This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

c++/6828: g++3 makes wrong mangled name for template cast operator

>Number:         6828
>Category:       c++
>Synopsis:       g++3 makes wrong mangled name for template cast operator
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Mon May 27 03:16:01 PDT 2002
>Originator:     Grigory Zagorodnev
>Release:        3.1
host: i686-pc-linux-gnu
build: i686-pc-linux-gnu
target: i686-pc-linux-gnu
configured with: ./configure
1. The issue

In the given testcase, for operator "template <class T> C::operator T ()" with 
T = {int}, g++3 compiler produces the following mangled name:

but it's expected to have mangled name "_ZN1CcvT_IiEEv" here.

--- testcase (fail.cpp) -------------------------------------
struct C {
  template <class T>
  operator T ();

template <class T>
C::operator T () { return 0; }

template C::operator int ();

2. Clarify the issue
Unfortunately, the mangling grammer given in the ABI is ambiguous in this area, so we need to think more about it.

First of all, let's decompose the mangled name taking GCC's encoding (it does not matter which one to take).
We are interesting in the following parts of '_ZN1CcviIiEEv' mangled name:
	a) cvi		# <operator-name> ::= cv <type>
	b) IiE		# <template-args> ::= I <type>+ E

In both parts, value of <type> used to be the same. And it is really the same 8-).
But, only one question is there: 
	Which part should be encoded/decoded first? (the hen or the egg)

As I can see, part a) depends on type value of b). 
So the <type> in the a should be a reference to template argument from b.

3. Clarify compiler behaviour
Let's change our example to make the behaviour more clean and see what exactly compilers are doing. 
According to the ABI, built-in types are not substituted, so it's better to change built-in type 'int' to some class 'A'.

Now we that g++3 encodes operator as:
i.e. type 1A in the part (a) of mangled name becomes major one, and part (b) reference it by <substitution>

While the correct code is:
where part (a) depends from type specified in part (b) so it reference it as <template-param>.

4. Decoder
It's true that the 'c++filt' is not able to decode given names. But decoder is not guilty.
The root of the problem is the mangling grammer.

1. compile testcase 
    g++3 -c fail.cpp

2. get list of produced mangled names in the object file
    nm fail.o | grep _Z

3. You will see
**** T _ZN1CcviIiEEv
Which is wrong.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]