I don't understand why this is necessary. Where is the downcast happening?

I'm sorry, I forgot to copy the compiler warning. I looked at a recent AMD64 build on Windows buildbots:

d:\buildarea\3.x.bolen-windows10\build\python\ast.c(4316): warning C4244: '+=': conversion from '__int64' to 'int', possible loss of data [D:\buildarea\3.x.bolen-windows10\build\PCbuild\pythoncore.vcxproj]

ast.c:4316:

    int cols = parent->n_col_offset;
(...)
            cols += substr - start;  // <--- HERE

The warning is because substr and start are addresses so they're 64 bit. But their subtraction is never going to exceed 32 bits (a single line of code exceeding 4 billion characters? Nope).

Rather than change col_offset from an int to Py_ssize_t, maybe let's just cast the subtraction on line 4316 to (int)?

If you'd rather keep Py_ssize_t, also change lineno to this type.

The warning is because substr and start are addresses so they're 64 bit. But their subtraction is never going to exceed 32 bits (a single line of code exceeding 4 billion characters? Nope).

I looked again at the code, and I think that using Py_ssize_t is wrong because node uses int:

typedef struct _node {
    ...
    int                 n_lineno;
    int                 n_col_offset;
    ...
} node;

I modified my PR to explicitly downcast to int.

@ambv: Please replace # with GH- in the commit message next time. Thanks!

Exactly! Thanks! ✨ 🍰 ✨