Minutes, February 8, 2000

Issue discussion:

991026.3    Duplicate Dwarf data deletion
    The proposal has evolved since it was originally raised. The current
    idea is to add an extension to TAG_compilation_unit which will refer
    to another TAG_compilation_unit, which contains information not provided
    in the current comp unit. Some concern was expressed that this did not
    address all of the duplicate data issues raised in the original proposal.
    Jason Merrill will rewrite the proposal. He will also provide a description
    of the COMDAT proposal, which can be used to eliminate duplicate data.

991108.2    Virtual functions extensions
    The existing standard does not support using multiple vtables to reference
    multiple base classes. One suggestion was to use TAG_member to identify the
    vtable. Mike Eager suggested that it would be better to create a new tag,
    rather than overload TAG_member (5.5.4 describes TAG_member as describing
    data members). Felix Burton suggested that a generalized call operation in
    the location expression could be used to compute vtable addresses, as well
    as have other uses. He will draft a proposal to add OP_call. The Dwarf 2
    spec should describe the state of the stack when a location expression is
    evaluated. In many cases, the address of the object must be on the stack when
    the location expression is evaluated. Jason will rewrite the proposal with
    only DW_AT_vptr_* specifications.

000131.1    Prolog identification
    A suggestion was made to use the line table to identify the line number where
    the prolog ends. This may be not provide adequate precision, since multiple C
    statements may start on the same line. After discussion, Dave Anderson said
    that he would propose extensions to the line number state machine which would
    identify the end of multiple prologs and the start of multiple epilogs.