Allocator cleanup: move tcmalloc-specific calls to base

third_party/tcmalloc/chromium/src/gperftools/tcmalloc.h

 /* Copyright (c) 2003, Google Inc.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are
  * met:
  *
  *     * Redistributions of source code must retain the above copyright
  * notice, this list of conditions and the following disclaimer.
  *     * Redistributions in binary form must reproduce the above
@@ skipping 75 matching lines @@
 
 extern "C" {
 #endif
   // Returns a human-readable version string.  If major, minor,
   // and/or patch are not NULL, they are set to the major version,
   // minor version, and patch-code (a string, usually "").
   PERFTOOLS_DLL_DECL const char* tc_version(int* major, int* minor,
                                             const char** patch) __THROW;
 
   PERFTOOLS_DLL_DECL void* tc_malloc(size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_malloc_skip_new_handler(size_t size);

[Will Harris, 2016/01/21 17:56:34]

now this is a first class exported API from tcmalloc, called from base I feel it
should be exposed from a different place. It seems weird to expose this via the
gperftools api. if base always will depend on allocator then none of the weak
symbols will be ever needed, and instead we should have a clean API from base
into the allocator, no?

[Primiano Tucci (use gerrit), 2016/01/21 18:01:09]

Right. So let me see if I'm getting it right. Are you suggesting to not expose
an API but just fwd declare from base and use it there?

My long-term plan is the following:
once we have a unified shim layer API [1], the "malloc behaves like new" should
be done at the shim layer level, not at the allocator level. In that case, we'd
have an unchecked_malloc directly in the shim layer API, instead of punching
holes (and forking) tcmalloc.
But I'd like to first cleanup the current situation and think to the shim API
later.

Maybe an internal fwd declaration in base is the best interim solution?


[1]
https://docs.google.com/document/d/1yKlO1AO4XjpDad9rjcBOI15EKdAGsuGO_IeZy0g0k...

   PERFTOOLS_DLL_DECL void tc_free(void* ptr) __THROW;
   PERFTOOLS_DLL_DECL void* tc_realloc(void* ptr, size_t size) __THROW;
   PERFTOOLS_DLL_DECL void* tc_calloc(size_t nmemb, size_t size) __THROW;
   PERFTOOLS_DLL_DECL void tc_cfree(void* ptr) __THROW;
 
   PERFTOOLS_DLL_DECL void* tc_memalign(size_t __alignment,
                                        size_t __size) __THROW;
   PERFTOOLS_DLL_DECL int tc_posix_memalign(void** ptr,
                                            size_t align, size_t size) __THROW;
   PERFTOOLS_DLL_DECL void* tc_valloc(size_t __size) __THROW;
@@ skipping 23 matching lines @@
   PERFTOOLS_DLL_DECL void* tc_newarray(size_t size);
   PERFTOOLS_DLL_DECL void* tc_newarray_nothrow(size_t size,
                                                const std::nothrow_t&) __THROW;
   PERFTOOLS_DLL_DECL void tc_deletearray(void* p) __THROW;
   PERFTOOLS_DLL_DECL void tc_deletearray_nothrow(void* p,
                                                  const std::nothrow_t&) __THROW;
 }
 #endif
 
 #endif  // #ifndef TCMALLOC_TCMALLOC_H_