[llvm-readobj]Fix error messages for bad archive members and add testing for archive handling

[jhenderson]

Summary

llvm-readobj's error messages were broken for bad archive members. This patch fixes them, and also adds testing for archive and thin archive handling within llvm-readobj.

[Higuoxing]

LGTM

[rupprecht]

test/tools/llvm-readobj/thin-archive.test

16 ↗

(On Diff #188390)

Probably not in this patch, but I'd like to see the full filename verified for a thin archive in a nested dir (which I found to be full of errors when dealing with llvm-ar), e.g. something like:

$ mkdir -p %t/x/y/z
$ cd %t # So we can invoke ar/readobj with relative paths
$ cp 1.o 2.o 3.o x/y/z
$ rm -f x/%t.a
$ llvm-ar rcT x/%t.a x/y/z/*.o
$ llvm-readobj --file-headers x/%t.a | FileCheck %s

# Full filename should be printed; possibly(?) replacing / with \ on Windows
# (IIRC it's stored as / in the raw archive contents, although readobj may translate that on the fly to \)
CHECK: File: y/z/1.o
...

[jhenderson]

test/tools/llvm-readobj/thin-archive.test

16 ↗

(On Diff #188390)

Sounds reasonable. I'll take a look. If it's broken, I'll probably just file a bug, unless it's an obvious one-line fix (currently I'm focusing on testing).

[grimar]

LGTM.

llvm/trunk/test/tools/llvm-readobj/archive.test

+# Show that dumping operations apply to all members in an archive.
+
+# RUN: rm -f %t.a
+# RUN: llvm-ar rc %t.a %p/Inputs/trivial.obj.elf-x86-64 %p/Inputs/trivial.obj.elf-i386 %p/Inputs/trivial.obj.coff-arm
+# RUN: llvm-readobj --all %t.a | FileCheck %s --check-prefixes=LLVM,COFF
+# RUN: llvm-readelf --all %t.a | FileCheck %s --check-prefixes=GNU,COFF
+
+# LLVM: File: trivial.obj.elf-x86-64
+# LLVM: Format: ELF64-x86-64
+# LLVM: ElfHeader {
+# LLVM:   Machine: EM_X86_64
+# LLVM: Sections [
+# LLVM: Relocations [
+# LLVM: Symbols [
+
+# GNU: ELF Header:
+# GNU:   Machine: Advanced Micro Devices X86-64
+# GNU: Section Headers:
+# GNU: Relocation section '.rela.text'
+# GNU: Symbol table '.symtab'
+
+# LLVM: File: trivial.obj.elf-i386
+# LLVM: Format: ELF32-i386
+# LLVM: ElfHeader {
+# LLVM:   Machine: EM_386
+# LLVM: Sections [
+# LLVM: Relocations [
+# LLVM: Symbols [
+
+# GNU: ELF Header:
+# GNU:   Machine: Intel 80386
+# GNU: Section Headers:
+# GNU: Relocation section '.rel.text'
+# GNU: Symbol table '.symtab'
+
+# LLVM: File: trivial.obj.coff-arm
+# LLVM: Format: COFF-ARM
+# COFF: ImageFileHeader {
+# COFF:   Machine: IMAGE_FILE_MACHINE_ARMNT
+# COFF: Sections [
+# COFF: Relocations [
+# COFF: Symbols [

llvm/trunk/test/tools/llvm-readobj/thin-archive.test

+# RUN: rm -f %t.a
+# RUN: rm -rf %t
+# RUN: mkdir -p %t
+
+# Make copies of the test inputs for placing in the archive so that we can mess
+# about with them later on.
+# RUN: cp %p/Inputs/trivial.obj.elf-x86-64 %t/1.o
+# RUN: cp %p/Inputs/relocs.obj.elf-x86_64 %t/2.o
+# RUN: cp %p/Inputs/trivial.obj.coff-arm %t/3.o
+# RUN: llvm-ar rcT %t.a %t/1.o %t/2.o %t/3.o
+
+# Test that basic dumping works for all members.
+# RUN: llvm-readobj --all %t.a | FileCheck %s --check-prefixes=LLVM,COFF
+# RUN: llvm-readelf --all %t.a | FileCheck %s --check-prefixes=GNU,COFF
+
+# LLVM: File: {{.*}}1.o
+# LLVM: Format: ELF64-x86-64
+# LLVM: ElfHeader {
+# LLVM:   SectionHeaderCount: 10
+# LLVM: Sections [
+# LLVM: Relocations [
+# LLVM: Symbols [
+
+# GNU: ELF Header:
+# GNU:   Number of section headers: 10
+# GNU: Section Headers:
+# GNU: Relocation section '.rela.text'
+# GNU: Symbol table '.symtab'
+
+# LLVM: File: {{.*}}2.o
+# LLVM: Format: ELF64-x86-64
+# LLVM: ElfHeader {
+# LLVM:   SectionHeaderCount: 8
+# LLVM: Sections [
+# LLVM: Relocations [
+# LLVM: Symbols [
+
+# GNU: ELF Header:
+# GNU:   Number of section headers: 8
+# GNU: Section Headers:
+# GNU: Relocation section '.rela.text'
+# GNU: Symbol table '.symtab'
+
+# LLVM: File: {{.*}}3.o
+# LLVM: Format: COFF-ARM
+# COFF: ImageFileHeader {
+# COFF: Sections [
+# COFF: Relocations [
+# COFF: Symbols [
+
+# Overwrite one of the members with a member of a different size to show that
+# the size field in the member header is not relevant.
+# RUN: cp %t/1.o %t/2.o
+# RUN: llvm-readobj --file-headers %t.a | FileCheck %s --check-prefix=RESIZED
+
+# RESIZED: File: {{.*}}1.o
+# RESIZED: SectionHeaderCount: 10
+# RESIZED: File: {{.*}}2.o
+# RESIZED: SectionHeaderCount: 10
+
+# Remove the second member and show that the first can still be dumped, but that the last isn't.
+# RUN: rm %t/2.o
+# RUN: not llvm-readobj --file-headers %t.a 2> %t.err | FileCheck %s --check-prefix=MISSING
+# RUN: FileCheck %s --check-prefix=ERR --input-file=%t.err
+
+# MISSING:     File: {{.*}}1.o
+# MISSING:     SectionHeaderCount: 10
+# MISSING-NOT: File: {{.*}}3.o
+# ERR: Error reading file: {{.*}}.a: '{{.*}}2.o': {{[Nn]}}o such file or directory

llvm/trunk/tools/llvm-readobj/llvm-readobj.cpp

 
 /// Dumps each object file in \a Arc;
 static void dumpArchive(const Archive *Arc, ScopedPrinter &Writer) {
   Error Err = Error::success();
   for (auto &Child : Arc->children(Err)) {
     Expected<std::unique_ptr<Binary>> ChildOrErr = Child.getAsBinary();
     if (!ChildOrErr) {
       if (auto E = isNotObjectErrorInvalidFileType(ChildOrErr.takeError())) {
-        reportError(Arc->getFileName(), ChildOrErr.takeError());
+        reportError(Arc->getFileName(), std::move(E));
       }
       continue;
     }
     if (ObjectFile *Obj = dyn_cast<ObjectFile>(&*ChildOrErr.get()))
       dumpObject(Obj, Writer);
     else if (COFFImportFile *Imp = dyn_cast<COFFImportFile>(&*ChildOrErr.get()))
       dumpCOFFImportFile(Imp, Writer);
     else