Re: ASIS (Was Re: IrisAda) REPOST - REFORMATTED TO 80 COLUMNS

Re: ASIS (Was Re: IrisAda) REPOST - REFORMATTED TO 80 COLUMNS

[Alex Blakemore]

ryer@inmet.camb.inmet.com (Mike Ryer) writes:
> ASIS has a cost beyond the dollars you pay for it.  It
> can result in slower compilers and larger program libraries.
> Our compiler (for MVS) currently generates DIANA; it would be a very simple
> matter to add ASIS generation -- just send money.  

In Intermetric's case, since you already store DIANA, providing an ASIS
interface to the DIANA should not add significant time, space, costs to
your compiler beyond what's there to create and traverse DIANA.  DIANA is
a very rich intermediate form.  You are correct that a vendor which
already used a lean intermediate form may have more work & tradeoffs to
make to support ASIS.

One important point - ASIS is NOT an intermediate form or a data structure
(like DIANA).  It is a relatively simple INTERFACE that allows you to
query information from whatever intermediate form the compiler uses in a
standard way.  Depending upon the details of the particular intermediate
form, supporting ASIS may be simple or complex.  If the information is
already stored in the intermediate form, the compiler should not have to
be made slower and more complex (regardless of the complexity of the ASIS
interface) The compiler may not need to be changed at all.  If ASIS
requires some information that is not currently stored, there are several
approaches.
  - modify the compiler to store the needed info,
  - recompute the info on demand
  - not implement the particular ASIS feature
    (several things have been made optional that compiler vendors
     typically dont store, there are query routines that allow tools to
     test if an implementation does this)

For example, source position info may not be preserved by some compilers.
If its not, dont implement the package ASIS_Text that provides that info.
Tools can call the routine IS_Text_Supported to check whether that info is
available.  Tools that dont need source position info can work.  Other
tools wont & people that want those tools to run can then pressure/pay you
to provide that feature.

I know one very fast lean compiler that already has an intermediate form
that will easily support most of ASIS.  Most compiler vendors already
provide some interface, but its always a different one.  Until we have
some kind of standard, its very expensive and difficult to develop Ada
tools - and when you are done they only work with one compiler.

> In my opinion, ASIS has been defined as "everything that Telesoft and Rationa
l
> happen to have in their libraries", rather than "everything that a reasonable
> tool vendor needs and cannot easily compute for him/herself".

ASIS is evolving - and the latest drafts have tried to address your
concerns in several ways.  Several compiler vendors have attended working
group sessions and pointed out what is hard or easy for them to support.
One concept that is shaping up in 1.1 is to provide a standard interface
to some information that could be computed (with difficulty) from lower
level ASIS primitives.  There would be a public implementation of this
higher level layer that any vendor could use - so they only have to
provide the primitives

Perhaps someone from Intermetrics could attend an ASIS working group
meeting and contribute to developing ASIS into a practical standard.

--
---------------------------------------------------
Alex Blakemore   alex@cs.umd.edu NeXT mail accepted


-- 
---------------------------------------------------
Alex Blakemore alex@cs.umd.edu   NeXT mail accepted