How to store invalid text typed at arbitrary places?

[enikao]

Especially unexperienced MPS users often try to type too much text without pressing Ctrl-Space, or type at incorrect places ("red text").
MPS stores these entries at some weird places, but not with the model. Thus, the unbound text is not part of version control, cannot be shared, etc.

How do we handle this in LIonWeb?

Option A: Don't store

Do the same as MPS: Assume the "red text" is not part of the model.
(A tool might keep track of it independent of the model.)

Pro:

• Compatible with MPS
• No weird structures in serialization

Con:

• Our tool either "forgets" user input, or needs additional storage

Option B: Extra serialization entry per node

A tool finds the "closest" node for each invalid text, and stores the text (+ some localization info) with the node.

Example:

M2:

concept Person
  property name: String
  property age: Int
  containment address: Address

concept Address
  property street: String
  property city: String

Projection (plain: constant text, italic: user input, bold: error):

My firstname is Alice with age of 29years, I live in main streetAvenue in Tokyo.

[
{
  "type": "Person",
  "id": "aaaa",
  "properties": {
    "name": "Alice",
    "age": "29"
  },
  "children": {
    "address": [
      "bbbb"
    ]
  },
  "redText": [
    {
      "position": "projection start + 3 characters",
      "text": "first"
    },
    {
      "position": "after age property",
      "text": "years"
    }
  ]
},
{
  "type": "Address",
  "id": "bbbb",
  "properties": {
    "street": "main",
    "city": "Tokyo"
  },
  "redText": [
    {
      "position": "+7",
      "text": "Avenue"
    }
  ]
}
]

Pro:

• Keeps red text context as much as possible
• No additional storage needed
• Non-model content clearly separated

Con:

• Position info depends on projection
• Complicates serialization format

Option C: Annotation on node

Assumes we support annotations

Same as option B, but store redText entry as annotation to the node (instead of separate member in the node serialization).

Pro:

• Keeps red text context as much as possible
• No additional storage needed
• Less invasive on serialization format

Con:

• Needs annotation support

Option D: Store globally with model

Keep a list of "redText" separate of nodes in serialization.

Pro:

• Don't need to find context node

Con:

• No scoping: If we request only a part of the model, we always have to send all red text known to the whole system
• position info gets even more complicated

[enikao]

Can somebody think of other options?

[ftomassetti]

I cannot think of other options, but I would be in favor of annotations as this seems something relevant only for certain use cases and not for all usages of models (i.e., this is useful for projectional editors and perhaps for textual editors, but not for code generators or transformets, I think)

[joswarmer]

A. I am ok with this
B. Position is a rather vague term here, so I would not like this. Also having a special property to store this feels like a hack.
C. If we have annotations support in the M# (has this been decide yet?), this might be a good option.
D. Seems useless of not connected to a node.