[Bug]: str(b'', encoding) raises LookupError for an unknown encoding instead of returning '' #1223

Describe the bug

str() with an encoding or errors argument looks up the codec even when the input is empty, so str(b'', 'x') raises LookupError: unknown encoding x. A CPython release build returns '' for empty input without looking up the codec.

Operating system

Linux

CPU architecture

ARM64

GraalPy version

25.0.2 (Python 3.12.8); also reproduced on 25.4.4.1.1 (Python 3.13.14)

JDK version

No response

Context configuration

No response

Steps to reproduce

print(repr(str(b'', 'x')))

GraalPy 25.0.2:

LookupError: unknown encoding x

Expected behavior

CPython 3.12.13 (same on 3.13.16):

''

Stack trace

Additional context

str(b'', 'x', 'y'), str(bytearray(), 'x'), str(memoryview(b''), 'x'), b''.decode('x') and bytearray().decode('x') diverge the same way. Non-empty input agrees: str(b'a', 'x') raises LookupError on both.

CPython raises for the empty cases too under -X dev. Outside dev mode, devmode.rst#L80-L82 documents that "the encoding argument is sometimes ignored for empty strings".

Root Cause

StrNewNode.doBuffer (StringBuiltins.java#L278-L314) copies the buffer into a new PBytes and hands it to BytesCommonBuiltins.DecodeNode (#L302-L304), which looks up the codec without checking the length (BytesCommonBuiltins.java#L203-L214). GraalPy's own PyUnicodeFromEncodedObject already returns '' for empty input (#L84-L87, #L116-L119), but str() does not go through it. CPython's unicode_new_impl calls PyUnicode_FromEncodedObject (unicodeobject.c#L14723), which returns '' for an empty bytes or buffer (#L3311-L3317, #L3337-L3343) and validates encoding and errors only in dev mode or a debug build (#L584-L588):

// Objects/unicodeobject.c, PyUnicode_FromEncodedObject (v3.12.13, line 3310)
/* Decoding bytes objects is the most common case and should be fast */
if (PyBytes_Check(obj)) {
    if (PyBytes_GET_SIZE(obj) == 0) {
        if (unicode_check_encoding_errors(encoding, errors) < 0) {
            return NULL;
        }
        _Py_RETURN_UNICODE_EMPTY();
    }

Fix Suggestion

Return the empty string when bufferLib.getBufferLength(buffer) == 0 in doBuffer and in doNativeSubclassEncodeErr (#L339-L374), as PyUnicodeFromEncodedObject.doBuffer does. Putting the check in BytesCommonBuiltins.DecodeNode.decode instead also covers bytes.decode and bytearray.decode.