Re: [gold] Merging string literals with bigger alignment

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Wed, May 1, 2013 at 11:28 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Wed, May 1, 2013 at 11:24 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
>> On Wed, May 1, 2013 at 10:43 AM, Cary Coutant <ccoutant@google.com> wrote:
>>>> `echo g++ -W -Wall    -Werror -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -fmerge-constants -g -O2   -o basic_static_test | sed -e 's/-Wp,-D_FORTIFY_SOURCE=[0-9][0-9]*//'` -Bgcctestdir/ -static basic_test.o
>>>> collect2: ld terminated with signal 6 [Aborted]
>>>> terminate called after throwing an instance of 'std::bad_alloc'
>>>>   what():  std::bad_alloc
>>>> make[5]: *** [basic_static_test] Error 1
>>>
>>> I'm not seeing this. What target?
>>>
>>>> Fixing new_key_offset() is easy, but the gold bootstrap test fails
>>>>
>>>> (cd gcctestdir2 && ln -s ../ld1 ld)
>>>> g++ -m64 -W -Wall    -Werror -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -frandom-seed=ld2 -g -O2 -Bgcctestdir2/ -Wl,--build-id=tree -Wl,--build-id-chunk-size-for-treehash=12345 -Wl,--build-id-min-file-size-for-treehash=0   -o ld2 main.o powerpc.o libgold.a ../libiberty/libiberty.a    -ldl -lz
>>>> gcctestdir2/ld: error: cannot find main.o
>>>> gcctestdir2/ld: error: cannot find powerpc.o
>>>> gcctestdir2/ld: error: cannot find libgold.a
>>>> gcctestdir2/ld: error: cannot find ../libiberty/libiberty.a
>>>> .zdebug_line: error: undefined reference to 'main'
>>>> collect2: error: ld returned 1 exit status
>>>
>>> I don't see this one either.
>>>
>>> -cary
>>
>> I can reproduce it on Fedora 18 with GCC 4.7.2.  ld1 is mis-linked by ld-new.
>>
>> --
>> H.J.
>
> String merging is broken:
>
> Breakpoint 1, gold::Input_file::try_extra_search_path (
>     pindex=pindex@entry=0x799530,
>     input_argument=input_argument@entry=0x757c98, filename="main.o",
>     found_name=found_name@entry=0x879578, namep=namep@entry=0x7fffffff7110)
>     at /export/gnu/import/git/binutils/gold/fileread.cc:975
> 975      if (*pindex > 0 || ::stat(name.c_str(), &dummy_stat) < 0)
> (gdb) p name
> $1 = "Group end without group start/main.o"
> (gdb)
>
> --
> H.J.

This change:

   while (p < pend0)
     {
       size_t len = string_length(p);

-      Stringpool::Key key;
-      this->stringpool_.add_with_length(p, len, true, &key);
+      if (len != 0)
+       {
+         // Within merge input section each string must be aligned.
+         if ((reinterpret_cast<uintptr_t>(p) & (this->addralign() - 1))
+             != init_align_modulo)
+           has_misaligned_strings = true;

-      merged_strings.push_back(Merged_string(i, key));
+         Stringpool::Key key;
+         this->stringpool_.add_with_length(p, len, true, &key);

+         merged_strings.push_back(Merged_string(i, key));
+       }
       p += len + 1;
       i += (len + 1) * sizeof(Char_type);
     }

skips merged_strings.push_back for empty strings so that
empty string merge is mishandled.

This patch restores merged_strings.push_back for emty
strings and fixes the regression.

--
H.J.
---
diff --git a/gold/merge.cc b/gold/merge.cc
index 6480bd9..2d01462 100644
--- a/gold/merge.cc
+++ b/gold/merge.cc
@@ -565,18 +565,16 @@
Output_merge_string<Char_type>::do_add_input_section(Relobj* object,
     {
       size_t len = string_length(p);

-      if (len != 0)
-    {
-      // Within merge input section each string must be aligned.
-      if ((reinterpret_cast<uintptr_t>(p) & (this->addralign() - 1))
-          != init_align_modulo)
-        has_misaligned_strings = true;
+      // Within merge input section each string must be aligned.
+      if (len != 0
+      && ((reinterpret_cast<uintptr_t>(p) & (this->addralign() - 1))
+          != init_align_modulo))
+      has_misaligned_strings = true;

-      Stringpool::Key key;
-      this->stringpool_.add_with_length(p, len, true, &key);
+      Stringpool::Key key;
+      this->stringpool_.add_with_length(p, len, true, &key);

-      merged_strings.push_back(Merged_string(i, key));
-    }
+      merged_strings.push_back(Merged_string(i, key));
       p += len + 1;
       i += (len + 1) * sizeof(Char_type);
     }