Kenneth.Fowler@SEI.CMU.EDU (08/15/89)
A Software Engineering Institute Special Report
entitled-
"Ada Compiler Pragmas and Attributes,
An Implementation-Dependent Study"
has recently been completed for the Ada9X Project
Office. This task is one of several being conducted
as "complex study issues" in preparation for revision
to the Ada language. Ada 9X Project team activities
at the SEI are led by John Goodenough. John reports
directly to Chris Anderson, Ada 9X Project Manager
at Eglin AFB, Florida. A recitation of the task
directive and study abstract follows:
Task Description
"Implementation-dependent pragmas and attributes:
A variety of implementation-dependent pragmas and
attributes have been developed by different compiler
vendors. The SEI shall survey all such implementations
dependiecies, documented in Appendix F by each vendor,
to determine if some of the pragmas and attributes
should be added to the Ada language standard in order
to improve uniformity among compilers. The product of
this task is an Implementation-Dependent Pragmas and
Attributes Study Report."
Abstract
The Ada Language Reference Manual describes an
assortment of compiler pragmas and attributes which
a validated compiler must support. A compiler may add
other pragmas and attributes as long as the legality
of the program cannot be influenced. This study examines
most of the current offering of the latter group with
the goal of identifying new candidates for language
definition. Compiler support for language-defined
pragmas is first characterized and we show that the
level of support varies surprisingly between
implementations. Based on trends in the data,
we point to specific language rule inconsistencies
and to a lack of requisite validation testing as the
most probable cause. A change in the LRM is recommended
to remedy the problem along with improved conformance
checking. The growing body of implementation-defined
pragmas is profiled and the largest class (26%) are
shown to substitute for pragma INTERFACE. An increase
in functionality for INTERFACE is suggested which meets
the same user needs and improves overall uniformity
between compilers. A variety of implementation-defined
pragmas are offered to improve program performance.
Some of these pragmas rely on the user to voluntarily
give-up certain forms of Ada expression. When usage
restrictions are abrogated in the presence of these
pragmas undefined behavior may result. As a basic
precaution, compilers should warn the user when risk
conditions arise in this context. In contrast to pragmas,
we found very few implementation-dependent attributes
and none which provided a functionality significantly
greater than that existing in the language. Attributes
thus remain a largely unexplored realm for innovation.