From dbbdf057406111be1a699adc134a834f455ca18d Mon Sep 17 00:00:00 2001
From: Zequan Wu <zequanwu@google.com>
Date: Wed, 2 Jun 2021 13:41:31 -0700
Subject: [PATCH] Load .symtab and .strtab sections before .debug_info to
 deduplicate PUBLIC records.

For chrome, the symbol file size dropped from 661 MB to 577 MB.
The number of PUBLIC records dropped from 559416 to 91930.

Change-Id: I7f703d619f2acc7c83f002e6f588f6a6569e5c87
Reviewed-on: https://chromium-review.googlesource.com/c/breakpad/breakpad/+/2916731
Reviewed-by: Lei Zhang <thestig@chromium.org>
---
 src/common/linux/dump_symbols.cc | 54 +++++++++++++++++---------------
 1 file changed, 28 insertions(+), 26 deletions(-)

diff --git a/src/common/linux/dump_symbols.cc b/src/common/linux/dump_symbols.cc
index c939c669..e8d0bd80 100644
--- a/src/common/linux/dump_symbols.cc
+++ b/src/common/linux/dump_symbols.cc
@@ -702,32 +702,6 @@ bool LoadSymbols(const string& obj_file,
     }
 #endif  // NO_STABS_SUPPORT
 
-    // Look for DWARF debugging information, and load it if present.
-    const Shdr* dwarf_section =
-      FindElfSectionByName<ElfClass>(".debug_info", SHT_PROGBITS,
-                                     sections, names, names_end,
-                                     elf_header->e_shnum);
-
-    // .debug_info section type is SHT_PROGBITS for mips on pnacl toolchains,
-    // but MIPS_DWARF for regular gnu toolchains, so both need to be checked
-    if (elf_header->e_machine == EM_MIPS && !dwarf_section) {
-      dwarf_section =
-        FindElfSectionByName<ElfClass>(".debug_info", SHT_MIPS_DWARF,
-                                       sections, names, names_end,
-                                       elf_header->e_shnum);
-    }
-
-    if (dwarf_section) {
-      found_debug_info_section = true;
-      found_usable_info = true;
-      info->LoadedSection(".debug_info");
-      if (!LoadDwarf<ElfClass>(obj_file, elf_header, big_endian,
-                               options.handle_inter_cu_refs, module)) {
-        fprintf(stderr, "%s: \".debug_info\" section found, but failed to load "
-                "DWARF debugging information\n", obj_file.c_str());
-      }
-    }
-
     // See if there are export symbols available.
     const Shdr* symtab_section =
         FindElfSectionByName<ElfClass>(".symtab", SHT_SYMTAB,
@@ -785,6 +759,34 @@ bool LoadSymbols(const string& obj_file,
         found_usable_info = found_usable_info || result;
       }
     }
+
+    // Only Load .debug_info after loading symbol table to avoid duplicate
+    // PUBLIC records.
+    // Look for DWARF debugging information, and load it if present.
+    const Shdr* dwarf_section =
+      FindElfSectionByName<ElfClass>(".debug_info", SHT_PROGBITS,
+                                     sections, names, names_end,
+                                     elf_header->e_shnum);
+
+    // .debug_info section type is SHT_PROGBITS for mips on pnacl toolchains,
+    // but MIPS_DWARF for regular gnu toolchains, so both need to be checked
+    if (elf_header->e_machine == EM_MIPS && !dwarf_section) {
+      dwarf_section =
+        FindElfSectionByName<ElfClass>(".debug_info", SHT_MIPS_DWARF,
+                                       sections, names, names_end,
+                                       elf_header->e_shnum);
+    }
+
+    if (dwarf_section) {
+      found_debug_info_section = true;
+      found_usable_info = true;
+      info->LoadedSection(".debug_info");
+      if (!LoadDwarf<ElfClass>(obj_file, elf_header, big_endian,
+                               options.handle_inter_cu_refs, module)) {
+        fprintf(stderr, "%s: \".debug_info\" section found, but failed to load "
+                "DWARF debugging information\n", obj_file.c_str());
+      }
+    }
   }
 
   if (options.symbol_data != NO_CFI) {