PEP657 Column position of raised exceptions

[tonybaloney]

In Debugpy, exceptions raised in Python 3.11 are missing the fine-grained position:

For example, in the following code:

d = {
    "x": {
        "y": {
            "z": {
                "theta": 1
            }
        }
    }
}

result = d["x"]["y"]["z"]["beta"]

On the CLI I get the position of the KeyError offset in the line underlined using PEP657 API:

$ python exploding_kitten.py
...
  File "/Users/anthonyshaw/tmp/exploding_kitten.py", line 11, in <module>
    result = d["x"]["y"]["z"]["beta"]
             ~~~~~~~~~~~~~~~~^^^^^^^^
KeyError: 'beta'

But in VS Code it only highlights the line and not the position of the exception, making it harder to work out where the error came from.

Please can PEP657 be implemented in debugpy so that the Stopped event sent back via DAP include the position (start col, end col) of the raised exception.

[pablogsal]

Notice that the traceback module implements the highlighting so you can use it if you don't want to make a custom highlighter:

import traceback

d = {
    "x": {
        "y": {
            "z": {
                "theta": 1
            }
        }
    }
}

try:
    result = d["x"]["y"]["z"]["beta"]
except Exception as e:
    f = e

print("\n".join(traceback.extract_tb(f.__traceback__).format()))

This prints:

  File "example.py", line 14, in <module>
    result = d["x"]["y"]["z"]["beta"]
             ~~~~~~~~~~~~~~~~^^^^^^^^

Note: this needs to be executed from a file as source is not displayed in the REPL

[tonybaloney]

Thanks for the update. Is there a way to propagate the col offset through DAP to the stop-point message? That way, the arrow inside the editor will point to the column and not highlight the whole line

[fabioz]

It seems that the stack trace from the DAP does indeed have a column. I'll reopen to investigate further.

[fabioz]

Getting the actual column where the frame is stopped seems to have a number of intricacies related to it...

@pablogsal I see that there's a way to format the traceback, but following the code it seems there's a lot of internal things to map the bytecode offset to the character (going from bytes -> char, parsing using the ast to find a BinOp or Subscript and then computing anchors based on that...).

I guess I can copy all that code to get the start column for the current pos in a frame or analyze the formatted string looking for ^, but I was wondering if there was a better API to deal with that somewhere...

[pablogsal]

I am not sure I follow... why you need to copy that code when you can just execute it? If you format the trace back with the traceback module, you will get all those things computed for you.

I am clearly missing something here. Could you please explain in detail what is what you are trying to do and why you cannot leverage existing APIs?

[fabioz]

I'd like to get the column (because I want to show the executing column in the editor and not just in the traceback string).

[fabioz]

@pablogsal the image below shows the use-case (the traceback module does many additional operations to infer the final position during the traceback printing).

[pablogsal]

I am afraid you are going to need to explain the problem with more details because I am missing a ton of context here. What exactly are you trying to achieve?

Precisely because the traceback module does all these extra things, I am asking you why you cannot just use the APIs provided in the traceback module and just print the traceback?

[pablogsal]

If you are just trying to get the column you can just use the co_positions attribute in the code object in combination with the instruction offset from the exception or the frame.