[llvm-objdump] Add --build-id flag for debuginfod lookups without binary.

[mysterymath]

Summary

Adding a --build-id flag allows handling binaries that are referenced in
logs from remote systems, but that aren't necessarily present on the
local machine. These are fetched via debuginfod and handled as if they
were input filenames.

[mysterymath]

Nits.

[jhenderson]

New options need adding to the CommandGuide documentation. Please update the relevant llvm/CommandGuide file.

llvm/tools/llvm-objdump/ObjdumpOpts.td

42

Other options don't have a trailing full stop for their description in this file.

llvm/tools/llvm-objdump/llvm-objdump.cpp

2797–2798

Test case for this error?

2975–2976

Test case?

[mysterymath]

Add error tests, docs.

[jhenderson]

Two nits, otherwise looks good.

llvm/test/tools/llvm-objdump/debuginfod.test

40

On first glance, it wasn't clear why this one was malformed, and the next one is not found. Perhaps worth extending the comment here explaining how it is malformed (i.e. something like "not a hex string").

llvm/tools/llvm-objdump/llvm-objdump.cpp

3061–3065

Nit: to reduce git blame noise, you should be able to do this move as part of the change which introduces the debuginfod stuff, right?

[mysterymath]

Add comment explaining how build ID is malformed.
Move otool ordering to earlier commit in stack.

[MaskRay]

[objdump]

We commonly use [llvm-objdump]. objdump refers to the GNU binutils one.

llvm/test/tools/llvm-objdump/debuginfod.test

34

Optional: The lines in this file appear to be wrapped too aggressively. Two lines seem sufficient if you place --build-id on the previous line.

41

FileCheck line should not be indented.

47

FileCheck line should not be indented.

[mysterymath]

Address review comments.

llvm/docs/CommandGuide/llvm-objdump.rst

.. option:: --adjust-vma=<offset>
   Increase the displayed address in disassembly or section header printing by
   the specified offset.
 
 .. option:: --arch-name=<string>
 
   Specify the target architecture when disassembling. Use :option:`--version`
   for a list of available targets.
 
+.. option:: --build-id=<string>
+
+  Look up the object using the given build ID, specified as a hexadecimal
+  string. The found object is handled as if it were an input filename.
+
 .. option:: -C, --demangle
 
   Demangle symbol names in the output.
 
 .. option:: --debuginfod, --no-debuginfod
 
   Whether or not to try debuginfod lookups for debug binaries. Unless specified,
   debuginfod is only enabled if libcurl was compiled in (``LLVM_ENABLE_CURL``)

llvm/test/tools/llvm-objdump/debuginfod.test

 RUN:   llvm-objdump -d --source --no-debuginfod %t/stripped | \
 RUN:   FileCheck %s --check-prefix=NOTFOUND
 
 # Use debuginfod without URLs if requested.
 RUN: env DEBUGINFOD_CACHE_PATH=%t llvm-objdump -d --source --debuginfod \
 RUN:   %t/stripped | \
 RUN:   FileCheck %s --check-prefix=FOUND
 
+# Use debuginfod to look up build IDs.
+RUN: env DEBUGINFOD_CACHE_PATH=%t llvm-objdump -d --source \
+RUN:   --build-id 1512f769114c011387393822af15dd660c080295 | \
+RUN:   FileCheck %s --check-prefix=FOUND
+
+# Produce an error if malformed.
+RUN: not env DEBUGINFOD_CACHE_PATH=%t llvm-objdump -d --source \
+RUN:   --build-id foo 2> %t.err
+RUN:   FileCheck %s --check-prefix=MALFORMEDERROR --input-file %t.err
+MALFORMEDERROR: error: --build-id: expected a build ID, but got 'foo'
+
+# Produce an error if not found.
+RUN: not env DEBUGINFOD_CACHE_PATH=%t llvm-objdump -d --source \
+RUN:   --build-id abc 2> %t.err
+RUN:   FileCheck %s --check-prefix=NOTFOUNDERROR --input-file %t.err
+NOTFOUNDERROR: error: --build-id: could not find build ID 'abc'
+
 # Use debuginfod to replace very-stripped binaries entirely.
 RUN: llvm-strip --strip-sections %t/stripped
 RUN: env DEBUGINFOD_CACHE_PATH=%t llvm-objdump -d --debuginfod \
 RUN:   %t/stripped | \
 RUN:   FileCheck %s --check-prefix=SYMBOLS
 
 SYMBOLS: <main>:
 
 FOUND: int main(int argc, char *argv[]) {
 NOTFOUND-NOT: int main(int argc, char *argv[]) {

llvm/tools/llvm-objdump/ObjdumpOpts.td

 def : Flag<["-"], "x">, Alias<all_headers>, HelpText<"Alias for --all-headers">;
 
 def arch_name_EQ : Joined<["--"], "arch-name=">,
   HelpText<"Target arch to disassemble for, "
             "see --version for available targets">;
 def archive_headers : Flag<["--"], "archive-headers">,
   HelpText<"Display archive header information">;
 
+defm build_id :
+  Eq<"build-id", "Build ID to look up. Once found, added as an input file">,
+  MetaVarName<"<dir>">;
+
 def : Flag<["-"], "a">, Alias<archive_headers>,
   HelpText<"Alias for --archive-headers">;
 
 def demangle : Flag<["--"], "demangle">, HelpText<"Demangle symbol names">;
 def : Flag<["-"], "C">, Alias<demangle>, HelpText<"Alias for --demangle">;
 
 defm debug_file_directory :
   Eq<"debug-file-directory", "Path to directory where to look for debug files">,

llvm/tools/llvm-objdump/llvm-objdump.cpp

if (const opt::Arg *A = InputArgs.getLastArg(ID)) {
     if (!llvm::to_integer(V, Value, 0)) {
       reportCmdLineError(A->getSpelling() +
                          ": expected a non-negative integer, but got '" + V +
                          "'");
     }
   }
 }
 
+static object::BuildID parseBuildIDArg(const opt::Arg *A) {
+  StringRef V(A->getValue());
+  std::string Bytes;
+  if (!tryGetFromHex(V, Bytes))
+    reportCmdLineError(A->getSpelling() + ": expected a build ID, but got '" +
+                       V + "'");
+  ArrayRef<uint8_t> BuildID(reinterpret_cast<const uint8_t *>(Bytes.data()),
+                            Bytes.size());
+  return object::BuildID(BuildID.begin(), BuildID.end());
+}
+
 static void invalidArgValue(const opt::Arg *A) {
   reportCmdLineError("'" + StringRef(A->getValue()) +
                      "' is not a valid value for '" + A->getSpelling() + "'");
 }
 
 static std::vector<std::string>
 commaSeparatedValues(const llvm::opt::InputArgList &InputArgs, int ID) {
   std::vector<std::string> Values;
for (StringRef V : Values) {
         DisassemblerOptions.push_back(V.str());
     }
   }
   if (AsmSyntax) {
     const char *Argv[] = {"llvm-objdump", AsmSyntax};
     llvm::cl::ParseCommandLineOptions(2, Argv);
   }
 
+  // Look up any provided build IDs, then append them to the input filenames.
+  for (const opt::Arg *A : InputArgs.filtered(OBJDUMP_build_id)) {
+    object::BuildID BuildID = parseBuildIDArg(A);
+    Optional<std::string> Path = BIDFetcher->fetch(BuildID);
+    if (!Path) {
+      reportCmdLineError(A->getSpelling() + ": could not find build ID '" +
+                         A->getValue() + "'");
+    }
+    InputFilenames.push_back(std::move(*Path));
+  }
+
   // objdump defaults to a.out if no filenames specified.
   if (InputFilenames.empty())
     InputFilenames.push_back("a.out");
 }
 
 int main(int argc, char **argv) {
   using namespace llvm;
   InitLLVM X(argc, argv);
if (InputArgs.hasArg(VersionFlag)) {
     cl::PrintVersionMessage();
     if (!Is("otool")) {
       outs() << '\n';
       TargetRegistry::printRegisteredTargetsForVersion(outs());
     }
     return 0;
   }
 
-  if (Is("otool"))
-    parseOtoolOptions(InputArgs);
-  else
-    parseObjdumpOptions(InputArgs);
-
   // Initialize debuginfod.
   const bool ShouldUseDebuginfodByDefault =
-      HTTPClient::isAvailable() &&
-      !ExitOnErr(getDefaultDebuginfodUrls()).empty();
+      InputArgs.hasArg(OBJDUMP_build_id) ||
+      (HTTPClient::isAvailable() &&
+       !ExitOnErr(getDefaultDebuginfodUrls()).empty());
   if (InputArgs.hasFlag(OBJDUMP_debuginfod, OBJDUMP_no_debuginfod,
                         ShouldUseDebuginfodByDefault)) {
     HTTPClient::initialize();
     BIDFetcher = std::make_unique<DebuginfodFetcher>(DebugFileDirectories);
   } else {
     BIDFetcher = std::make_unique<BuildIDFetcher>(DebugFileDirectories);
   }
 
+  if (Is("otool"))
+    parseOtoolOptions(InputArgs);
+  else
+    parseObjdumpOptions(InputArgs);
+
   if (StartAddress >= StopAddress)
     reportCmdLineError("start address should be less than stop address");
 
   // Removes trailing separators from prefix.
   while (!Prefix.empty() && sys::path::is_separator(Prefix.back()))
     Prefix.pop_back();
 
   if (AllHeaders)