Minutes, April 17, 2001
010322.1 Permit AT_location in non-defining declaration
This proposal removes the restriction that a non-defining variable
declaration not have an location specification. No one was aware
of any reason for this restriction and at least one person thought
that the existing text was meant to be descriptive (that a non-
defining declaration wouldn't usually have a location) rather than
prescriptive. There was one clarification to the proposal: Change
the third paragraph in item 3 to begin "In a defining declaration".
The proposal was adopted as clarified.
010228.1 AT_specification generalization
This proposal describes the use of AT_specification in circumstances
other than previously anticipated, such as with a TAG_module or
TAG_class_type. Appendix A indicates the expected use of attributes,
but in most cases, Dwarf 2 does not prescribe or restrict the use of
attributes which are not mentioned. This proposal only clarifies the
use of AT_specification. Adopted with the clarification that text
under 2.13.1 should read "may have an DW_AT_specification" instead of
"has an DW_AT_specification".
010301.1 Out of line package initialization
An Ada module may have initialization code. The current specification
anticipates that this is within the module. After discussion, it was
decided that changing the text in Section 3.2.1 which specifies that
a TAG_module which has initialization code has either AT_low_pc/AT_high_pc
or AT_ranges, to indicate that it may have these attributes would have
substantially the same result. This change was adopted.
010301.3 Thin inline functions
This proposal adds three attributes: AT_call_file, AT_call_line, and
AT_call_column. When there is nested inlining (inlined code has an
inlined code section) it may not be clear where the nested inline appears
in the code. These attributes permit this to be described. The
description "thin inline" come from the common case where the outer
function call is a wrapper for the nested function call.
A question was raised about a number of sections of the specification which mention
code addresses or ranges. The intent is that if code is not generated for a
function or block, that these attributes will not be present. This was suggested as
an editorial change to Section 2.16.
We discussed plans for a full document review. We would like to schedule a
part or full day, in person, editorial review for mid-June. It was felt, based
on the experience of the previous editorial review, that this was more effective
than individual review and comments.