python.md

Python and your language server

Starter code for Python has no from typing import ..., and no definition of ListNode / TreeNode / a problem-specific Node / Interval. The judge supplies all of it implicitly — Optional, List, ListNode and friends are already in scope by the time your Solution class runs. A language server does not know that, so a perfectly valid solution lights up with undefined-name diagnostics.

C++ solves the same problem by force-including a header only clangd sees (see cpp.md) — the solution file itself is never touched. Pyright and pylsp have no equivalent: there is no way to put a name in scope without it actually being there. So instead, the plugin inserts a real, clearly-marked block at the top of your solution file, and strips exactly that block back out before every local run and every submission. runner.python.auto_imports = false turns it off.

What gets inserted

# --- meatcode: auto-imports for your editor (stripped before running/submitting) ---
from typing import *


class ListNode:
    def __init__(self, val=0, next=None):
        self.val = val
        self.next = next
# --- meatcode: end auto-imports ---

class Solution:
    def addTwoNumbers(self, l1: Optional[ListNode], l2: Optional[ListNode]) -> Optional[ListNode]:
        ...

• from typing import * only appears when the starter actually mentions a typing name (Optional, List, Dict, Union, ...). A problem like "A + B" that only uses int gets no header at all.
• The helper type is lifted from the starter's own leading comment or docstring — the same # class ListNode: ... (LeetCode, NeetCode) or """Definition of ListNode: class ListNode(object): ...""" (LintCode) text every provider already ships to document the type its judge injects. This is per-problem, the same way the C++ header is: "Copy List with Random Pointer" gets a Node with a random field, "Clone Graph" gets one with neighbors, because each is lifted from that problem's own starter instead of a single generic shape.
• When a starter names ListNode/TreeNode but carries no such comment, the plugin falls back to the same shape lua/meatcode/runner/harness/python.py's local-run harness already injects for every problem, so editing and running never disagree about the shape.

Guarantees

• A local run, a judge test run (the cloud oracle) and a submission all see your solution with the block removed — never anything beyond what you wrote. lua/meatcode/runner/python_prelude.lua is the single strip point they all flow through.
• Resetting to starter code (<leader>nR) re-inserts the block; it is re-derived from the starter each time, never hand-maintained.
• The block only ever appears when something in the starter looks like it needs it. Nothing is inserted for a problem that needs none of it.