Using List and Map of user classes in a Record fails to compile

[arezaii]

Summary of Problem

Using both Map and List in a record, without specifying deallocation strategy keywords owned or shared for user classes, results in a compilation error.

Steps to Reproduce

Source Code:

use List;
use Map;

class A {
  var msg:string;
}

class B {
  var name:string;
}

record R {
  var foo: list(A);
  var bar: map(string, B);
}

Compile command:
chpl exampleCode.chpl

Output

exampleCode.chpl:13: In method 'foo':
exampleCode.chpl:13: internal error: invalid attempt to get reference type [AST/primitive.cpp:235]

Note
Curiously, the code above compiles if either map or list is omitted, but not when both are present in the record. I would expect to see similar behavior to the above in either of these cases.

so changing the example code to either of the below examples will result in successful compilation:

record R {
  var foo: A;
  var bar: map(string, B);
}

OR

record R {
  var foo: list(A);
  var bar: B;
}

Alternatively, specifying a deallocation strategy for user classes A and B will successfully compile, e.g:

record R {
 var foo: list(owned A);
 var bar: map(string, shared B);
}

Configuration Information

• Output of chpl --version: chpl version 1.25.0 pre-release (75a0eeea87)
• Output of $CHPL_HOME/util/printchplenv --anonymize:

CHPL_TARGET_PLATFORM: darwin
CHPL_TARGET_COMPILER: clang
CHPL_TARGET_ARCH: x86_64
CHPL_TARGET_CPU: native
CHPL_LOCALE_MODEL: flat
CHPL_COMM: none
CHPL_TASKS: qthreads
CHPL_LAUNCHER: none
CHPL_TIMERS: generic
CHPL_UNWIND: none
CHPL_MEM: jemalloc
CHPL_ATOMICS: cstdlib
CHPL_GMP: bundled
CHPL_HWLOC: bundled
CHPL_RE2: bundled
CHPL_LLVM: none *
CHPL_AUX_FILESYS: none

• Back-end compiler and version, e.g. gcc --version or clang --version:

Apple clang version 12.0.5 (clang-1205.0.22.9)
Target: x86_64-apple-darwin20.5.0
Thread model: posix
InstalledDir: /Library/Developer/CommandLineTools/usr/bin

[mppf]

I think this is a case where the program shouldn't compile but we need a better error message.

With this version, I see the same (unhelpful) error message with both foo and bar fields or either one commented out.

use List;
use Map;

class A {
  var msg:string;
}

class B {
  var name:string;
}

record R {
  //var foo: list(A);
  var bar: map(string, B);
}

var x = new R();
writeln(x);

The problem here is that A and B refer to the classes with generic management. As a result, list(A) and map(string, B) are generic types. Which is probably not what you intended.

It might be possible or reasonable to make list/map issue an error if the element type is generic. However this might go against some partial instantiation scenarios (maybe I want to say var x: list(SomeGenericType) = constructTheList; as a way to be clearer about the type without specifying everything, say).

[arezaii]

today this provides a much nicer, helpful error message that helped me change the code to compile, so I will close this as it seems addressed.

[bradcray]

@arezaii : Would you consider filing a test locking this behavior in (if there isn't obviously one already)?

[arezaii]

I found the test partial-no-q-warning covers this type of problem (partial instantiation of a generic without a ?) and demonstrates the warning/error that I get from this example now, but for completeness I added this example in #25493