Empty arrays?

[orionll]

Object[] array = {};
browse(new LJV(), array);

Graphviz Online doesn't render and reports an error:

in label of node n1

[inponomarev]

Hi @orionll , thanks for pointing this out!

Another thing I don't like about arrays now is that their type isn't shown.

I see two possibilities for array visualization fix:

1. Use xlabel for array type, use shape="point", or circled {} for an empty array

Pros: arrays are distinguishable from other objects. Cons: xlabel is often poorly placed.

2. Use first <td> for an array type, other <td>s for array elements.

Pros: simplicity. Cons: array can be confused with other objects

What do you think?

[orionll]

I definitely like the second option the best:

• It solves the problem with empty arrays.
• Exact types give more information to viewers (e.g. make int[] vs byte[] arrays distinguishable).
• This is better for pedagogical reasons: an array looks more or less like it's actually stored in JVM (the Foo[] part looks like an array header and then values are following).

I also think we should give users the full flexibility of how objects headers are represented as strings. There can be some method like:

// probably not the best name for the method
public <T> LJV setStringRepresentation(Class<T> cz, Function<T, String> function) {
      ...
}

Having this method, I can draw a graph like I want. For example, this is the default representation:

var list = new ArrayList<>();
list.add(1);
list.add(2);
new LJV()
        .setTreatAsPrimitive(Integer.class)
        .setStringRepresentation(List.class, list -> list.getClass().getSimpleName());

With a full package name and identity:

...
    .setStringRepresentation(List.class, list -> list.getClass().getName() + "@"  + System.identityHashCode(list))

String example:

...
    .setStringRepresentation(String.class, str -> str)
    .setByteStringRepresentation(b -> "0x" + Integer.toHexString(Byte.toUnsignedInt(b)))

I can create a separate task if you like this proposal.

[inponomarev]

• Hi, concerning array visualization design: agreed, first <td> must contain the array type. We will implement it this way.

• Concerning the string representation: yes, I believe this idea deserves a separate Github issue. To begin with, I'd say that inventing a good design for it must be a difficult task:

  • Should it be class-based, as you proposed, or object-based or even field-based (that is, what if we want to have a hexadecimal representation for a byte only when it's a specific field of a specific object, and decimal representation for all the other cases?)
  • Can we implement it in the spirit of attribute providers?
  • Maybe we should globally refactor LJV first? (btw -- our next big aim is to use JOL instead of Reflection API for LJV)

[orionll]

I created #30. I looked at attributes providers. For me, they don't look like a suitable solution for this (but I may be wrong). I think we need a new type of provider (ObjectLabelProvider). In the future, we can add other types of providers (e.g. FieldLabelProvider).

I like the idea of using JOL for more precise visualization of objects graphs (exact order of fields, object sizes...). However, I think LJV should stay reflection-based by default and JOL should be an optional plugin (probably even in a separate module).

[inponomarev]

Hi @orionll ! An update 3 years later: finally project LJV now is the part of JOL and I am able to continue maintaining it as a part of JOL (before that it was not possible or reasonable, as introspection via reflection is not working in modern Java). This ticket is the first thing we are fixing :-)

[inponomarev]

Closed by https://github.com/openjdk/jol/pull/57