gh-155499: Fix curses border() and box() rejecting 0 mixed with str/bytes

[Lohitha0-0]

Fixes #155499

border() and box() were raising TypeError when you mix a one-character str/bytes arg with a plain 0, even though 0 is documented as meaning "use the default character" for that spot. This broke in GH-151758, which added a check to stop mixing wide-char types with int/bytes types, but didn't account for 0 being a valid sentinel value regardless of type.

For border(), the fix is small — just treat an int argument as fine when its value is 0, in addition to the existing wide-char case.

box() needed a bit more work. It was checking verch and horch together (t1 != 2 || t2 != 2), so it would already reject the call before it even got to check whether the individual value was 0. I restructured it to resolve each side's cchar_t pointer on its own, so 0 is accepted per-side instead of requiring both sides to match types.

Added a test (test_border_box_zero_sentinel) covering both functions with 0 mixed in, and also checking that a real type mismatch (like 1 mixed with a str) still correctly raises TypeError.

• Issue: curses border() and box() no longer accept 0 with a str #155499

[picnixz]

You can merge that check with the check at L2708

[Lohitha0-0]

Done, merged both branches into a single objs[i] == NULL || (types[i] == 1 && ch[i] == 0) condition.

[picnixz]

I feel that we could have a claener if-else if-else and keep an if (cond) { raise } and later wch1p = c2 ? NULL : &wch1; and same for wch2p (namely we can simplify the checks and diff). Unfortunately it's too late now for me to think in terms of boolean arithmetic so I'll let you figure it out.

[Lohitha0-0]

reworked it to a single upfront check (t1_ok/t2_ok) plus a ternary for the pointer assignment, no more goto/label.

[serhiy-storchaka]

Thank you for the patch.

Only the integer 0 is documented as the request for the default character, but the patch gives b'\0' the same meaning — it is an ordinary byte character.

The sentinel is also not the only regression from GH-151758: a string character could be mixed with byte and integer characters, including the ACS_* constants, in 3.14, but not in main.

#155871 solves both: border() and box() now choose the drawing function by what the arguments need, so only a complexchar cannot be mixed with integer and byte characters, and a string character is narrowed for them.

One note for future patches: test_border_box_zero_sentinel asserts TypeError for mixing but is not guarded by @requires_wide_build — a narrow build has no mixing check, so it would fail there.