Re: [PATCH][gold] compiling gold with libc++ and __stl_hash_string

[Dimitry Andric <dimitry at andric dot com>]

On 19 Dec 2013, at 00:25, Cary Coutant <ccoutant@google.com> wrote:
>> That's not the right fix.  That code should only be used if gold is
>> using <ext/hash_map>, which is a fall back for older compilers.
>> Currently gold prefers <tr1/unordered_map>.  I think we need to add a
>> case to configure.ac and system.h to use <unordered_map> if it is
>> available, as I assume it is for libc++.
> 
> Ian, does the attached patch look OK?
> 
> Would we want to turn on C++11 at some point? I believe we get
> <unordered_map> in G++ if we turn on --std=c++11.

We had almost exactly the same patch for FreeBSD 10/11 (which uses libc++ by default), but with two additional changes:

diff --git a/gold/stringpool.cc b/gold/stringpool.cc
index 665fcc8..00ed184 100644
--- a/gold/stringpool.cc
+++ b/gold/stringpool.cc
@@ -73,7 +73,10 @@ Stringpool_template<Stringpool_char>::reserve(unsigned int n)
 {
   this->key_to_offset_.reserve(n);

-#if defined(HAVE_TR1_UNORDERED_MAP)
+#if defined(HAVE_UNORDERED_MAP)
+  this->string_set_.rehash(this->string_set_.size() + n);
+  return;
+#elif defined(HAVE_TR1_UNORDERED_MAP)
   // rehash() implementation is broken in gcc 4.0.3's stl
   //this->string_set_.rehash(this->string_set_.size() + n);
   //return;
@@ -503,7 +506,7 @@ template<typename Stringpool_char>
 void
 Stringpool_template<Stringpool_char>::print_stats(const char* name) const
 {
-#if defined(HAVE_TR1_UNORDERED_MAP) || defined(HAVE_EXT_HASH_MAP)
+#if defined(HAVE_UNORDERED_MAP) || defined(HAVE_TR1_UNORDERED_MAP) || defined(HAVE_EXT_HASH_MAP)
   fprintf(stderr, _("%s: %s entries: %zu; buckets: %zu\n"),
          program_name, name, this->string_set_.size(),
          this->string_set_.bucket_count());

-Dimitry

Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail