[PATCH] clone_function_name_1: Retain any stdcall suffix

[PATCH] clone_function_name_1: Retain any stdcall suffix

[Ray Donnelly]

[-- Attachment #1: Type: text/plain, Size: 175 bytes --]

I'm not familiar with setting up GCC testcases yet so I expect to have
to do that at least. To aid discussion, the commit message contains a
testcase.

--
Best regards,

Ray.

[-- Attachment #2: 0022-clone_function_name_1-Retain-any-stdcall-suffix.patch --]
[-- Type: application/octet-stream, Size: 3584 bytes --]

From 38274147bb2a3fa481d0a1853c185c2337550260 Mon Sep 17 00:00:00 2001
From: Ray Donnelly <mingw.android@gmail.com>
Date: Mon, 17 Aug 2015 22:57:46 +0100
Subject: [PATCH 22/22] clone_function_name_1: Retain any stdcall suffix

Previously, clone_function_name_1 would add a suffix after
any existing stdcall suffix, for example ipa-split.c would
clone test@4 as test@4.part.0.

Later, i386_pe_strip_name_encoding_full would come along
and strip off everything from the last @ onwards which had
the effect of generating incorrect section names which
would then fall over with errors such as:

error: void test() causes a section type conflict with \
  void test@4.part.0()

The following testcase, reduced from Firefox can be used
to reproduce this.

test.ii:
class ClassA {
public:
  virtual int __attribute__((__stdcall__)) Dispatch() = 0;
};
class ClassB {
public:
  ClassA* __attribute__((__stdcall__)) operator->();
};
class ClassC : ClassA {
  int *some_int_ptr_variable;
  int __attribute__((__stdcall__)) Dispatch() {
    return some_int_ptr_variable
           ? 42
           : m_ClassInstanceB->Dispatch();
  }
  ClassB m_ClassInstanceB;
};
ClassC ClassInstanceC;

Compile for i686-w64-mingw32 with:
cc1plus -O -fpartial-inlining -fdevirtualize \
  -fdevirtualize-speculatively test.ii

Outputs:
test.ii: In member function 'virtual int ClassC::Dispatch()':
test.ii:11:36: error: virtual int ClassC::Dispatch() causes \
  a section type conflict with int ClassC::_ZN6ClassC8DispatchEv@4.part.0()
   int __attribute__((CALLTYPE)) Dispatch() {
                                    ^
test.ii:11:36: note: \
  'int ClassC::_ZN6ClassC8DispatchEv@4.part.0()' was declared here
---
 gcc/cgraphclones.c | 13 ++++++++++++-
 gcc/defaults.h     |  2 +-
 2 files changed, 13 insertions(+), 2 deletions(-)

diff --git a/gcc/cgraphclones.c b/gcc/cgraphclones.c
index 9e9f1a0..69b8a91 100644
--- a/gcc/cgraphclones.c
+++ b/gcc/cgraphclones.c
@@ -519,15 +519,24 @@ cgraph_node::create_clone (tree new_decl, gcov_type gcov_count, int freq,
 static GTY(()) unsigned int clone_fn_id_num;
 
 /* Return a new assembler name for a clone with SUFFIX of a decl named
-   NAME.  */
+   NAME. Final stdcall @N suffixes are maintained. */
 
 tree
 clone_function_name_1 (const char *name, const char *suffix)
 {
   size_t len = strlen (name);
   char *tmp_name, *prefix;
+  char *at_suffix = NULL;
 
   prefix = XALLOCAVEC (char, len + strlen (suffix) + 2);
+  /* name + 1 to skip fastcall which begins with '@' */
+  at_suffix = strchr (name + 1, '@');
+  size_t at_suffix_len = 0;
+  if (at_suffix)
+    {
+      at_suffix_len = strlen (at_suffix);
+      len -= at_suffix_len;
+    }
   memcpy (prefix, name, len);
   strcpy (prefix + len + 1, suffix);
 #ifndef NO_DOT_IN_LABEL
@@ -538,6 +547,8 @@ clone_function_name_1 (const char *name, const char *suffix)
   prefix[len] = '_';
 #endif
   ASM_FORMAT_PRIVATE_NAME (tmp_name, prefix, clone_fn_id_num++);
+  if (at_suffix)
+    strcat (tmp_name, at_suffix);
   return get_identifier (tmp_name);
 }
 
diff --git a/gcc/defaults.h b/gcc/defaults.h
index 9d38ba1..c34478a 100644
--- a/gcc/defaults.h
+++ b/gcc/defaults.h
@@ -51,7 +51,7 @@ see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
 # define ASM_FORMAT_PRIVATE_NAME(OUTPUT, NAME, LABELNO) \
   do { const char *const name_ = (NAME); \
        char *const output_ = (OUTPUT) = \
-	 (char *) alloca (strlen (name_) + 32); \
+	 (char *) alloca (strlen (name_) + 35); \
        sprintf (output_, ASM_PN_FORMAT, name_, (unsigned long)(LABELNO)); \
   } while (0)
 #endif
-- 
2.5.0

Re: [PATCH] clone_function_name_1: Retain any stdcall suffix

[Kai Tietz]

2015-08-18 22:23 GMT+02:00 Ray Donnelly <mingw.android@gmail.com>:
> I'm not familiar with setting up GCC testcases yet so I expect to have
> to do that at least. To aid discussion, the commit message contains a
> testcase.

No problem.  Patch looks fine to me.

Thanks,
Kai

> --
> Best regards,
>
> Ray.