[Up] [SIGAda] [ACM]

ASIS for GNAT
Update - June 1995

We would like to bring to the attention of the ASISWG a little information about the the current state of the ASIS for GNAT implementation and our position concerning some open ASIS issues.

    Sergey Rybin                   Alfred Strohmeier
(rybin@alex.srcc.msu.su)       (alfred.strohmeier@di.epfl.ch)
    Eugene Zueff
(zueff@such.srcc.msu.su)        Swiss Federal Institute of
                                        Technology
  Moscow State University

I. ASIS for GNAT Implementation: the current state.

The initial version of the prototype implementation of the ASIS interface for the GNAT compiler has been already implemented. It allows to make up sample ASIS applications which could perform simple structural decomposition of Compilation Unit down to the statements/declarations level (for some kinds of statements and declarations more detailed decomposition is also possible).

The ASIS specification to be implemented in the project has been obtained by applying the ASIS 2.0.A Drafts to the ASIS 1.1.1 specification (so we are trying to implement some pre-prototype of ASIS 2.X for GNAT) and we are looking forward for appearing the first complete Draft ASIS specification for Ada 95.

II. ASIS Open Issues.

2.1. Library Model for ASIS 95.

The starting points for revising the ASIS Library Model for ASIS/Ada 95 should include:

Ada 95 does not even formally define the notion of the Compiler Library and of the Obsolete Compilation Unit;
All what is defined by Ada 95 LRM is related to the source texts of the Ada Compilation Units.
Some Ada 95 compilers could inherit their Library models (together with the notions of the Program Library and Obsolete Compilation Unit, which are defined rather strictly in RM83!) from their Ada 83 "parents", but the other could use "novel" approaches based on the "loose" definition of the compilation process in Ada 95 LRM -- GNAT is an excellent example!
ASIS Library-level and Compilation-Unit-level queries could be subdivided into two parts:
- related to the content (=sources) of the library;
- related to the history of the (creation and modification of the) library;

History of the library:: the principles of handling the history may differ essentially among different compilers (depending on at leas the smart recompilation techniques), GNAT does not handle the history at all;
Content of the library:: the question is who should define the content of the (Ada Program?) Library handled by ASIS implementation and (through the ASIS interface) by Ada tool working as ASIS application?

Our suggestions concerning the new ASIS Library model are:

ASIS Library should have the compiler-independent definition and it should be source-based;
ASIS should work only with the content of the library but not with the history of the library; all history-related queries should be removed from ASIS or reformulated to make them more (or completely, if it could be possible) content-oriented -- it would be a very few queries indeed;
The content of the ASIS Library handled by an ASIS implementation and processed by an Ada tool (ASIS application) should be determined by the ASIS implementation, but not by the compiler. Moreover, ASIS implementation could be even a stand-alone tool (which does not require any compiler at all and which could handle the arbitrary collections of Ada sources);
ASIS Library should contain the Compilable Compilation Units only (see RM95. 10.1.4); the notion of compilable unit may be considered as the "weak" Ada 95 analog of the Ada 83 rules defining the belonging of the Unit to an (Ada 83!) Library;
The content of the ASIS Library should be frozen from the Initialization till the Finalization of the ASIS Environment. That means that ASIS is to process the snapshots of the compilation environment of a compiler or of some set of Ada sources;

We think that the GNAT compilation model could be used as the initial prototype for defining the ASIS 95 Library Model.

2.2 ASIS Semantics for Ada Implicit Constructs.

Not long ago Sergey Rybin asked some questions concerning the reasons for having differences between ASIS and Ada LRM semantics of the Ada implicit constructs. He got the explanations from ASIS "gurus" which could be summarized as:

ASIS treatment of the Ada implicit constructs is much more convenient for the vendor libraries, because implicit elements are more hard and resource-consuming to process and to store by a compiler library.

But what about GNAT? It could represent and process implicit constructs just in the same way as explicit ones.

ASIS treatment of the Ada implicit constructs is much more convenient for ASIS users, because ASIS application usually would like to see namely those constructs which have been "explicitly" typed by programmer.

But the questions are:

Is this the result of the consensus among ASIS users?
Do ASIS users (as well as Ada compiler vendors) really like to use two different formal models for just the same object -- the content of the Ada Compilation Unit (even if these models could be proved as being completely equivalent)?

And one more question concerning the latest plans to standardize ASIS: could it be really possible to get the acceptance of the ASIS as an (ISO) information technology standard having the direct relation to the Ada 95 standard and having also formal contradiction with Ada Standard?

We would suggest to revisit the ASIS semantics of Ada implicit constructs once again. Our proposal is that ASIS should confirm with Ada LRM in this aspect.

III. Announcement of a workshop about ASIS.

There are plans to organize an ASIS Workshop, on Friday October 6, 1995, during the Ada in Europe 95 conference (October 2-6, Frankfurt, Germany). We will be the organizers of the Workshop. We hope to mail out the corresponding announcement soon.

In no case, this workshop will be a "competitor" with the work done by your working group (the ASIS working group). Our goal is to improve awareness of ASIS in Europe. The workshop may well start by some kind of short tutorial for new-comers to ASIS, but more technical issues will be tackled too. You are all welcome to attend!