[ELF] - Fix incorrect object file use with --reproduce option.

[grimar]

Summary

This patch is mostly for discussion, it is not ideal, but I have no better solution in mind atm.

I observed this issue few times earlier, but last days I am facing it too often when running
other people reproduces.

Problem is next. I can following invocation under ubuntu:
~/LLVM/llvm-build/bin/clang-5.0 -fuse-ld=lld test.c -l stdc++ -o out -v -Wl,--reproduce,out.tar

Result out.tar contains usr\lib64\libc.so which is a linker script and has GROUP command:
GROUP ( /lib64/libc.so.6 /usr/lib64/libc_nonshared.a AS_NEEDED ( /lib64/ld-linux-x86-64.so.2 ) )

Our --reproduce logic correctly includes these 3 files into result tar. Final reproduce file works fine
under the same ubuntu, but if I want to run this reproduce under windows or on system that does not have these
files at absolute path, e.g. there is no file /lib64/libc.so.6, then LLD will fail.

It happens because even running the reproduce, such files from GROUP options are taken
from outside, what breaks concept of "black box".
It is sure undesirable and incorrect and makes sometimes hard to run wild repro files.

I though about we can rewrite such scripts before including them into Tar, but at the same time
it is probably not too nice to modify original files used for reproduce, so I refused from this idea.

In this patch I added new option --reproduce-use which informs linker that it runs under
debug reproduce environment and should never use absolute pathes, but convert them to relative.
That fixes issue I am observing.

No testcase included beause would like to discuss possible solutions at first.

[grimar]

May be it is better to modify scripts before pushing them to Tar ? Adding one more gnu incompatible option is inconvinent
for case when we link reproduce using bfd/gold/LLD together for output comparsion.

[ruiu]

I think I'm not convinced that this would worth to be added. I usually add fakechroot chroot before ld.lld and it'd just work, so unless there's a strong reason to avoid doing that, I wouldn't add this function.

[ruiu]

I might be OK with the feature, but this particular patch needs fixing from my point of view.

• --reproduce-use doesn't sound quite right as an option name. I'd vote for --chroot <dirname>.
• Handling this case in ScriptParser::addFile is not right. You should do this in Driver::addFile.
• If you do this, you probably should remove rewritePath as it is no longer needed.

[ruiu]

I did an experiment based on that idea, and it seems to be working, so I sent it to you as https://reviews.llvm.org/D35517.

[grimar]

D35517 was landed instead.

ELF/Config.h

struct Configuration {
   // Each ABI defines its relocation type. IsRela is true if target
   // uses RELA. As far as we know, all 64-bit ABIs are using RELA. A
   // few 32-bit ABIs are using RELA too.
   bool IsRela;
 
   // True if we are creating position-independent code.
   bool Pic;
 
+  // True if we are running invocation from reproduce file for debug.
+  bool UsingReproduce;
+
   // 4 for ELF32, 8 for ELF64.
   int Wordsize;
 };
 
 // The only instance of Configuration struct.
 extern Configuration *Config;
 
 } // namespace elf
 } // namespace lld
 
 #endif

ELF/Driver.cpp

if (ErrOrWriter) {
       Tar->append("version.txt", getLLDVersion() + "\n");
       make<std::unique_ptr<TarWriter>>(std::move(*ErrOrWriter));
     } else {
       error(Twine("--reproduce: failed to open ") + Path + ": " +
             toString(ErrOrWriter.takeError()));
     }
   }
 
+  Config->UsingReproduce = Args.hasArg(OPT_reproduce_use);
+
   readConfigs(Args);
   initLLVM(Args);
   createFiles(Args);
   inferMachineType();
   setConfigs();
   checkOptions(Args);
   if (ErrorCount)
     return;

ELF/DriverUtils.cpp

 }
 
 // Reconstructs command line arguments so that so that you can re-run
 // the same command with the same inputs. This is for --reproduce.
 std::string elf::createResponseFile(const opt::InputArgList &Args) {
   SmallString<0> Data;
   raw_svector_ostream OS(Data);
 
+  // Place marker argument informing it is debug invocation.
+  OS << "--reproduce-use\n";
+
   // Copy the command line to the output while rewriting paths.
   for (auto *Arg : Args) {
     switch (Arg->getOption().getID()) {
     case OPT_reproduce:
       break;
     case OPT_INPUT:
       OS << quote(rewritePath(Arg->getValue())) << "\n";
       break;

ELF/Options.td

def print_gc_sections: F<"print-gc-sections">,
   HelpText<"List removed unused sections">;
 
 def print_map: F<"print-map">,
   HelpText<"Print a link map to the standard output">;
 
 def reproduce: S<"reproduce">,
   HelpText<"Dump linker invocation and input files for debugging">;
 
+def reproduce_use: F<"reproduce-use">,
+  HelpText<"Informs linker about its running debug reproduce invocation">;
+
 def rpath: S<"rpath">, HelpText<"Add a DT_RUNPATH to the output">;
 
 def relocatable: F<"relocatable">, HelpText<"Create relocatable object file">;
 
 def retain_symbols_file: J<"retain-symbols-file=">, MetaVarName<"<file>">,
   HelpText<"Retain only the symbols listed in the file">;
 
 def script: S<"script">, HelpText<"Read linker script">;

ELF/ScriptParser.cpp

if (IsUnderSysroot && S.startswith("/")) {
     SmallString<128> PathData;
     StringRef Path = (Config->Sysroot + S).toStringRef(PathData);
     if (sys::fs::exists(Path)) {
       Driver->addFile(Saver.save(Path), /*WithLOption=*/false);
       return;
     }
   }
 
+  if (Config->UsingReproduce &&
+      sys::path::is_absolute(S, sys::path::Style::posix))
+    S = sys::path::relative_path(S, sys::path::Style::posix);
+
   if (sys::path::is_absolute(S)) {
+    assert(!Config->UsingReproduce);
     Driver->addFile(S, /*WithLOption=*/false);
   } else if (S.startswith("=")) {
     if (Config->Sysroot.empty())
       Driver->addFile(S.substr(1), /*WithLOption=*/false);
     else
       Driver->addFile(Saver.save(Config->Sysroot + "/" + S.substr(1)),
                       /*WithLOption=*/false);
   } else if (S.startswith("-l")) {