/ThinLTO/X86/alias_resolution.ll is gone). Should this test now be part of D19308?
Yes,


http://reviews.llvm.org/D19096