Fix PNaCl bitcode to LLVM IR reader to match new API.

src/llvm2ice.cpp

Index: src/llvm2ice.cpp
diff --git a/src/llvm2ice.cpp b/src/llvm2ice.cpp
index a28b07f924c61a3b9dd45a17a72f30746014e27c..d92ee06fe1043a2cbd7b71c9f461a12ee814d26f 100644
--- a/src/llvm2ice.cpp
+++ b/src/llvm2ice.cpp
@@ -165,6 +165,14 @@ static cl::opt<bool>
                 cl::desc("Build ICE instructions when reading bitcode"),
                 cl::init(true));
 
+
+static cl::opt<bool>
+LLVMVerboseErrors(
+    "verbose-llvm-parse-errors",
+    cl::desc("Print out more descriptive PNaCl bitcode parse errors when "
+             "building LLVM IR first"),
+    cl::init(false));

[Jim Stichnoth, 2014/12/03 20:35:15]

I realize the default false value is consistent with the LLVM-side changes, but
why wouldn't we by default want more verbose, more awesome error messages?

[Karl, 2014/12/03 20:57:29]

I didn't worry about the verbose messages. Verbose messages give more detail
about the record that was bad. However, they do not give you a location within
the bitcode file (making it hard to find, even using pnacl-bcdis).

In addition, the use case in subzero for building LLVM IR is based on using
frozen valid LLVM IR. Hence, there shouldn't be any error messages in the first
place.

+
 static cl::opt<bool>
     UseIntegratedAssembler("integrated-as",
                            cl::desc("Use integrated assembler (default yes)"),
@@ -308,8 +316,10 @@ int main(int argc, char **argv) {
     // Parse the input LLVM IR file into a module.
     SMDiagnostic Err;
     Ice::TimerMarker T1(Ice::TimerStack::TT_parse, &Ctx);
+    raw_ostream *Verbose = LLVMVerboseErrors ? &errs() : nullptr;
     Module *Mod =
-        NaClParseIRFile(IRFilename, InputFileFormat, Err, getGlobalContext());
+        NaClParseIRFile(IRFilename, InputFileFormat, Err, Verbose,
+                        getGlobalContext());
 
     if (!Mod) {
       Err.print(argv[0], errs());