[lldb][LoongArch] Preserve temporary symbols starting with .L in lldb symbol table

[zhaoqi5]

LoongArch64 always uses symbols for relocations, so temporary symbols starting with ".L" should be preserved so that relocations in .debug_info can be fixed correctly.

After this commit, three tests passed:

lldb-shell :: SymbolFile/DWARF/anon_class_w_and_wo_export_symbols.ll
lldb-shell :: SymbolFile/DWARF/clang-ast-from-dwarf-unamed-and-anon-structs.cpp
lldb-shell :: SymbolFile/DWARF/clang-gmodules-type-lookup.c

[llvmbot]

@llvm/pr-subscribers-backend-loongarch

@llvm/pr-subscribers-lldb

Author: ZhaoQi (zhaoqi5)

Changes

LoongArch64 always uses symbols for relocations, so temporary symbols starting with ".L" should be preserved so that relocations in .debug_info can be fixed correctly.

After this commit, three tests passed:

lldb-shell :: SymbolFile/DWARF/anon_class_w_and_wo_export_symbols.ll
lldb-shell :: SymbolFile/DWARF/clang-ast-from-dwarf-unamed-and-anon-structs.cpp
lldb-shell :: SymbolFile/DWARF/clang-gmodules-type-lookup.c

---

Full diff: https://github.com/llvm/llvm-project/pull/158551.diff

1 Files Affected:

• (modified) lldb/source/Plugins/ObjectFile/ELF/ObjectFileELF.cpp (+6-2)

diff --git a/lldb/source/Plugins/ObjectFile/ELF/ObjectFileELF.cpp b/lldb/source/Plugins/ObjectFile/ELF/ObjectFileELF.cpp
index 931baf5927a04..0f8bc5fc93e07 100644
--- a/lldb/source/Plugins/ObjectFile/ELF/ObjectFileELF.cpp
+++ b/lldb/source/Plugins/ObjectFile/ELF/ObjectFileELF.cpp
@@ -2119,8 +2119,12 @@ ObjectFileELF::ParseSymbols(Symtab *symtab, user_id_t start_id,
     // generated local labels used for internal purposes (e.g. debugging,
     // optimization) and are not relevant for symbol resolution or external
     // linkage.
-    if (llvm::StringRef(symbol_name).starts_with(".L"))
-      continue;
+    // LoongArch64 always uses symbols for relocations, so temporary symbols
+    // starting with ".L" should be preserved.
+    if (arch.GetMachine() != llvm::Triple::loongarch64) {
+      if (llvm::StringRef(symbol_name).starts_with(".L"))
+        continue;
+    }
     // No need to add non-section symbols that have no names
     if (symbol.getType() != STT_SECTION &&
         (symbol_name == nullptr || symbol_name[0] == '\0'))

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[SixWeining]

cc @barsolo2000

[DavidSpickett]

I assume this is fallout from https://discourse.llvm.org/t/rfc-should-we-omit-local-symbols-in-elf-files-from-the-lldb-symbol-table/87384.

Which I have no context for but I'll add some of those folks on review.

[MaskRay]

I assume this is fallout from discourse.llvm.org/t/rfc-should-we-omit-local-symbols-in-elf-files-from-the-lldb-symbol-table/87384.

Which I have no context for but I'll add some of those folks on review.

I am not familiar with lldb. When RISC-V and LoongArch linker relaxation is enabled for assemblers, there are typically a lot of .L0 symbols (#89693). These numerous symbols have an identical name. What does "preserve" in the context mean?

[zhaoqi5]

I am not familiar with lldb. When RISC-V and LoongArch linker relaxation is enabled for assemblers, there are typically a lot of .L0 symbols (#89693). These numerous symbols have an identical name. What does "preserve" in the context mean?

There are .L0 symbols even linker relaxation is not enabled, because RISC-V and LoongArch always use symbols for relocations.

If these symbols are omitted from symbol table, lldb will fail to analyze the symbols info and fix the relocations in .debug_info, leading to the three tests mentioned earlier which test lldb-test symbols -dump-clang-ast ... failed (maybe both on riscv and loongarch). So we should preserve these symbols rather than omit them.