[debug] Jump to Cursor feedback

[vadimcn]

After having tried to implement the new "Jump to Cursor" functionality, I'd like to share some feedback:

• VSCode does not surface errors returned from GotoRequest. Because GotoTargetsRequest does not specify which thread will be jumped, it is not possible to validate locations at that stage. (For example, language runtime may not allow moving execution point outside of the current function).
• The lifetime of GotoTarget not not clearly spelled out in the DAP spec. Does adapter need to remember ids of all GotoTargets it had ever returned?
• It would be helpful if GotoTarget contained adapter-defined data, that were round-tripped via subsequent GotoRequest. This way adapter wouldn't need to remember GotoTarget<->id association at all.

[gregg-miskelly]

FYI the first issue is also a problem for the C# DA.

Also, I think the answer for the second question is that target ids should be valid until the next continue. Though specifying this in the spec sounds reasonable to me.

[isidorn]

@vadimcn @gregg-miskelly thanks for feedback.
I have pushed a commit which tackles the first issue. Please try it out and let me know how it behaves for you.

@weinand leaving up the other two issues for you.

[weinand]

The general rule for ID lifetime in DAP is:

• IDs of "ephemeral" objects are valid while in the stopped state (or as Gregg said: "until the next 'continue'"). Examples are variable references, goto targets, step-in target, frames.
• IDs of long living objects are valid at least for the lifetime of the debug session (e.g. processes, threads), or even longer (breakpoints, data breakpoints).

I have created an issue in the DAP-repo for the clarification request: microsoft/debug-adapter-protocol#60.

Round-tripping adapter-defined data is not an option any more because the DAP already uses IDs to allow for communicating arbitrary data between "GotoTarget" and "Goto" requests. In addition it is cheaper to round-trip only an ID instead of the whole data.

[weinand]

Assigning to @isidor because he provided the only real code fix for the three issues.

[isidorn]

@vadimcn @gregg-miskelly can you please verify the 1) is fixed for you so I add a verified label. Thanks

[gregg-miskelly]

LGTM

[vadimcn]

Yeah, #1 looks fixed.

[isidorn]

Thanks. Adding verfieid label.