This is the mail archive of the gcc-bugs@gcc.gnu.org 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]

Re: debug/7032: arm-elf-gcc 3.1 generates wrong DWARF-2, unlike 3.0.4


>     This sounds more like a bug in AXD than in gcc.  GCC is designed to
work with GDB as the primary consumer of the debug information.  If you want
this problem to be fixed then you will have to supply details of precisely
what debug information is causing the problem.  Remember that most gcc
developers don't have access to AXD.
>

    You might be right, of course.  However it is hard to tell
because the error message from AXD was very uninformative.
Still, the bug might also be in gcc, especially since it appeared
after a version change.  That is the reason why I reported
the bug, which completely prevents the use of gcc in
such an environment.

    Unfortunately, I did not have time to investigate the exact cause
and I have rolled back to gcc 3.0.4, which works.

    Nonetheless, someone who knows what has changed in the
DWARF generation between 3.0.4 and 3.1 (and also 3.1.1,
which exhibits identical behavior) should be able to discover
the origin.

    AXD crashed regardless of the compiled program, and
the problem was located in the two sections indicated
in my bug report.

    All these facts (changes between 3.0.4 and 3.1 in the two
specified DWARF sections and independence between  the
crash and the code of the compiled program) should narrow
enough the possible choices for the cause.

    When I will find some time, I will further investigate the problem.
I also had problems with binutils, but there I was able to find
quickly the bug and patch the binutils sources.

    I would rather use gdb than AXD, which is full of other bugs,
but there is no way to connect gdb to my boards.  The only
ARM debug interfaces that gdb knows about are those from
Macraigor but their support is incomplete because there is
no way to define a JTAG chain with multiple devices.

    Therefore, until I will have time to reverse-engineer any
of the existing ARM debug interfaces and add the right code
to gdb, I must use AXD.  And even if I do the reverse-engineering,
it is probable that I will not be able to contribute the corresponding
device drivers to the main distribution of gdb because of the
existing abusive laws for copyright protection.

                                                    Best regards !











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