rosen@gypsy.UUCP (07/17/86)
If you are currently in the process of trying to implement an Ada compiler
(god help you!) or are wondering just how good your compiler is, you may be
interested in the following programs.
A co-worker and myself happended to stumble across a bunch of Ada programs
that caused our Ada compilers to do unexpected things. You may be interested
in trying the programs out with your compiler to see what happends. I am
purposely leaving out the compilers that I used (I used more than one) so
that it won't appear as if I am picking on any one compiler.
If you run the tests and send me the results based on the classifications
that I list below, I will post to the net in the near future. These tests
should probably end up in the ACVC test, but that isn't my concern here.
You should tell me the version of Ada compiler that you used.
Here are the classifications:
******************************************************************************
Compilation classes of test programs:
CLASS MEANING
----- --------------------------------------------------------------
A Should definitely compile and did compile.
B Should definitely compile and did not compile.
C Should definitely not compile and did compile.
D Should definitely not compile and did not compile.
E Should defintely not compile and compiler printed
internal diagnostic.
F Should defintely compile and compiler printed
internal diagnostic.
G Compiler dies without giving any useful diagnostic.
------------------------------------------------------------------------------
******************************************************************************
Here is what my compilers did:
TEST # CLASS compiler #1 compiler #2
------ -----------------------------------------
01 (rename1.a) E D
02 (rename2.a) A A
03 (separate1.a) B B
04 (separate2.a) B B
05 (type1.a) C D
06 (type2.a) E D
07 (use.a) B B
******************************************************************************
What is real surprising is that these two validated compilers both failed to
compile the same program in a couple of cases where it should have (case 03,
04, and 07 for instance). In the code I tried to give an explaination of
why I think the code should compile. If you tend to think that I am wrong
about any of my assertions (I claim the program should compile, but you
disagree), please let me know what your point of view is. I tried to check
these pretty carefully and I think I am right in all cases, but Ada is so
complicated I admit I could easily have overlooked something.
Here are the programs in Unix shar format. If you have VMS and want me to
send you the programs in non-shar format send me mail.
Actually I have come to the conclusion that current Ada compilers have
reached their theoretically limit of purity and it is not possible to ever
make a compiler any more correct :-) .
-----CUT HERE------
#! /bin/sh
# This is a shell archive, meaning:
# 1. Remove everything above the #! /bin/sh line.
# 2. Save the resulting text in a file.
# 3. Execute the file with /bin/sh (not csh) to create the files:
# rename1.a
# rename2.a
# separate1.a
# separate2.a
# type1.a
# type2.a
# use.a
# This archive created: Wed Jul 16 18:40:55 1986
export PATH; PATH=/bin:$PATH
if test -f 'rename1.a'
then
echo shar: will not over-write existing file "'rename1.a'"
else
echo Extracting rename1.a
cat << \SHAR_EOF > 'rename1.a'
--------
-- 01 --
--------
--
-- Should not work. Below program is not legal. Cannot reference an
-- item during its own declaration.
--
package SOME_PACKAGE is
procedure PROC renames SOME_PACKAGE.PROC;
end SOME_PACKAGE;
SHAR_EOF
fi # end of overwriting check
if test -f 'rename2.a'
then
echo shar: will not over-write existing file "'rename2.a'"
else
echo Extracting rename2.a
cat << \SHAR_EOF > 'rename2.a'
--------
-- 02 --
--------
--
-- This should work. It's legal Ada.
--
package A is
package B renames A;
package C renames B;
package D renames C;
package E renames D;
end A;
with A;
package X is
use A.B.C.D.E.B.C.D.E.B.C.D.E.B.C.D.E.B.C.D.E;
Z: INTEGER;
end X;
SHAR_EOF
fi # end of overwriting check
if test -f 'separate1.a'
then
echo shar: will not over-write existing file "'separate1.a'"
else
echo Extracting separate1.a
cat << \SHAR_EOF > 'separate1.a'
--------
-- 03 --
--------
--
-- Should work according to semantics of rename (LRM 8.5(3)), but doesn't
--
package A is
package B renames A;
procedure C;
end A;
package body A is
procedure C is separate;
end A;
separate (A.B)
procedure C is
begin
null;
end C;
SHAR_EOF
fi # end of overwriting check
if test -f 'separate2.a'
then
echo shar: will not over-write existing file "'separate2.a'"
else
echo Extracting separate2.a
cat << \SHAR_EOF > 'separate2.a'
--------
-- 04 --
--------
--
-- Should work according to semantics of rename (LRM 8.5(3)), but doesn't
--
package A is
procedure B;
procedure C renames B;
end A;
package body A is
procedure B is separate;
end A;
separate (A)
procedure B is
procedure X is separate;
begin
null;
end B;
separate (A.C)
procedure X is
begin
null;
end X;
SHAR_EOF
fi # end of overwriting check
if test -f 'type1.a'
then
echo shar: will not over-write existing file "'type1.a'"
else
echo Extracting type1.a
cat << \SHAR_EOF > 'type1.a'
--------
-- 05 --
--------
--
-- Should not work, there is a mistaken identifier in case statement
--
procedure PROC is
type FRUIT is (APPLE, PEAR, PEACH, PLUM, GRAPEFRUIT, ORANGE);
X: FRUIT := GRAPEFRUIT;
begin
case FRUIT is -- Uh-oh, wrong identifier here, meant X
when OTHERS =>
X := FRUIT'FIRST;
end case;
end PROC;
SHAR_EOF
fi # end of overwriting check
if test -f 'type2.a'
then
echo shar: will not over-write existing file "'type2.a'"
else
echo Extracting type2.a
cat << \SHAR_EOF > 'type2.a'
--------
-- 06 --
--------
--
-- Should not work, there is a mistaken identifier in case statement
--
procedure PROC is
type FRUIT is (APPLE, PEAR, PEACH, PLUM, GRAPEFRUIT, ORANGE);
X: FRUIT := GRAPEFRUIT;
begin
case FRUIT is -- Uh-oh, wrong identifier here, meant X
when APPLE =>
null;
when OTHERS =>
X := FRUIT'FIRST;
end case;
end PROC;
SHAR_EOF
fi # end of overwriting check
if test -f 'use.a'
then
echo shar: will not over-write existing file "'use.a'"
else
echo Extracting use.a
cat << \SHAR_EOF > 'use.a'
--------
-- 07 --
--------
--
-- Should work. Compiler complains saying a visible declaration for
-- FRUIT is hidden by the use of the 'use' clause in the context clause
-- for the body of package B.
--
-- The 'use' clause in the body of package B should have no effect since
-- there is already a 'use' for package A in the spec of package B that will
-- allow declarations in A to be visible in the body of B.
--
-- Proof:
--
-- 8.1(2) - Declaritive region of package spec includes its own body
-- 8.4(3) - Use of 'use' clause covers entire declarative region
-- 8.4(9) - Repeated 'use' clauses have no effect.
--
-- Therefore, there is no visible declaration to be hidden
--
package A is
type FRUIT is (apple, banana, cherry);
end A;
with A;
package B is
use A;
SOME_FRUIT: FRUIT;
end B;
with A;
use A; -- Take out this 'use' clause to make units compile
-- Whether it is there or not should have no effect
package body B is
OTHER_FRUIT: FRUIT;
end B;
SHAR_EOF
fi # end of overwriting check
# End of shell archive
exit 0
******************************************************************************
----------------
| Steve Rosen
| Siemens Research and Technology Laboratories
| Princeton, NJ
USENET: adrvax\
ihnp4 |
princeton |-->!siemens!gypsy!rosen
topaz |
vrdxhq/
ARPA: siemens!gypsy!rosen@topaz.rutgers.edusdl@linus.UUCP (07/21/86)
I tried your seven compiler breakers on the DEC VAX Ada compiler,
T1.2-12. (VAX/VMS).
Two results don't fit your categories. For (illegal) programs 05 and
06, the compiler reported the correct diagnostic, but then hung (or
infinite looped). I have taken the liberty of calling this new
category of behavior, category H (program shouldn't compile, and
compiler complains, but hangs anyway).
Also, note that the DEC VAX Ada also believes that 03 and 04 are
illegal. That's three compilers so far.
So, here are the results:
01 D (program illegal, compiler complains)
02 A (program legal, compiler compiles it)
03 B (program legal, compiler complains)
04 B (program legal, compiler complains)
05 H (program illegal, compiler complains but hangs)
06 H (program illegal, compiler complains but hangs)
07 A (program legal, compiler compiles it)
Steven Litvintchouk
MITRE Corporation
Burlington Road
Bedford, MA 01730
Fone: (617)271-7753
ARPA: sdl@mitre-bedford
UUCP: ...{allegra,decvax,genrad,ihnp4,philabs,security,utzoo}!linus!sdl