Minutes, Oct 31, 2000

David Anderson acting for Michael Eager.
David Anderson, George Beshers, Ron Brender, Felix Burton,
Jim Dehnert, and David Weatherford attended (physically or
by phone).

001018.1 Unspecified Type (void, void * and similar for other languages)

    The proposal (which was discussed in the previous minutes)
    was adopted. Ron Brender plans to put this as a section
    between 5.1 and 5.2, causing renumbering of the rest of
    chapter 5 (this editorial aspect is not part of the
    proposal, it is simply the Editor's current plan for this topic).


001012.1 Factored Offset

    Felix Burton asked "Why not use dwarf expressions
    instead of new operators?" The suspicion is
    that there is a meaningful difference in the space
    required using expressions. Tabled till the next meeting
    so the proposal champion (Anderson) can address this question
    and get some real numbers together.


001016.1 Interludes/Trampolines

    Weatherford pointed out yet a new name for these. "slipperies"
    is a term often used at Sun (to add to the "thunks, trampolines,
    and interludes" list).

    The possibility of using DW_AT_artificial was raised as an
    alternative to a new attribute.
    But this was not felt to be direct or specific enough.
    When used to describe adjusting a 'this' pointer
    to point to a derived-type on entry, it was pointed out
    that there could be 'adjust it back' code at the exit point,
    which also fits the description.

    The proposal was accepted with the stipulation that
    the new attribute be named DW_AT_trampoline.
    Additionally, it was stipulated that the FORM of the
    DIE can be any of at least the following:
    A string, an actual function name
    This function name lookup is implementation-dependent
    and could reference an ABI/implementation/object
    specific table, such as an Elf symbol table.
    An Address of a function that will be called.
    A reference to a DIE, the function definition DIE
    of the function that is being called.
    A flag (when there is no way to know the
    function address or name).
    When it's just a flag the debugger must step (or equivalent)
    to get to the target function.