summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorEric V. Smith <eric@trueblade.com>2015-09-19 18:51:32 (GMT)
committerEric V. Smith <eric@trueblade.com>2015-09-19 18:51:32 (GMT)
commit235a6f09847ad554d8bf073d4e1d58d1e398ae8c (patch)
tree36ff217247cfcd108914065cea8ddf3ad056d192
parentaed8830af3bb5a79878cf0f603ebbd8a37f5b36e (diff)
downloadcpython-235a6f09847ad554d8bf073d4e1d58d1e398ae8c.zip
cpython-235a6f09847ad554d8bf073d4e1d58d1e398ae8c.tar.gz
cpython-235a6f09847ad554d8bf073d4e1d58d1e398ae8c.tar.bz2
Issue #24965: Implement PEP 498 "Literal String Interpolation". Documentation is still needed, I'll open an issue for that.
-rw-r--r--Include/Python-ast.h23
-rw-r--r--Lib/test/test_fstring.py715
-rw-r--r--Misc/NEWS5
-rw-r--r--Parser/Python.asdl2
-rw-r--r--Parser/tokenizer.c8
-rw-r--r--Python/Python-ast.c165
-rw-r--r--Python/ast.c985
-rw-r--r--Python/compile.c117
-rw-r--r--Python/symtable.c8
9 files changed, 1965 insertions, 63 deletions
diff --git a/Include/Python-ast.h b/Include/Python-ast.h
index 3bc015f..ea6679c 100644
--- a/Include/Python-ast.h
+++ b/Include/Python-ast.h
@@ -201,9 +201,10 @@ enum _expr_kind {BoolOp_kind=1, BinOp_kind=2, UnaryOp_kind=3, Lambda_kind=4,
SetComp_kind=9, DictComp_kind=10, GeneratorExp_kind=11,
Await_kind=12, Yield_kind=13, YieldFrom_kind=14,
Compare_kind=15, Call_kind=16, Num_kind=17, Str_kind=18,
- Bytes_kind=19, NameConstant_kind=20, Ellipsis_kind=21,
- Attribute_kind=22, Subscript_kind=23, Starred_kind=24,
- Name_kind=25, List_kind=26, Tuple_kind=27};
+ FormattedValue_kind=19, JoinedStr_kind=20, Bytes_kind=21,
+ NameConstant_kind=22, Ellipsis_kind=23, Attribute_kind=24,
+ Subscript_kind=25, Starred_kind=26, Name_kind=27,
+ List_kind=28, Tuple_kind=29};
struct _expr {
enum _expr_kind kind;
union {
@@ -297,6 +298,16 @@ struct _expr {
} Str;
struct {
+ expr_ty value;
+ int conversion;
+ expr_ty format_spec;
+ } FormattedValue;
+
+ struct {
+ asdl_seq *values;
+ } JoinedStr;
+
+ struct {
bytes s;
} Bytes;
@@ -543,6 +554,12 @@ expr_ty _Py_Call(expr_ty func, asdl_seq * args, asdl_seq * keywords, int
expr_ty _Py_Num(object n, int lineno, int col_offset, PyArena *arena);
#define Str(a0, a1, a2, a3) _Py_Str(a0, a1, a2, a3)
expr_ty _Py_Str(string s, int lineno, int col_offset, PyArena *arena);
+#define FormattedValue(a0, a1, a2, a3, a4, a5) _Py_FormattedValue(a0, a1, a2, a3, a4, a5)
+expr_ty _Py_FormattedValue(expr_ty value, int conversion, expr_ty format_spec,
+ int lineno, int col_offset, PyArena *arena);
+#define JoinedStr(a0, a1, a2, a3) _Py_JoinedStr(a0, a1, a2, a3)
+expr_ty _Py_JoinedStr(asdl_seq * values, int lineno, int col_offset, PyArena
+ *arena);
#define Bytes(a0, a1, a2, a3) _Py_Bytes(a0, a1, a2, a3)
expr_ty _Py_Bytes(bytes s, int lineno, int col_offset, PyArena *arena);
#define NameConstant(a0, a1, a2, a3) _Py_NameConstant(a0, a1, a2, a3)
diff --git a/Lib/test/test_fstring.py b/Lib/test/test_fstring.py
new file mode 100644
index 0000000..a6ff9cf
--- /dev/null
+++ b/Lib/test/test_fstring.py
@@ -0,0 +1,715 @@
+import ast
+import types
+import decimal
+import unittest
+
+a_global = 'global variable'
+
+# You could argue that I'm too strict in looking for specific error
+# values with assertRaisesRegex, but without it it's way too easy to
+# make a syntax error in the test strings. Especially with all of the
+# triple quotes, raw strings, backslashes, etc. I think it's a
+# worthwhile tradeoff. When I switched to this method, I found many
+# examples where I wasn't testing what I thought I was.
+
+class TestCase(unittest.TestCase):
+ def assertAllRaise(self, exception_type, regex, error_strings):
+ for str in error_strings:
+ with self.subTest(str=str):
+ with self.assertRaisesRegex(exception_type, regex):
+ eval(str)
+
+ def test__format__lookup(self):
+ # Make sure __format__ is looked up on the type, not the instance.
+ class X:
+ def __format__(self, spec):
+ return 'class'
+
+ x = X()
+
+ # Add a bound __format__ method to the 'y' instance, but not
+ # the 'x' instance.
+ y = X()
+ y.__format__ = types.MethodType(lambda self, spec: 'instance', y)
+
+ self.assertEqual(f'{y}', format(y))
+ self.assertEqual(f'{y}', 'class')
+ self.assertEqual(format(x), format(y))
+
+ # __format__ is not called this way, but still make sure it
+ # returns what we expect (so we can make sure we're bypassing
+ # it).
+ self.assertEqual(x.__format__(''), 'class')
+ self.assertEqual(y.__format__(''), 'instance')
+
+ # This is how __format__ is actually called.
+ self.assertEqual(type(x).__format__(x, ''), 'class')
+ self.assertEqual(type(y).__format__(y, ''), 'class')
+
+ def test_ast(self):
+ # Inspired by http://bugs.python.org/issue24975
+ class X:
+ def __init__(self):
+ self.called = False
+ def __call__(self):
+ self.called = True
+ return 4
+ x = X()
+ expr = """
+a = 10
+f'{a * x()}'"""
+ t = ast.parse(expr)
+ c = compile(t, '', 'exec')
+
+ # Make sure x was not called.
+ self.assertFalse(x.called)
+
+ # Actually run the code.
+ exec(c)
+
+ # Make sure x was called.
+ self.assertTrue(x.called)
+
+ def test_literal_eval(self):
+ # With no expressions, an f-string is okay.
+ self.assertEqual(ast.literal_eval("f'x'"), 'x')
+ self.assertEqual(ast.literal_eval("f'x' 'y'"), 'xy')
+
+ # But this should raise an error.
+ with self.assertRaisesRegex(ValueError, 'malformed node or string'):
+ ast.literal_eval("f'x{3}'")
+
+ # As should this, which uses a different ast node
+ with self.assertRaisesRegex(ValueError, 'malformed node or string'):
+ ast.literal_eval("f'{3}'")
+
+ def test_ast_compile_time_concat(self):
+ x = ['']
+
+ expr = """x[0] = 'foo' f'{3}'"""
+ t = ast.parse(expr)
+ c = compile(t, '', 'exec')
+ exec(c)
+ self.assertEqual(x[0], 'foo3')
+
+ def test_literal(self):
+ self.assertEqual(f'', '')
+ self.assertEqual(f'a', 'a')
+ self.assertEqual(f' ', ' ')
+ self.assertEqual(f'\N{GREEK CAPITAL LETTER DELTA}',
+ '\N{GREEK CAPITAL LETTER DELTA}')
+ self.assertEqual(f'\N{GREEK CAPITAL LETTER DELTA}',
+ '\u0394')
+ self.assertEqual(f'\N{True}', '\u22a8')
+ self.assertEqual(rf'\N{True}', r'\NTrue')
+
+ def test_escape_order(self):
+ # note that hex(ord('{')) == 0x7b, so this
+ # string becomes f'a{4*10}b'
+ self.assertEqual(f'a\u007b4*10}b', 'a40b')
+ self.assertEqual(f'a\x7b4*10}b', 'a40b')
+ self.assertEqual(f'a\x7b4*10\N{RIGHT CURLY BRACKET}b', 'a40b')
+ self.assertEqual(f'{"a"!\N{LATIN SMALL LETTER R}}', "'a'")
+ self.assertEqual(f'{10\x3a02X}', '0A')
+ self.assertEqual(f'{10:02\N{LATIN CAPITAL LETTER X}}', '0A')
+
+ self.assertAllRaise(SyntaxError, "f-string: single '}' is not allowed",
+ [r"""f'a{\u007b4*10}b'""", # mis-matched brackets
+ ])
+ self.assertAllRaise(SyntaxError, 'unexpected character after line continuation character',
+ [r"""f'{"a"\!r}'""",
+ r"""f'{a\!r}'""",
+ ])
+
+ def test_unterminated_string(self):
+ self.assertAllRaise(SyntaxError, 'f-string: unterminated string',
+ [r"""f'{"x'""",
+ r"""f'{"x}'""",
+ r"""f'{("x'""",
+ r"""f'{("x}'""",
+ ])
+
+ def test_mismatched_parens(self):
+ self.assertAllRaise(SyntaxError, 'f-string: mismatched',
+ ["f'{((}'",
+ ])
+
+ def test_double_braces(self):
+ self.assertEqual(f'{{', '{')
+ self.assertEqual(f'a{{', 'a{')
+ self.assertEqual(f'{{b', '{b')
+ self.assertEqual(f'a{{b', 'a{b')
+ self.assertEqual(f'}}', '}')
+ self.assertEqual(f'a}}', 'a}')
+ self.assertEqual(f'}}b', '}b')
+ self.assertEqual(f'a}}b', 'a}b')
+
+ self.assertEqual(f'{{{10}', '{10')
+ self.assertEqual(f'}}{10}', '}10')
+ self.assertEqual(f'}}{{{10}', '}{10')
+ self.assertEqual(f'}}a{{{10}', '}a{10')
+
+ self.assertEqual(f'{10}{{', '10{')
+ self.assertEqual(f'{10}}}', '10}')
+ self.assertEqual(f'{10}}}{{', '10}{')
+ self.assertEqual(f'{10}}}a{{' '}', '10}a{}')
+
+ # Inside of strings, don't interpret doubled brackets.
+ self.assertEqual(f'{"{{}}"}', '{{}}')
+
+ self.assertAllRaise(TypeError, 'unhashable type',
+ ["f'{ {{}} }'", # dict in a set
+ ])
+
+ def test_compile_time_concat(self):
+ x = 'def'
+ self.assertEqual('abc' f'## {x}ghi', 'abc## defghi')
+ self.assertEqual('abc' f'{x}' 'ghi', 'abcdefghi')
+ self.assertEqual('abc' f'{x}' 'gh' f'i{x:4}', 'abcdefghidef ')
+ self.assertEqual('{x}' f'{x}', '{x}def')
+ self.assertEqual('{x' f'{x}', '{xdef')
+ self.assertEqual('{x}' f'{x}', '{x}def')
+ self.assertEqual('{{x}}' f'{x}', '{{x}}def')
+ self.assertEqual('{{x' f'{x}', '{{xdef')
+ self.assertEqual('x}}' f'{x}', 'x}}def')
+ self.assertEqual(f'{x}' 'x}}', 'defx}}')
+ self.assertEqual(f'{x}' '', 'def')
+ self.assertEqual('' f'{x}' '', 'def')
+ self.assertEqual('' f'{x}', 'def')
+ self.assertEqual(f'{x}' '2', 'def2')
+ self.assertEqual('1' f'{x}' '2', '1def2')
+ self.assertEqual('1' f'{x}', '1def')
+ self.assertEqual(f'{x}' f'-{x}', 'def-def')
+ self.assertEqual('' f'', '')
+ self.assertEqual('' f'' '', '')
+ self.assertEqual('' f'' '' f'', '')
+ self.assertEqual(f'', '')
+ self.assertEqual(f'' '', '')
+ self.assertEqual(f'' '' f'', '')
+ self.assertEqual(f'' '' f'' '', '')
+
+ self.assertAllRaise(SyntaxError, "f-string: expecting '}'",
+ ["f'{3' f'}'", # can't concat to get a valid f-string
+ ])
+
+ def test_comments(self):
+ # These aren't comments, since they're in strings.
+ d = {'#': 'hash'}
+ self.assertEqual(f'{"#"}', '#')
+ self.assertEqual(f'{d["#"]}', 'hash')
+
+ self.assertAllRaise(SyntaxError, "f-string cannot include '#'",
+ ["f'{1#}'", # error because the expression becomes "(1#)"
+ "f'{3(#)}'",
+ ])
+
+ def test_many_expressions(self):
+ # Create a string with many expressions in it. Note that
+ # because we have a space in here as a literal, we're actually
+ # going to use twice as many ast nodes: one for each literal
+ # plus one for each expression.
+ def build_fstr(n, extra=''):
+ return "f'" + ('{x} ' * n) + extra + "'"
+
+ x = 'X'
+ width = 1
+
+ # Test around 256.
+ for i in range(250, 260):
+ self.assertEqual(eval(build_fstr(i)), (x+' ')*i)
+
+ # Test concatenating 2 largs fstrings.
+ self.assertEqual(eval(build_fstr(255)*256), (x+' ')*(255*256))
+
+ s = build_fstr(253, '{x:{width}} ')
+ self.assertEqual(eval(s), (x+' ')*254)
+
+ # Test lots of expressions and constants, concatenated.
+ s = "f'{1}' 'x' 'y'" * 1024
+ self.assertEqual(eval(s), '1xy' * 1024)
+
+ def test_format_specifier_expressions(self):
+ width = 10
+ precision = 4
+ value = decimal.Decimal('12.34567')
+ self.assertEqual(f'result: {value:{width}.{precision}}', 'result: 12.35')
+ self.assertEqual(f'result: {value:{width!r}.{precision}}', 'result: 12.35')
+ self.assertEqual(f'result: {value:{width:0}.{precision:1}}', 'result: 12.35')
+ self.assertEqual(f'result: {value:{1}{0:0}.{precision:1}}', 'result: 12.35')
+ self.assertEqual(f'result: {value:{ 1}{ 0:0}.{ precision:1}}', 'result: 12.35')
+ self.assertEqual(f'{10:#{1}0x}', ' 0xa')
+ self.assertEqual(f'{10:{"#"}1{0}{"x"}}', ' 0xa')
+ self.assertEqual(f'{-10:-{"#"}1{0}x}', ' -0xa')
+ self.assertEqual(f'{-10:{"-"}#{1}0{"x"}}', ' -0xa')
+ self.assertEqual(f'{10:#{3 != {4:5} and width}x}', ' 0xa')
+
+ self.assertAllRaise(SyntaxError, "f-string: expecting '}'",
+ ["""f'{"s"!r{":10"}}'""",
+
+ # This looks like a nested format spec.
+ ])
+
+ self.assertAllRaise(SyntaxError, "invalid syntax",
+ [# Invalid sytax inside a nested spec.
+ "f'{4:{/5}}'",
+ ])
+
+ self.assertAllRaise(SyntaxError, "f-string: expressions nested too deeply",
+ [# Can't nest format specifiers.
+ "f'result: {value:{width:{0}}.{precision:1}}'",
+ ])
+
+ self.assertAllRaise(SyntaxError, 'f-string: invalid conversion character',
+ [# No expansion inside conversion or for
+ # the : or ! itself.
+ """f'{"s"!{"r"}}'""",
+ ])
+
+ def test_side_effect_order(self):
+ class X:
+ def __init__(self):
+ self.i = 0
+ def __format__(self, spec):
+ self.i += 1
+ return str(self.i)
+
+ x = X()
+ self.assertEqual(f'{x} {x}', '1 2')
+
+ def test_missing_expression(self):
+ self.assertAllRaise(SyntaxError, 'f-string: empty expression not allowed',
+ ["f'{}'",
+ "f'{ }'"
+ "f' {} '",
+ "f'{!r}'",
+ "f'{ !r}'",
+ "f'{10:{ }}'",
+ "f' { } '",
+ r"f'{\n}'",
+ r"f'{\n \n}'",
+ ])
+
+ def test_parens_in_expressions(self):
+ self.assertEqual(f'{3,}', '(3,)')
+
+ # Add these because when an expression is evaluated, parens
+ # are added around it. But we shouldn't go from an invalid
+ # expression to a valid one. The added parens are just
+ # supposed to allow whitespace (including newlines).
+ self.assertAllRaise(SyntaxError, 'invalid syntax',
+ ["f'{,}'",
+ "f'{,}'", # this is (,), which is an error
+ ])
+
+ self.assertAllRaise(SyntaxError, "f-string: expecting '}'",
+ ["f'{3)+(4}'",
+ ])
+
+ self.assertAllRaise(SyntaxError, 'EOL while scanning string literal',
+ ["f'{\n}'",
+ ])
+
+ def test_newlines_in_expressions(self):
+ self.assertEqual(f'{0}', '0')
+ self.assertEqual(f'{0\n}', '0')
+ self.assertEqual(f'{0\r}', '0')
+ self.assertEqual(f'{\n0\n}', '0')
+ self.assertEqual(f'{\r0\r}', '0')
+ self.assertEqual(f'{\n0\r}', '0')
+ self.assertEqual(f'{\n0}', '0')
+ self.assertEqual(f'{3+\n4}', '7')
+ self.assertEqual(f'{3+\\\n4}', '7')
+ self.assertEqual(rf'''{3+
+4}''', '7')
+ self.assertEqual(f'''{3+\
+4}''', '7')
+
+ self.assertAllRaise(SyntaxError, 'f-string: empty expression not allowed',
+ [r"f'{\n}'",
+ ])
+
+ def test_lambda(self):
+ x = 5
+ self.assertEqual(f'{(lambda y:x*y)("8")!r}', "'88888'")
+ self.assertEqual(f'{(lambda y:x*y)("8")!r:10}', "'88888' ")
+ self.assertEqual(f'{(lambda y:x*y)("8"):10}', "88888 ")
+
+ # lambda doesn't work without parens, because the colon
+ # makes the parser think it's a format_spec
+ self.assertAllRaise(SyntaxError, 'unexpected EOF while parsing',
+ ["f'{lambda x:x}'",
+ ])
+
+ def test_yield(self):
+ # Not terribly useful, but make sure the yield turns
+ # a function into a generator
+ def fn(y):
+ f'y:{yield y*2}'
+
+ g = fn(4)
+ self.assertEqual(next(g), 8)
+
+ def test_yield_send(self):
+ def fn(x):
+ yield f'x:{yield (lambda i: x * i)}'
+
+ g = fn(10)
+ the_lambda = next(g)
+ self.assertEqual(the_lambda(4), 40)
+ self.assertEqual(g.send('string'), 'x:string')
+
+ def test_expressions_with_triple_quoted_strings(self):
+ self.assertEqual(f"{'''x'''}", 'x')
+ self.assertEqual(f"{'''eric's'''}", "eric's")
+ self.assertEqual(f'{"""eric\'s"""}', "eric's")
+ self.assertEqual(f"{'''eric\"s'''}", 'eric"s')
+ self.assertEqual(f'{"""eric"s"""}', 'eric"s')
+
+ # Test concatenation within an expression
+ self.assertEqual(f'{"x" """eric"s""" "y"}', 'xeric"sy')
+ self.assertEqual(f'{"x" """eric"s"""}', 'xeric"s')
+ self.assertEqual(f'{"""eric"s""" "y"}', 'eric"sy')
+ self.assertEqual(f'{"""x""" """eric"s""" "y"}', 'xeric"sy')
+ self.assertEqual(f'{"""x""" """eric"s""" """y"""}', 'xeric"sy')
+ self.assertEqual(f'{r"""x""" """eric"s""" """y"""}', 'xeric"sy')
+
+ def test_multiple_vars(self):
+ x = 98
+ y = 'abc'
+ self.assertEqual(f'{x}{y}', '98abc')
+
+ self.assertEqual(f'X{x}{y}', 'X98abc')
+ self.assertEqual(f'{x}X{y}', '98Xabc')
+ self.assertEqual(f'{x}{y}X', '98abcX')
+
+ self.assertEqual(f'X{x}Y{y}', 'X98Yabc')
+ self.assertEqual(f'X{x}{y}Y', 'X98abcY')
+ self.assertEqual(f'{x}X{y}Y', '98XabcY')
+
+ self.assertEqual(f'X{x}Y{y}Z', 'X98YabcZ')
+
+ def test_closure(self):
+ def outer(x):
+ def inner():
+ return f'x:{x}'
+ return inner
+
+ self.assertEqual(outer('987')(), 'x:987')
+ self.assertEqual(outer(7)(), 'x:7')
+
+ def test_arguments(self):
+ y = 2
+ def f(x, width):
+ return f'x={x*y:{width}}'
+
+ self.assertEqual(f('foo', 10), 'x=foofoo ')
+ x = 'bar'
+ self.assertEqual(f(10, 10), 'x= 20')
+
+ def test_locals(self):
+ value = 123
+ self.assertEqual(f'v:{value}', 'v:123')
+
+ def test_missing_variable(self):
+ with self.assertRaises(NameError):
+ f'v:{value}'
+
+ def test_missing_format_spec(self):
+ class O:
+ def __format__(self, spec):
+ if not spec:
+ return '*'
+ return spec
+
+ self.assertEqual(f'{O():x}', 'x')
+ self.assertEqual(f'{O()}', '*')
+ self.assertEqual(f'{O():}', '*')
+
+ self.assertEqual(f'{3:}', '3')
+ self.assertEqual(f'{3!s:}', '3')
+
+ def test_global(self):
+ self.assertEqual(f'g:{a_global}', 'g:global variable')
+ self.assertEqual(f'g:{a_global!r}', "g:'global variable'")
+
+ a_local = 'local variable'
+ self.assertEqual(f'g:{a_global} l:{a_local}',
+ 'g:global variable l:local variable')
+ self.assertEqual(f'g:{a_global!r}',
+ "g:'global variable'")
+ self.assertEqual(f'g:{a_global} l:{a_local!r}',
+ "g:global variable l:'local variable'")
+
+ self.assertIn("module 'unittest' from", f'{unittest}')
+
+ def test_shadowed_global(self):
+ a_global = 'really a local'
+ self.assertEqual(f'g:{a_global}', 'g:really a local')
+ self.assertEqual(f'g:{a_global!r}', "g:'really a local'")
+
+ a_local = 'local variable'
+ self.assertEqual(f'g:{a_global} l:{a_local}',
+ 'g:really a local l:local variable')
+ self.assertEqual(f'g:{a_global!r}',
+ "g:'really a local'")
+ self.assertEqual(f'g:{a_global} l:{a_local!r}',
+ "g:really a local l:'local variable'")
+
+ def test_call(self):
+ def foo(x):
+ return 'x=' + str(x)
+
+ self.assertEqual(f'{foo(10)}', 'x=10')
+
+ def test_nested_fstrings(self):
+ y = 5
+ self.assertEqual(f'{f"{0}"*3}', '000')
+ self.assertEqual(f'{f"{y}"*3}', '555')
+ self.assertEqual(f'{f"{\'x\'}"*3}', 'xxx')
+
+ self.assertEqual(f"{r'x' f'{\"s\"}'}", 'xs')
+ self.assertEqual(f"{r'x'rf'{\"s\"}'}", 'xs')
+
+ def test_invalid_string_prefixes(self):
+ self.assertAllRaise(SyntaxError, 'unexpected EOF while parsing',
+ ["fu''",
+ "uf''",
+ "Fu''",
+ "fU''",
+ "Uf''",
+ "uF''",
+ "ufr''",
+ "urf''",
+ "fur''",
+ "fru''",
+ "rfu''",
+ "ruf''",
+ "FUR''",
+ "Fur''",
+ ])
+
+ def test_leading_trailing_spaces(self):
+ self.assertEqual(f'{ 3}', '3')
+ self.assertEqual(f'{ 3}', '3')
+ self.assertEqual(f'{\t3}', '3')
+ self.assertEqual(f'{\t\t3}', '3')
+ self.assertEqual(f'{3 }', '3')
+ self.assertEqual(f'{3 }', '3')
+ self.assertEqual(f'{3\t}', '3')
+ self.assertEqual(f'{3\t\t}', '3')
+
+ self.assertEqual(f'expr={ {x: y for x, y in [(1, 2), ]}}',
+ 'expr={1: 2}')
+ self.assertEqual(f'expr={ {x: y for x, y in [(1, 2), ]} }',
+ 'expr={1: 2}')
+
+ def test_character_name(self):
+ self.assertEqual(f'{4}\N{GREEK CAPITAL LETTER DELTA}{3}',
+ '4\N{GREEK CAPITAL LETTER DELTA}3')
+ self.assertEqual(f'{{}}\N{GREEK CAPITAL LETTER DELTA}{3}',
+ '{}\N{GREEK CAPITAL LETTER DELTA}3')
+
+ def test_not_equal(self):
+ # There's a special test for this because there's a special
+ # case in the f-string parser to look for != as not ending an
+ # expression. Normally it would, while looking for !s or !r.
+
+ self.assertEqual(f'{3!=4}', 'True')
+ self.assertEqual(f'{3!=4:}', 'True')
+ self.assertEqual(f'{3!=4!s}', 'True')
+ self.assertEqual(f'{3!=4!s:.3}', 'Tru')
+
+ def test_conversions(self):
+ self.assertEqual(f'{3.14:10.10}', ' 3.14')
+ self.assertEqual(f'{3.14!s:10.10}', '3.14 ')
+ self.assertEqual(f'{3.14!r:10.10}', '3.14 ')
+ self.assertEqual(f'{3.14!a:10.10}', '3.14 ')
+
+ self.assertEqual(f'{"a"}', 'a')
+ self.assertEqual(f'{"a"!r}', "'a'")
+ self.assertEqual(f'{"a"!a}', "'a'")
+
+ # Not a conversion.
+ self.assertEqual(f'{"a!r"}', "a!r")
+
+ # Not a conversion, but show that ! is allowed in a format spec.
+ self.assertEqual(f'{3.14:!<10.10}', '3.14!!!!!!')
+
+ self.assertEqual(f'{"\N{GREEK CAPITAL LETTER DELTA}"}', '\u0394')
+ self.assertEqual(f'{"\N{GREEK CAPITAL LETTER DELTA}"!r}', "'\u0394'")
+ self.assertEqual(f'{"\N{GREEK CAPITAL LETTER DELTA}"!a}', "'\\u0394'")
+
+ self.assertAllRaise(SyntaxError, 'f-string: invalid conversion character',
+ ["f'{3!g}'",
+ "f'{3!A}'",
+ "f'{3!A}'",
+ "f'{3!A}'",
+ "f'{3!!}'",
+ "f'{3!:}'",
+ "f'{3!\N{GREEK CAPITAL LETTER DELTA}}'",
+ "f'{3! s}'", # no space before conversion char
+ "f'{x!\\x00:.<10}'",
+ ])
+
+ self.assertAllRaise(SyntaxError, "f-string: expecting '}'",
+ ["f'{x!s{y}}'",
+ "f'{3!ss}'",
+ "f'{3!ss:}'",
+ "f'{3!ss:s}'",
+ ])
+
+ def test_assignment(self):
+ self.assertAllRaise(SyntaxError, 'invalid syntax',
+ ["f'' = 3",
+ "f'{0}' = x",
+ "f'{x}' = x",
+ ])
+
+ def test_del(self):
+ self.assertAllRaise(SyntaxError, 'invalid syntax',
+ ["del f''",
+ "del '' f''",
+ ])
+
+ def test_mismatched_braces(self):
+ self.assertAllRaise(SyntaxError, "f-string: single '}' is not allowed",
+ ["f'{{}'",
+ "f'{{}}}'",
+ "f'}'",
+ "f'x}'",
+ "f'x}x'",
+
+ # Can't have { or } in a format spec.
+ "f'{3:}>10}'",
+ r"f'{3:\\}>10}'",
+ "f'{3:}}>10}'",
+ ])
+
+ self.assertAllRaise(SyntaxError, "f-string: expecting '}'",
+ ["f'{3:{{>10}'",
+ "f'{3'",
+ "f'{3!'",
+ "f'{3:'",
+ "f'{3!s'",
+ "f'{3!s:'",
+ "f'{3!s:3'",
+ "f'x{'",
+ "f'x{x'",
+ "f'{3:s'",
+ "f'{{{'",
+ "f'{{}}{'",
+ "f'{'",
+ ])
+
+ self.assertAllRaise(SyntaxError, 'invalid syntax',
+ [r"f'{3:\\{>10}'",
+ ])
+
+ # But these are just normal strings.
+ self.assertEqual(f'{"{"}', '{')
+ self.assertEqual(f'{"}"}', '}')
+ self.assertEqual(f'{3:{"}"}>10}', '}}}}}}}}}3')
+ self.assertEqual(f'{2:{"{"}>10}', '{{{{{{{{{2')
+
+ def test_if_conditional(self):
+ # There's special logic in compile.c to test if the
+ # conditional for an if (and while) are constants. Exercise
+ # that code.
+
+ def test_fstring(x, expected):
+ flag = 0
+ if f'{x}':
+ flag = 1
+ else:
+ flag = 2
+ self.assertEqual(flag, expected)
+
+ def test_concat_empty(x, expected):
+ flag = 0
+ if '' f'{x}':
+ flag = 1
+ else:
+ flag = 2
+ self.assertEqual(flag, expected)
+
+ def test_concat_non_empty(x, expected):
+ flag = 0
+ if ' ' f'{x}':
+ flag = 1
+ else:
+ flag = 2
+ self.assertEqual(flag, expected)
+
+ test_fstring('', 2)
+ test_fstring(' ', 1)
+
+ test_concat_empty('', 2)
+ test_concat_empty(' ', 1)
+
+ test_concat_non_empty('', 1)
+ test_concat_non_empty(' ', 1)
+
+ def test_empty_format_specifier(self):
+ x = 'test'
+ self.assertEqual(f'{x}', 'test')
+ self.assertEqual(f'{x:}', 'test')
+ self.assertEqual(f'{x!s:}', 'test')
+ self.assertEqual(f'{x!r:}', "'test'")
+
+ def test_str_format_differences(self):
+ d = {'a': 'string',
+ 0: 'integer',
+ }
+ a = 0
+ self.assertEqual(f'{d[0]}', 'integer')
+ self.assertEqual(f'{d["a"]}', 'string')
+ self.assertEqual(f'{d[a]}', 'integer')
+ self.assertEqual('{d[a]}'.format(d=d), 'string')
+ self.assertEqual('{d[0]}'.format(d=d), 'integer')
+
+ def test_invalid_expressions(self):
+ self.assertAllRaise(SyntaxError, 'invalid syntax',
+ [r"f'{a[4)}'",
+ r"f'{a(4]}'",
+ ])
+
+ def test_loop(self):
+ for i in range(1000):
+ self.assertEqual(f'i:{i}', 'i:' + str(i))
+
+ def test_dict(self):
+ d = {'"': 'dquote',
+ "'": 'squote',
+ 'foo': 'bar',
+ }
+ self.assertEqual(f'{d["\'"]}', 'squote')
+ self.assertEqual(f"{d['\"']}", 'dquote')
+
+ self.assertEqual(f'''{d["'"]}''', 'squote')
+ self.assertEqual(f"""{d['"']}""", 'dquote')
+
+ self.assertEqual(f'{d["foo"]}', 'bar')
+ self.assertEqual(f"{d['foo']}", 'bar')
+ self.assertEqual(f'{d[\'foo\']}', 'bar')
+ self.assertEqual(f"{d[\"foo\"]}", 'bar')
+
+ def test_escaped_quotes(self):
+ d = {'"': 'a',
+ "'": 'b'}
+
+ self.assertEqual(fr"{d['\"']}", 'a')
+ self.assertEqual(fr'{d["\'"]}', 'b')
+ self.assertEqual(fr"{'\"'}", '"')
+ self.assertEqual(fr'{"\'"}', "'")
+ self.assertEqual(f'{"\\"3"}', '"3')
+
+ self.assertAllRaise(SyntaxError, 'f-string: unterminated string',
+ [r'''f'{"""\\}' ''', # Backslash at end of expression
+ ])
+ self.assertAllRaise(SyntaxError, 'unexpected character after line continuation',
+ [r"rf'{3\}'",
+ ])
+
+
+if __name__ == '__main__':
+ unittest.main()
diff --git a/Misc/NEWS b/Misc/NEWS
index a2c76f5..46f7a3e 100644
--- a/Misc/NEWS
+++ b/Misc/NEWS
@@ -19,6 +19,11 @@ Core and Builtins
argument list of a function declaration. For example, "def f(*, a =
3,): pass" is now legal. Patch from Mark Dickinson.
+- Issue #24965: Implement PEP 498 "Literal String Interpolation". This
+ allows you to embed expressions inside f-strings, which are
+ converted to normal strings at run time. Given x=3, then
+ f'value={x}' == 'value=3'. Patch by Eric V. Smith.
+
Library
-------
diff --git a/Parser/Python.asdl b/Parser/Python.asdl
index cd0832d..22775c6 100644
--- a/Parser/Python.asdl
+++ b/Parser/Python.asdl
@@ -71,6 +71,8 @@ module Python
| Call(expr func, expr* args, keyword* keywords)
| Num(object n) -- a number as a PyObject.
| Str(string s) -- need to specify raw, unicode, etc?
+ | FormattedValue(expr value, int? conversion, expr? format_spec)
+ | JoinedStr(expr* values)
| Bytes(bytes s)
| NameConstant(singleton value)
| Ellipsis
diff --git a/Parser/tokenizer.c b/Parser/tokenizer.c
index 5046fa5..2369be4 100644
--- a/Parser/tokenizer.c
+++ b/Parser/tokenizer.c
@@ -1477,17 +1477,19 @@ tok_get(struct tok_state *tok, char **p_start, char **p_end)
nonascii = 0;
if (is_potential_identifier_start(c)) {
/* Process b"", r"", u"", br"" and rb"" */
- int saw_b = 0, saw_r = 0, saw_u = 0;
+ int saw_b = 0, saw_r = 0, saw_u = 0, saw_f = 0;
while (1) {
- if (!(saw_b || saw_u) && (c == 'b' || c == 'B'))
+ if (!(saw_b || saw_u || saw_f) && (c == 'b' || c == 'B'))
saw_b = 1;
/* Since this is a backwards compatibility support literal we don't
want to support it in arbitrary order like byte literals. */
- else if (!(saw_b || saw_u || saw_r) && (c == 'u' || c == 'U'))
+ else if (!(saw_b || saw_u || saw_r || saw_f) && (c == 'u' || c == 'U'))
saw_u = 1;
/* ur"" and ru"" are not supported */
else if (!(saw_r || saw_u) && (c == 'r' || c == 'R'))
saw_r = 1;
+ else if (!(saw_f || saw_b || saw_u) && (c == 'f' || c == 'F'))
+ saw_f = 1;
else
break;
c = tok_nextc(tok);
diff --git a/Python/Python-ast.c b/Python/Python-ast.c
index fd7f17e..a2e9816 100644
--- a/Python/Python-ast.c
+++ b/Python/Python-ast.c
@@ -285,6 +285,18 @@ _Py_IDENTIFIER(s);
static char *Str_fields[]={
"s",
};
+static PyTypeObject *FormattedValue_type;
+_Py_IDENTIFIER(conversion);
+_Py_IDENTIFIER(format_spec);
+static char *FormattedValue_fields[]={
+ "value",
+ "conversion",
+ "format_spec",
+};
+static PyTypeObject *JoinedStr_type;
+static char *JoinedStr_fields[]={
+ "values",
+};
static PyTypeObject *Bytes_type;
static char *Bytes_fields[]={
"s",
@@ -917,6 +929,11 @@ static int init_types(void)
if (!Num_type) return 0;
Str_type = make_type("Str", expr_type, Str_fields, 1);
if (!Str_type) return 0;
+ FormattedValue_type = make_type("FormattedValue", expr_type,
+ FormattedValue_fields, 3);
+ if (!FormattedValue_type) return 0;
+ JoinedStr_type = make_type("JoinedStr", expr_type, JoinedStr_fields, 1);
+ if (!JoinedStr_type) return 0;
Bytes_type = make_type("Bytes", expr_type, Bytes_fields, 1);
if (!Bytes_type) return 0;
NameConstant_type = make_type("NameConstant", expr_type,
@@ -2063,6 +2080,42 @@ Str(string s, int lineno, int col_offset, PyArena *arena)
}
expr_ty
+FormattedValue(expr_ty value, int conversion, expr_ty format_spec, int lineno,
+ int col_offset, PyArena *arena)
+{
+ expr_ty p;
+ if (!value) {
+ PyErr_SetString(PyExc_ValueError,
+ "field value is required for FormattedValue");
+ return NULL;
+ }
+ p = (expr_ty)PyArena_Malloc(arena, sizeof(*p));
+ if (!p)
+ return NULL;
+ p->kind = FormattedValue_kind;
+ p->v.FormattedValue.value = value;
+ p->v.FormattedValue.conversion = conversion;
+ p->v.FormattedValue.format_spec = format_spec;
+ p->lineno = lineno;
+ p->col_offset = col_offset;
+ return p;
+}
+
+expr_ty
+JoinedStr(asdl_seq * values, int lineno, int col_offset, PyArena *arena)
+{
+ expr_ty p;
+ p = (expr_ty)PyArena_Malloc(arena, sizeof(*p));
+ if (!p)
+ return NULL;
+ p->kind = JoinedStr_kind;
+ p->v.JoinedStr.values = values;
+ p->lineno = lineno;
+ p->col_offset = col_offset;
+ return p;
+}
+
+expr_ty
Bytes(bytes s, int lineno, int col_offset, PyArena *arena)
{
expr_ty p;
@@ -3161,6 +3214,34 @@ ast2obj_expr(void* _o)
goto failed;
Py_DECREF(value);
break;
+ case FormattedValue_kind:
+ result = PyType_GenericNew(FormattedValue_type, NULL, NULL);
+ if (!result) goto failed;
+ value = ast2obj_expr(o->v.FormattedValue.value);
+ if (!value) goto failed;
+ if (_PyObject_SetAttrId(result, &PyId_value, value) == -1)
+ goto failed;
+ Py_DECREF(value);
+ value = ast2obj_int(o->v.FormattedValue.conversion);
+ if (!value) goto failed;
+ if (_PyObject_SetAttrId(result, &PyId_conversion, value) == -1)
+ goto failed;
+ Py_DECREF(value);
+ value = ast2obj_expr(o->v.FormattedValue.format_spec);
+ if (!value) goto failed;
+ if (_PyObject_SetAttrId(result, &PyId_format_spec, value) == -1)
+ goto failed;
+ Py_DECREF(value);
+ break;
+ case JoinedStr_kind:
+ result = PyType_GenericNew(JoinedStr_type, NULL, NULL);
+ if (!result) goto failed;
+ value = ast2obj_list(o->v.JoinedStr.values, ast2obj_expr);
+ if (!value) goto failed;
+ if (_PyObject_SetAttrId(result, &PyId_values, value) == -1)
+ goto failed;
+ Py_DECREF(value);
+ break;
case Bytes_kind:
result = PyType_GenericNew(Bytes_type, NULL, NULL);
if (!result) goto failed;
@@ -6022,6 +6103,86 @@ obj2ast_expr(PyObject* obj, expr_ty* out, PyArena* arena)
if (*out == NULL) goto failed;
return 0;
}
+ isinstance = PyObject_IsInstance(obj, (PyObject*)FormattedValue_type);
+ if (isinstance == -1) {
+ return 1;
+ }
+ if (isinstance) {
+ expr_ty value;
+ int conversion;
+ expr_ty format_spec;
+
+ if (_PyObject_HasAttrId(obj, &PyId_value)) {
+ int res;
+ tmp = _PyObject_GetAttrId(obj, &PyId_value);
+ if (tmp == NULL) goto failed;
+ res = obj2ast_expr(tmp, &value, arena);
+ if (res != 0) goto failed;
+ Py_CLEAR(tmp);
+ } else {
+ PyErr_SetString(PyExc_TypeError, "required field \"value\" missing from FormattedValue");
+ return 1;
+ }
+ if (exists_not_none(obj, &PyId_conversion)) {
+ int res;
+ tmp = _PyObject_GetAttrId(obj, &PyId_conversion);
+ if (tmp == NULL) goto failed;
+ res = obj2ast_int(tmp, &conversion, arena);
+ if (res != 0) goto failed;
+ Py_CLEAR(tmp);
+ } else {
+ conversion = 0;
+ }
+ if (exists_not_none(obj, &PyId_format_spec)) {
+ int res;
+ tmp = _PyObject_GetAttrId(obj, &PyId_format_spec);
+ if (tmp == NULL) goto failed;
+ res = obj2ast_expr(tmp, &format_spec, arena);
+ if (res != 0) goto failed;
+ Py_CLEAR(tmp);
+ } else {
+ format_spec = NULL;
+ }
+ *out = FormattedValue(value, conversion, format_spec, lineno,
+ col_offset, arena);
+ if (*out == NULL) goto failed;
+ return 0;
+ }
+ isinstance = PyObject_IsInstance(obj, (PyObject*)JoinedStr_type);
+ if (isinstance == -1) {
+ return 1;
+ }
+ if (isinstance) {
+ asdl_seq* values;
+
+ if (_PyObject_HasAttrId(obj, &PyId_values)) {
+ int res;
+ Py_ssize_t len;
+ Py_ssize_t i;
+ tmp = _PyObject_GetAttrId(obj, &PyId_values);
+ if (tmp == NULL) goto failed;
+ if (!PyList_Check(tmp)) {
+ PyErr_Format(PyExc_TypeError, "JoinedStr field \"values\" must be a list, not a %.200s", tmp->ob_type->tp_name);
+ goto failed;
+ }
+ len = PyList_GET_SIZE(tmp);
+ values = _Py_asdl_seq_new(len, arena);
+ if (values == NULL) goto failed;
+ for (i = 0; i < len; i++) {
+ expr_ty value;
+ res = obj2ast_expr(PyList_GET_ITEM(tmp, i), &value, arena);
+ if (res != 0) goto failed;
+ asdl_seq_SET(values, i, value);
+ }
+ Py_CLEAR(tmp);
+ } else {
+ PyErr_SetString(PyExc_TypeError, "required field \"values\" missing from JoinedStr");
+ return 1;
+ }
+ *out = JoinedStr(values, lineno, col_offset, arena);
+ if (*out == NULL) goto failed;
+ return 0;
+ }
isinstance = PyObject_IsInstance(obj, (PyObject*)Bytes_type);
if (isinstance == -1) {
return 1;
@@ -7319,6 +7480,10 @@ PyInit__ast(void)
if (PyDict_SetItemString(d, "Call", (PyObject*)Call_type) < 0) return NULL;
if (PyDict_SetItemString(d, "Num", (PyObject*)Num_type) < 0) return NULL;
if (PyDict_SetItemString(d, "Str", (PyObject*)Str_type) < 0) return NULL;
+ if (PyDict_SetItemString(d, "FormattedValue",
+ (PyObject*)FormattedValue_type) < 0) return NULL;
+ if (PyDict_SetItemString(d, "JoinedStr", (PyObject*)JoinedStr_type) < 0)
+ return NULL;
if (PyDict_SetItemString(d, "Bytes", (PyObject*)Bytes_type) < 0) return
NULL;
if (PyDict_SetItemString(d, "NameConstant", (PyObject*)NameConstant_type) <
diff --git a/Python/ast.c b/Python/ast.c
index 1f7ddfc..735424b 100644
--- a/Python/ast.c
+++ b/Python/ast.c
@@ -257,6 +257,14 @@ validate_expr(expr_ty exp, expr_context_ty ctx)
}
return 1;
}
+ case JoinedStr_kind:
+ return validate_exprs(exp->v.JoinedStr.values, Load, 0);
+ case FormattedValue_kind:
+ if (validate_expr(exp->v.FormattedValue.value, Load) == 0)
+ return 0;
+ if (exp->v.FormattedValue.format_spec)
+ return validate_expr(exp->v.FormattedValue.format_spec, Load);
+ return 1;
case Bytes_kind: {
PyObject *b = exp->v.Bytes.s;
if (!PyBytes_CheckExact(b)) {
@@ -535,9 +543,7 @@ static stmt_ty ast_for_for_stmt(struct compiling *, const node *, int);
static expr_ty ast_for_call(struct compiling *, const node *, expr_ty);
static PyObject *parsenumber(struct compiling *, const char *);
-static PyObject *parsestr(struct compiling *, const node *n, int *bytesmode);
-static PyObject *parsestrplus(struct compiling *, const node *n,
- int *bytesmode);
+static expr_ty parsestrplus(struct compiling *, const node *n);
#define COMP_GENEXP 0
#define COMP_LISTCOMP 1
@@ -986,6 +992,8 @@ set_context(struct compiling *c, expr_ty e, expr_context_ty ctx, const node *n)
case Num_kind:
case Str_kind:
case Bytes_kind:
+ case JoinedStr_kind:
+ case FormattedValue_kind:
expr_name = "literal";
break;
case NameConstant_kind:
@@ -2001,7 +2009,6 @@ ast_for_atom(struct compiling *c, const node *n)
| '...' | 'None' | 'True' | 'False'
*/
node *ch = CHILD(n, 0);
- int bytesmode = 0;
switch (TYPE(ch)) {
case NAME: {
@@ -2023,7 +2030,7 @@ ast_for_atom(struct compiling *c, const node *n)
return Name(name, Load, LINENO(n), n->n_col_offset, c->c_arena);
}
case STRING: {
- PyObject *str = parsestrplus(c, n, &bytesmode);
+ expr_ty str = parsestrplus(c, n);
if (!str) {
const char *errtype = NULL;
if (PyErr_ExceptionMatches(PyExc_UnicodeError))
@@ -2050,14 +2057,7 @@ ast_for_atom(struct compiling *c, const node *n)
}
return NULL;
}
- if (PyArena_AddPyObject(c->c_arena, str) < 0) {
- Py_DECREF(str);
- return NULL;
- }
- if (bytesmode)
- return Bytes(str, LINENO(n), n->n_col_offset, c->c_arena);
- else
- return Str(str, LINENO(n), n->n_col_offset, c->c_arena);
+ return str;
}
case NUMBER: {
PyObject *pynum = parsenumber(c, STR(ch));
@@ -4002,12 +4002,838 @@ decode_unicode(struct compiling *c, const char *s, size_t len, int rawmode, cons
return v;
}
-/* s is a Python string literal, including the bracketing quote characters,
- * and r &/or b prefixes (if any), and embedded escape sequences (if any).
- * parsestr parses it, and returns the decoded Python string object.
- */
+/* Compile this expression in to an expr_ty. We know that we can
+ temporarily modify the character before the start of this string
+ (it's '{'), and we know we can temporarily modify the character
+ after this string (it is a '}'). Leverage this to create a
+ sub-string with enough room for us to add parens around the
+ expression. This is to allow strings with embedded newlines, for
+ example. */
+static expr_ty
+fstring_expression_compile(PyObject *str, Py_ssize_t expr_start,
+ Py_ssize_t expr_end, PyArena *arena)
+{
+ PyCompilerFlags cf;
+ mod_ty mod;
+ char *utf_expr;
+ Py_ssize_t i;
+ int all_whitespace;
+ PyObject *sub = NULL;
+
+ /* We only decref sub if we allocated it with a PyUnicode_Substring.
+ decref_sub records that. */
+ int decref_sub = 0;
+
+ assert(str);
+
+ /* If the substring is all whitespace, it's an error. We need to
+ catch this here, and not when we call PyParser_ASTFromString,
+ because turning the expression '' in to '()' would go from
+ being invalid to valid. */
+ /* Note that this code says an empty string is all
+ whitespace. That's important. There's a test for it: f'{}'. */
+ all_whitespace = 1;
+ for (i = expr_start; i < expr_end; i++) {
+ if (!Py_UNICODE_ISSPACE(PyUnicode_READ_CHAR(str, i))) {
+ all_whitespace = 0;
+ break;
+ }
+ }
+ if (all_whitespace) {
+ PyErr_SetString(PyExc_SyntaxError, "f-string: empty expression "
+ "not allowed");
+ goto error;
+ }
+
+ /* If the substring will be the entire source string, we can't use
+ PyUnicode_Substring, since it will return another reference to
+ our original string. Because we're modifying the string in
+ place, that's a no-no. So, detect that case and just use our
+ string directly. */
+
+ if (expr_start-1 == 0 && expr_end+1 == PyUnicode_GET_LENGTH(str)) {
+ /* No need to actually remember these characters, because we
+ know they must be braces. */
+ assert(PyUnicode_ReadChar(str, 0) == '{');
+ assert(PyUnicode_ReadChar(str, expr_end-expr_start+1) == '}');
+ sub = str;
+ } else {
+ /* Create a substring object. It must be a new object, with
+ refcount==1, so that we can modify it. */
+ sub = PyUnicode_Substring(str, expr_start-1, expr_end+1);
+ if (!sub)
+ goto error;
+ assert(sub != str); /* Make sure it's a new string. */
+ decref_sub = 1; /* Remember to deallocate it on error. */
+ }
+
+ if (PyUnicode_WriteChar(sub, 0, '(') < 0 ||
+ PyUnicode_WriteChar(sub, expr_end-expr_start+1, ')') < 0)
+ goto error;
+
+ cf.cf_flags = PyCF_ONLY_AST;
+
+ /* No need to free the memory returned here: it's managed by the
+ string. */
+ utf_expr = PyUnicode_AsUTF8(sub);
+ if (!utf_expr)
+ goto error;
+ mod = PyParser_ASTFromString(utf_expr, "<fstring>",
+ Py_eval_input, &cf, arena);
+ if (!mod)
+ goto error;
+ if (sub != str)
+ /* Clear instead of decref in case we ever modify this code to change
+ the error handling: this is safest because the XDECREF won't try
+ and decref it when it's NULL. */
+ /* No need to restore the chars in sub, since we know it's getting
+ ready to get deleted (refcount must be 1, since we got a new string
+ in PyUnicode_Substring). */
+ Py_CLEAR(sub);
+ else {
+ assert(!decref_sub);
+ /* Restore str, which we earlier modified directly. */
+ if (PyUnicode_WriteChar(str, 0, '{') < 0 ||
+ PyUnicode_WriteChar(str, expr_end-expr_start+1, '}') < 0)
+ goto error;
+ }
+ return mod->v.Expression.body;
+
+error:
+ /* Only decref sub if it was the result of a call to SubString. */
+ if (decref_sub)
+ Py_XDECREF(sub);
+ return NULL;
+}
+
+/* Return -1 on error.
+
+ Return 0 if we reached the end of the literal.
+
+ Return 1 if we haven't reached the end of the literal, but we want
+ the caller to process the literal up to this point. Used for
+ doubled braces.
+*/
+static int
+fstring_find_literal(PyObject *str, Py_ssize_t *ofs, PyObject **literal,
+ int recurse_lvl, struct compiling *c, const node *n)
+{
+ /* Get any literal string. It ends when we hit an un-doubled brace, or the
+ end of the string. */
+
+ Py_ssize_t literal_start, literal_end;
+ int result = 0;
+
+ enum PyUnicode_Kind kind = PyUnicode_KIND(str);
+ void *data = PyUnicode_DATA(str);
+
+ assert(*literal == NULL);
+
+ literal_start = *ofs;
+ for (; *ofs < PyUnicode_GET_LENGTH(str); *ofs += 1) {
+ Py_UCS4 ch = PyUnicode_READ(kind, data, *ofs);
+ if (ch == '{' || ch == '}') {
+ /* Check for doubled braces, but only at the top level. If
+ we checked at every level, then f'{0:{3}}' would fail
+ with the two closing braces. */
+ if (recurse_lvl == 0) {
+ if (*ofs + 1 < PyUnicode_GET_LENGTH(str) &&
+ PyUnicode_READ(kind, data, *ofs + 1) == ch) {
+ /* We're going to tell the caller that the literal ends
+ here, but that they should continue scanning. But also
+ skip over the second brace when we resume scanning. */
+ literal_end = *ofs + 1;
+ *ofs += 2;
+ result = 1;
+ goto done;
+ }
+
+ /* Where a single '{' is the start of a new expression, a
+ single '}' is not allowed. */
+ if (ch == '}') {
+ ast_error(c, n, "f-string: single '}' is not allowed");
+ return -1;
+ }
+ }
+
+ /* We're either at a '{', which means we're starting another
+ expression; or a '}', which means we're at the end of this
+ f-string (for a nested format_spec). */
+ break;
+ }
+ }
+ literal_end = *ofs;
+
+ assert(*ofs == PyUnicode_GET_LENGTH(str) ||
+ PyUnicode_READ(kind, data, *ofs) == '{' ||
+ PyUnicode_READ(kind, data, *ofs) == '}');
+done:
+ if (literal_start != literal_end) {
+ *literal = PyUnicode_Substring(str, literal_start, literal_end);
+ if (!*literal)
+ return -1;
+ }
+
+ return result;
+}
+
+/* Forward declaration because parsing is recursive. */
+static expr_ty
+fstring_parse(PyObject *str, Py_ssize_t *ofs, int recurse_lvl,
+ struct compiling *c, const node *n);
+
+/* Parse the f-string str, starting at ofs. We know *ofs starts an
+ expression (so it must be a '{'). Returns the FormattedValue node,
+ which includes the expression, conversion character, and
+ format_spec expression.
+
+ Note that I don't do a perfect job here: I don't make sure that a
+ closing brace doesn't match an opening paren, for example. It
+ doesn't need to error on all invalid expressions, just correctly
+ find the end of all valid ones. Any errors inside the expression
+ will be caught when we parse it later. */
+static int
+fstring_find_expr(PyObject *str, Py_ssize_t *ofs, int recurse_lvl,
+ expr_ty *expression, struct compiling *c, const node *n)
+{
+ /* Return -1 on error, else 0. */
+
+ Py_ssize_t expr_start;
+ Py_ssize_t expr_end;
+ expr_ty simple_expression;
+ expr_ty format_spec = NULL; /* Optional format specifier. */
+ Py_UCS4 conversion = -1; /* The conversion char. -1 if not specified. */
+
+ enum PyUnicode_Kind kind = PyUnicode_KIND(str);
+ void *data = PyUnicode_DATA(str);
+
+ /* 0 if we're not in a string, else the quote char we're trying to
+ match (single or double quote). */
+ Py_UCS4 quote_char = 0;
+
+ /* If we're inside a string, 1=normal, 3=triple-quoted. */
+ int string_type = 0;
+
+ /* Keep track of nesting level for braces/parens/brackets in
+ expressions. */
+ Py_ssize_t nested_depth = 0;
+
+ /* Can only nest one level deep. */
+ if (recurse_lvl >= 2) {
+ ast_error(c, n, "f-string: expressions nested too deeply");
+ return -1;
+ }
+
+ /* The first char must be a left brace, or we wouldn't have gotten
+ here. Skip over it. */
+ assert(PyUnicode_READ(kind, data, *ofs) == '{');
+ *ofs += 1;
+
+ expr_start = *ofs;
+ for (; *ofs < PyUnicode_GET_LENGTH(str); *ofs += 1) {
+ Py_UCS4 ch;
+
+ /* Loop invariants. */
+ assert(nested_depth >= 0);
+ assert(*ofs >= expr_start);
+ if (quote_char)
+ assert(string_type == 1 || string_type == 3);
+ else
+ assert(string_type == 0);
+
+ ch = PyUnicode_READ(kind, data, *ofs);
+ if (quote_char) {
+ /* We're inside a string. See if we're at the end. */
+ /* This code needs to implement the same non-error logic
+ as tok_get from tokenizer.c, at the letter_quote
+ label. To actually share that code would be a
+ nightmare. But, it's unlikely to change and is small,
+ so duplicate it here. Note we don't need to catch all
+ of the errors, since they'll be caught when parsing the
+ expression. We just need to match the non-error
+ cases. Thus we can ignore \n in single-quoted strings,
+ for example. Or non-terminated strings. */
+ if (ch == quote_char) {
+ /* Does this match the string_type (single or triple
+ quoted)? */
+ if (string_type == 3) {
+ if (*ofs+2 < PyUnicode_GET_LENGTH(str) &&
+ PyUnicode_READ(kind, data, *ofs+1) == ch &&
+ PyUnicode_READ(kind, data, *ofs+2) == ch) {
+ /* We're at the end of a triple quoted string. */
+ *ofs += 2;
+ string_type = 0;
+ quote_char = 0;
+ continue;
+ }
+ } else {
+ /* We're at the end of a normal string. */
+ quote_char = 0;
+ string_type = 0;
+ continue;
+ }
+ }
+ /* We're inside a string, and not finished with the
+ string. If this is a backslash, skip the next char (it
+ might be an end quote that needs skipping). Otherwise,
+ just consume this character normally. */
+ if (ch == '\\' && *ofs+1 < PyUnicode_GET_LENGTH(str)) {
+ /* Just skip the next char, whatever it is. */
+ *ofs += 1;
+ }
+ } else if (ch == '\'' || ch == '"') {
+ /* Is this a triple quoted string? */
+ if (*ofs+2 < PyUnicode_GET_LENGTH(str) &&
+ PyUnicode_READ(kind, data, *ofs+1) == ch &&
+ PyUnicode_READ(kind, data, *ofs+2) == ch) {
+ string_type = 3;
+ *ofs += 2;
+ } else {
+ /* Start of a normal string. */
+ string_type = 1;
+ }
+ /* Start looking for the end of the string. */
+ quote_char = ch;
+ } else if (ch == '[' || ch == '{' || ch == '(') {
+ nested_depth++;
+ } else if (nested_depth != 0 &&
+ (ch == ']' || ch == '}' || ch == ')')) {
+ nested_depth--;
+ } else if (ch == '#') {
+ /* Error: can't include a comment character, inside parens
+ or not. */
+ ast_error(c, n, "f-string cannot include '#'");
+ return -1;
+ } else if (nested_depth == 0 &&
+ (ch == '!' || ch == ':' || ch == '}')) {
+ /* First, test for the special case of "!=". Since '=' is
+ not an allowed conversion character, nothing is lost in
+ this test. */
+ if (ch == '!' && *ofs+1 < PyUnicode_GET_LENGTH(str) &&
+ PyUnicode_READ(kind, data, *ofs+1) == '=')
+ /* This isn't a conversion character, just continue. */
+ continue;
+
+ /* Normal way out of this loop. */
+ break;
+ } else {
+ /* Just consume this char and loop around. */
+ }
+ }
+ expr_end = *ofs;
+ /* If we leave this loop in a string or with mismatched parens, we
+ don't care. We'll get a syntax error when compiling the
+ expression. But, we can produce a better error message, so
+ let's just do that.*/
+ if (quote_char) {
+ ast_error(c, n, "f-string: unterminated string");
+ return -1;
+ }
+ if (nested_depth) {
+ ast_error(c, n, "f-string: mismatched '(', '{', or '['");
+ return -1;
+ }
+
+ /* Check for a conversion char, if present. */
+ if (*ofs >= PyUnicode_GET_LENGTH(str))
+ goto unexpected_end_of_string;
+ if (PyUnicode_READ(kind, data, *ofs) == '!') {
+ *ofs += 1;
+ if (*ofs >= PyUnicode_GET_LENGTH(str))
+ goto unexpected_end_of_string;
+
+ conversion = PyUnicode_READ(kind, data, *ofs);
+ *ofs += 1;
+
+ /* Validate the conversion. */
+ if (!(conversion == 's' || conversion == 'r'
+ || conversion == 'a')) {
+ ast_error(c, n, "f-string: invalid conversion character: "
+ "expected 's', 'r', or 'a'");
+ return -1;
+ }
+ }
+
+ /* Check for the format spec, if present. */
+ if (*ofs >= PyUnicode_GET_LENGTH(str))
+ goto unexpected_end_of_string;
+ if (PyUnicode_READ(kind, data, *ofs) == ':') {
+ *ofs += 1;
+ if (*ofs >= PyUnicode_GET_LENGTH(str))
+ goto unexpected_end_of_string;
+
+ /* Parse the format spec. */
+ format_spec = fstring_parse(str, ofs, recurse_lvl+1, c, n);
+ if (!format_spec)
+ return -1;
+ }
+
+ if (*ofs >= PyUnicode_GET_LENGTH(str) ||
+ PyUnicode_READ(kind, data, *ofs) != '}')
+ goto unexpected_end_of_string;
+
+ /* We're at a right brace. Consume it. */
+ assert(*ofs < PyUnicode_GET_LENGTH(str));
+ assert(PyUnicode_READ(kind, data, *ofs) == '}');
+ *ofs += 1;
+
+ /* Compile the expression. */
+ simple_expression = fstring_expression_compile(str, expr_start, expr_end,
+ c->c_arena);
+ if (!simple_expression)
+ return -1;
+
+ /* And now create the FormattedValue node that represents this entire
+ expression with the conversion and format spec. */
+ *expression = FormattedValue(simple_expression, (int)conversion,
+ format_spec, LINENO(n), n->n_col_offset,
+ c->c_arena);
+ if (!*expression)
+ return -1;
+
+ return 0;
+
+unexpected_end_of_string:
+ ast_error(c, n, "f-string: expecting '}'");
+ return -1;
+}
+
+/* Return -1 on error.
+
+ Return 0 if we have a literal (possible zero length) and an
+ expression (zero length if at the end of the string.
+
+ Return 1 if we have a literal, but no expression, and we want the
+ caller to call us again. This is used to deal with doubled
+ braces.
+
+ When called multiple times on the string 'a{{b{0}c', this function
+ will return:
+
+ 1. the literal 'a{' with no expression, and a return value
+ of 1. Despite the fact that there's no expression, the return
+ value of 1 means we're not finished yet.
+
+ 2. the literal 'b' and the expression '0', with a return value of
+ 0. The fact that there's an expression means we're not finished.
+
+ 3. literal 'c' with no expression and a return value of 0. The
+ combination of the return value of 0 with no expression means
+ we're finished.
+*/
+static int
+fstring_find_literal_and_expr(PyObject *str, Py_ssize_t *ofs, int recurse_lvl,
+ PyObject **literal, expr_ty *expression,
+ struct compiling *c, const node *n)
+{
+ int result;
+
+ assert(*literal == NULL && *expression == NULL);
+
+ /* Get any literal string. */
+ result = fstring_find_literal(str, ofs, literal, recurse_lvl, c, n);
+ if (result < 0)
+ goto error;
+
+ assert(result == 0 || result == 1);
+
+ if (result == 1)
+ /* We have a literal, but don't look at the expression. */
+ return 1;
+
+ assert(*ofs <= PyUnicode_GET_LENGTH(str));
+
+ if (*ofs >= PyUnicode_GET_LENGTH(str) ||
+ PyUnicode_READ_CHAR(str, *ofs) == '}')
+ /* We're at the end of the string or the end of a nested
+ f-string: no expression. The top-level error case where we
+ expect to be at the end of the string but we're at a '}' is
+ handled later. */
+ return 0;
+
+ /* We must now be the start of an expression, on a '{'. */
+ assert(*ofs < PyUnicode_GET_LENGTH(str) &&
+ PyUnicode_READ_CHAR(str, *ofs) == '{');
+
+ if (fstring_find_expr(str, ofs, recurse_lvl, expression, c, n) < 0)
+ goto error;
+
+ return 0;
+
+error:
+ Py_XDECREF(*literal);
+ *literal = NULL;
+ return -1;
+}
+
+#define EXPRLIST_N_CACHED 64
+
+typedef struct {
+ /* Incrementally build an array of expr_ty, so be used in an
+ asdl_seq. Cache some small but reasonably sized number of
+ expr_ty's, and then after that start dynamically allocating,
+ doubling the number allocated each time. Note that the f-string
+ f'{0}a{1}' contains 3 expr_ty's: 2 FormattedValue's, and one
+ Str for the literal 'a'. So you add expr_ty's about twice as
+ fast as you add exressions in an f-string. */
+
+ Py_ssize_t allocated; /* Number we've allocated. */
+ Py_ssize_t size; /* Number we've used. */
+ expr_ty *p; /* Pointer to the memory we're actually
+ using. Will point to 'data' until we
+ start dynamically allocating. */
+ expr_ty data[EXPRLIST_N_CACHED];
+} ExprList;
+
+#ifdef NDEBUG
+#define ExprList_check_invariants(l)
+#else
+static void
+ExprList_check_invariants(ExprList *l)
+{
+ /* Check our invariants. Make sure this object is "live", and
+ hasn't been deallocated. */
+ assert(l->size >= 0);
+ assert(l->p != NULL);
+ if (l->size <= EXPRLIST_N_CACHED)
+ assert(l->data == l->p);
+}
+#endif
+
+static void
+ExprList_Init(ExprList *l)
+{
+ l->allocated = EXPRLIST_N_CACHED;
+ l->size = 0;
+
+ /* Until we start allocating dynamically, p points to data. */
+ l->p = l->data;
+
+ ExprList_check_invariants(l);
+}
+
+static int
+ExprList_Append(ExprList *l, expr_ty exp)
+{
+ ExprList_check_invariants(l);
+ if (l->size >= l->allocated) {
+ /* We need to alloc (or realloc) the memory. */
+ Py_ssize_t new_size = l->allocated * 2;
+
+ /* See if we've ever allocated anything dynamically. */
+ if (l->p == l->data) {
+ Py_ssize_t i;
+ /* We're still using the cached data. Switch to
+ alloc-ing. */
+ l->p = PyMem_RawMalloc(sizeof(expr_ty) * new_size);
+ if (!l->p)
+ return -1;
+ /* Copy the cached data into the new buffer. */
+ for (i = 0; i < l->size; i++)
+ l->p[i] = l->data[i];
+ } else {
+ /* Just realloc. */
+ expr_ty *tmp = PyMem_RawRealloc(l->p, sizeof(expr_ty) * new_size);
+ if (!tmp) {
+ PyMem_RawFree(l->p);
+ l->p = NULL;
+ return -1;
+ }
+ l->p = tmp;
+ }
+
+ l->allocated = new_size;
+ assert(l->allocated == 2 * l->size);
+ }
+
+ l->p[l->size++] = exp;
+
+ ExprList_check_invariants(l);
+ return 0;
+}
+
+static void
+ExprList_Dealloc(ExprList *l)
+{
+ ExprList_check_invariants(l);
+
+ /* If there's been an error, or we've never dynamically allocated,
+ do nothing. */
+ if (!l->p || l->p == l->data) {
+ /* Do nothing. */
+ } else {
+ /* We have dynamically allocated. Free the memory. */
+ PyMem_RawFree(l->p);
+ }
+ l->p = NULL;
+ l->size = -1;
+}
+
+static asdl_seq *
+ExprList_Finish(ExprList *l, PyArena *arena)
+{
+ asdl_seq *seq;
+
+ ExprList_check_invariants(l);
+
+ /* Allocate the asdl_seq and copy the expressions in to it. */
+ seq = _Py_asdl_seq_new(l->size, arena);
+ if (seq) {
+ Py_ssize_t i;
+ for (i = 0; i < l->size; i++)
+ asdl_seq_SET(seq, i, l->p[i]);
+ }
+ ExprList_Dealloc(l);
+ return seq;
+}
+
+/* The FstringParser is designed to add a mix of strings and
+ f-strings, and concat them together as needed. Ultimately, it
+ generates an expr_ty. */
+typedef struct {
+ PyObject *last_str;
+ ExprList expr_list;
+} FstringParser;
+
+#ifdef NDEBUG
+#define FstringParser_check_invariants(state)
+#else
+static void
+FstringParser_check_invariants(FstringParser *state)
+{
+ if (state->last_str)
+ assert(PyUnicode_CheckExact(state->last_str));
+ ExprList_check_invariants(&state->expr_list);
+}
+#endif
+
+static void
+FstringParser_Init(FstringParser *state)
+{
+ state->last_str = NULL;
+ ExprList_Init(&state->expr_list);
+ FstringParser_check_invariants(state);
+}
+
+static void
+FstringParser_Dealloc(FstringParser *state)
+{
+ FstringParser_check_invariants(state);
+
+ Py_XDECREF(state->last_str);
+ ExprList_Dealloc(&state->expr_list);
+}
+
+/* Make a Str node, but decref the PyUnicode object being added. */
+static expr_ty
+make_str_node_and_del(PyObject **str, struct compiling *c, const node* n)
+{
+ PyObject *s = *str;
+ *str = NULL;
+ assert(PyUnicode_CheckExact(s));
+ if (PyArena_AddPyObject(c->c_arena, s) < 0) {
+ Py_DECREF(s);
+ return NULL;
+ }
+ return Str(s, LINENO(n), n->n_col_offset, c->c_arena);
+}
+
+/* Add a non-f-string (that is, a regular literal string). str is
+ decref'd. */
+static int
+FstringParser_ConcatAndDel(FstringParser *state, PyObject *str)
+{
+ FstringParser_check_invariants(state);
+
+ assert(PyUnicode_CheckExact(str));
+
+ if (PyUnicode_GET_LENGTH(str) == 0) {
+ Py_DECREF(str);
+ return 0;
+ }
+
+ if (!state->last_str) {
+ /* We didn't have a string before, so just remember this one. */
+ state->last_str = str;
+ } else {
+ /* Concatenate this with the previous string. */
+ PyObject *temp = PyUnicode_Concat(state->last_str, str);
+ Py_DECREF(state->last_str);
+ Py_DECREF(str);
+ state->last_str = temp;
+ if (!temp)
+ return -1;
+ }
+ FstringParser_check_invariants(state);
+ return 0;
+}
+
+/* Parse an f-string. The f-string is in str, starting at ofs, with no 'f'
+ or quotes. str is not decref'd, since we don't know if it's used elsewhere.
+ And if we're only looking at a part of a string, then decref'ing is
+ definitely not the right thing to do! */
+static int
+FstringParser_ConcatFstring(FstringParser *state, PyObject *str,
+ Py_ssize_t *ofs, int recurse_lvl,
+ struct compiling *c, const node *n)
+{
+ FstringParser_check_invariants(state);
+
+ /* Parse the f-string. */
+ while (1) {
+ PyObject *literal = NULL;
+ expr_ty expression = NULL;
+
+ /* If there's a zero length literal in front of the
+ expression, literal will be NULL. If we're at the end of
+ the f-string, expression will be NULL (unless result == 1,
+ see below). */
+ int result = fstring_find_literal_and_expr(str, ofs, recurse_lvl,
+ &literal, &expression,
+ c, n);
+ if (result < 0)
+ return -1;
+
+ /* Add the literal, if any. */
+ if (!literal) {
+ /* Do nothing. Just leave last_str alone (and possibly
+ NULL). */
+ } else if (!state->last_str) {
+ state->last_str = literal;
+ literal = NULL;
+ } else {
+ /* We have a literal, concatenate it. */
+ assert(PyUnicode_GET_LENGTH(literal) != 0);
+ if (FstringParser_ConcatAndDel(state, literal) < 0)
+ return -1;
+ literal = NULL;
+ }
+ assert(!state->last_str ||
+ PyUnicode_GET_LENGTH(state->last_str) != 0);
+
+ /* We've dealt with the literal now. It can't be leaked on further
+ errors. */
+ assert(literal == NULL);
+
+ /* See if we should just loop around to get the next literal
+ and expression, while ignoring the expression this
+ time. This is used for un-doubling braces, as an
+ optimization. */
+ if (result == 1)
+ continue;
+
+ if (!expression)
+ /* We're done with this f-string. */
+ break;
+
+ /* We know we have an expression. Convert any existing string
+ to a Str node. */
+ if (!state->last_str) {
+ /* Do nothing. No previous literal. */
+ } else {
+ /* Convert the existing last_str literal to a Str node. */
+ expr_ty str = make_str_node_and_del(&state->last_str, c, n);
+ if (!str || ExprList_Append(&state->expr_list, str) < 0)
+ return -1;
+ }
+
+ if (ExprList_Append(&state->expr_list, expression) < 0)
+ return -1;
+ }
+
+ assert(*ofs <= PyUnicode_GET_LENGTH(str));
+
+ /* If recurse_lvl is zero, then we must be at the end of the
+ string. Otherwise, we must be at a right brace. */
+
+ if (recurse_lvl == 0 && *ofs < PyUnicode_GET_LENGTH(str)) {
+ ast_error(c, n, "f-string: unexpected end of string");
+ return -1;
+ }
+ if (recurse_lvl != 0 && PyUnicode_READ_CHAR(str, *ofs) != '}') {
+ ast_error(c, n, "f-string: expecting '}'");
+ return -1;
+ }
+
+ FstringParser_check_invariants(state);
+ return 0;
+}
+
+/* Convert the partial state reflected in last_str and expr_list to an
+ expr_ty. The expr_ty can be a Str, or a JoinedStr. */
+static expr_ty
+FstringParser_Finish(FstringParser *state, struct compiling *c,
+ const node *n)
+{
+ asdl_seq *seq;
+
+ FstringParser_check_invariants(state);
+
+ /* If we're just a constant string with no expressions, return
+ that. */
+ if(state->expr_list.size == 0) {
+ if (!state->last_str) {
+ /* Create a zero length string. */
+ state->last_str = PyUnicode_FromStringAndSize(NULL, 0);
+ if (!state->last_str)
+ goto error;
+ }
+ return make_str_node_and_del(&state->last_str, c, n);
+ }
+
+ /* Create a Str node out of last_str, if needed. It will be the
+ last node in our expression list. */
+ if (state->last_str) {
+ expr_ty str = make_str_node_and_del(&state->last_str, c, n);
+ if (!str || ExprList_Append(&state->expr_list, str) < 0)
+ goto error;
+ }
+ /* This has already been freed. */
+ assert(state->last_str == NULL);
+
+ seq = ExprList_Finish(&state->expr_list, c->c_arena);
+ if (!seq)
+ goto error;
+
+ /* If there's only one expression, return it. Otherwise, we need
+ to join them together. */
+ if (seq->size == 1)
+ return seq->elements[0];
+
+ return JoinedStr(seq, LINENO(n), n->n_col_offset, c->c_arena);
+
+error:
+ FstringParser_Dealloc(state);
+ return NULL;
+}
+
+/* Given an f-string (with no 'f' or quotes) that's in str starting at
+ ofs, parse it into an expr_ty. Return NULL on error. Does not
+ decref str. */
+static expr_ty
+fstring_parse(PyObject *str, Py_ssize_t *ofs, int recurse_lvl,
+ struct compiling *c, const node *n)
+{
+ FstringParser state;
+
+ FstringParser_Init(&state);
+ if (FstringParser_ConcatFstring(&state, str, ofs, recurse_lvl,
+ c, n) < 0) {
+ FstringParser_Dealloc(&state);
+ return NULL;
+ }
+
+ return FstringParser_Finish(&state, c, n);
+}
+
+/* n is a Python string literal, including the bracketing quote
+ characters, and r, b, u, &/or f prefixes (if any), and embedded
+ escape sequences (if any). parsestr parses it, and returns the
+ decoded Python string object. If the string is an f-string, set
+ *fmode and return the unparsed string object.
+*/
static PyObject *
-parsestr(struct compiling *c, const node *n, int *bytesmode)
+parsestr(struct compiling *c, const node *n, int *bytesmode, int *fmode)
{
size_t len;
const char *s = STR(n);
@@ -4027,15 +4853,24 @@ parsestr(struct compiling *c, const node *n, int *bytesmode)
quote = *++s;
rawmode = 1;
}
+ else if (quote == 'f' || quote == 'F') {
+ quote = *++s;
+ *fmode = 1;
+ }
else {
break;
}
}
}
+ if (*fmode && *bytesmode) {
+ PyErr_BadInternalCall();
+ return NULL;
+ }
if (quote != '\'' && quote != '\"') {
PyErr_BadInternalCall();
return NULL;
}
+ /* Skip the leading quote char. */
s++;
len = strlen(s);
if (len > INT_MAX) {
@@ -4044,12 +4879,17 @@ parsestr(struct compiling *c, const node *n, int *bytesmode)
return NULL;
}
if (s[--len] != quote) {
+ /* Last quote char must match the first. */
PyErr_BadInternalCall();
return NULL;
}
if (len >= 4 && s[0] == quote && s[1] == quote) {
+ /* A triple quoted string. We've already skipped one quote at
+ the start and one at the end of the string. Now skip the
+ two at the start. */
s += 2;
len -= 2;
+ /* And check that the last two match. */
if (s[--len] != quote || s[--len] != quote) {
PyErr_BadInternalCall();
return NULL;
@@ -4088,51 +4928,84 @@ parsestr(struct compiling *c, const node *n, int *bytesmode)
}
}
return PyBytes_DecodeEscape(s, len, NULL, 1,
- need_encoding ? c->c_encoding : NULL);
+ need_encoding ? c->c_encoding : NULL);
}
-/* Build a Python string object out of a STRING+ atom. This takes care of
- * compile-time literal catenation, calling parsestr() on each piece, and
- * pasting the intermediate results together.
- */
-static PyObject *
-parsestrplus(struct compiling *c, const node *n, int *bytesmode)
+/* Accepts a STRING+ atom, and produces an expr_ty node. Run through
+ each STRING atom, and process it as needed. For bytes, just
+ concatenate them together, and the result will be a Bytes node. For
+ normal strings and f-strings, concatenate them together. The result
+ will be a Str node if there were no f-strings; a FormattedValue
+ node if there's just an f-string (with no leading or trailing
+ literals), or a JoinedStr node if there are multiple f-strings or
+ any literals involved. */
+static expr_ty
+parsestrplus(struct compiling *c, const node *n)
{
- PyObject *v;
+ int bytesmode = 0;
+ PyObject *bytes_str = NULL;
int i;
- REQ(CHILD(n, 0), STRING);
- v = parsestr(c, CHILD(n, 0), bytesmode);
- if (v != NULL) {
- /* String literal concatenation */
- for (i = 1; i < NCH(n); i++) {
- PyObject *s;
- int subbm = 0;
- s = parsestr(c, CHILD(n, i), &subbm);
- if (s == NULL)
- goto onError;
- if (*bytesmode != subbm) {
- ast_error(c, n, "cannot mix bytes and nonbytes literals");
- Py_DECREF(s);
- goto onError;
- }
- if (PyBytes_Check(v) && PyBytes_Check(s)) {
- PyBytes_ConcatAndDel(&v, s);
- if (v == NULL)
- goto onError;
- }
- else {
- PyObject *temp = PyUnicode_Concat(v, s);
- Py_DECREF(s);
- Py_DECREF(v);
- v = temp;
- if (v == NULL)
- goto onError;
+
+ FstringParser state;
+ FstringParser_Init(&state);
+
+ for (i = 0; i < NCH(n); i++) {
+ int this_bytesmode = 0;
+ int this_fmode = 0;
+ PyObject *s;
+
+ REQ(CHILD(n, i), STRING);
+ s = parsestr(c, CHILD(n, i), &this_bytesmode, &this_fmode);
+ if (!s)
+ goto error;
+
+ /* Check that we're not mixing bytes with unicode. */
+ if (i != 0 && bytesmode != this_bytesmode) {
+ ast_error(c, n, "cannot mix bytes and nonbytes literals");
+ Py_DECREF(s);
+ goto error;
+ }
+ bytesmode = this_bytesmode;
+
+ assert(bytesmode ? PyBytes_CheckExact(s) : PyUnicode_CheckExact(s));
+
+ if (bytesmode) {
+ /* For bytes, concat as we go. */
+ if (i == 0) {
+ /* First time, just remember this value. */
+ bytes_str = s;
+ } else {
+ PyBytes_ConcatAndDel(&bytes_str, s);
+ if (!bytes_str)
+ goto error;
}
+ } else if (this_fmode) {
+ /* This is an f-string. Concatenate and decref it. */
+ Py_ssize_t ofs = 0;
+ int result = FstringParser_ConcatFstring(&state, s, &ofs, 0, c, n);
+ Py_DECREF(s);
+ if (result < 0)
+ goto error;
+ } else {
+ /* This is a regular string. Concatenate it. */
+ if (FstringParser_ConcatAndDel(&state, s) < 0)
+ goto error;
}
}
- return v;
+ if (bytesmode) {
+ /* Just return the bytes object and we're done. */
+ if (PyArena_AddPyObject(c->c_arena, bytes_str) < 0)
+ goto error;
+ return Bytes(bytes_str, LINENO(n), n->n_col_offset, c->c_arena);
+ }
+
+ /* We're not a bytes string, bytes_str should never have been set. */
+ assert(bytes_str == NULL);
+
+ return FstringParser_Finish(&state, c, n);
- onError:
- Py_XDECREF(v);
+error:
+ Py_XDECREF(bytes_str);
+ FstringParser_Dealloc(&state);
return NULL;
}
diff --git a/Python/compile.c b/Python/compile.c
index a6884ec..3a49ece 100644
--- a/Python/compile.c
+++ b/Python/compile.c
@@ -731,6 +731,7 @@ compiler_set_qualname(struct compiler *c)
return 1;
}
+
/* Allocate a new block and return a pointer to it.
Returns NULL on error.
*/
@@ -3209,6 +3210,117 @@ compiler_call(struct compiler *c, expr_ty e)
e->v.Call.keywords);
}
+static int
+compiler_joined_str(struct compiler *c, expr_ty e)
+{
+ /* Concatenate parts of a string using ''.join(parts). There are
+ probably better ways of doing this.
+
+ This is used for constructs like "'x=' f'{42}'", which have to
+ be evaluated at compile time. */
+
+ static PyObject *empty_string;
+ static PyObject *join_string;
+
+ if (!empty_string) {
+ empty_string = PyUnicode_FromString("");
+ if (!empty_string)
+ return 0;
+ }
+ if (!join_string) {
+ join_string = PyUnicode_FromString("join");
+ if (!join_string)
+ return 0;
+ }
+
+ ADDOP_O(c, LOAD_CONST, empty_string, consts);
+ ADDOP_NAME(c, LOAD_ATTR, join_string, names);
+ VISIT_SEQ(c, expr, e->v.JoinedStr.values);
+ ADDOP_I(c, BUILD_LIST, asdl_seq_LEN(e->v.JoinedStr.values));
+ ADDOP_I(c, CALL_FUNCTION, 1);
+ return 1;
+}
+
+/* Note that this code uses the builtin functions format(), str(),
+ repr(), and ascii(). You can break this code, or make it do odd
+ things, by redefining those functions. */
+static int
+compiler_formatted_value(struct compiler *c, expr_ty e)
+{
+ PyObject *conversion_name = NULL;
+
+ static PyObject *format_string;
+ static PyObject *str_string;
+ static PyObject *repr_string;
+ static PyObject *ascii_string;
+
+ if (!format_string) {
+ format_string = PyUnicode_InternFromString("format");
+ if (!format_string)
+ return 0;
+ }
+
+ if (!str_string) {
+ str_string = PyUnicode_InternFromString("str");
+ if (!str_string)
+ return 0;
+ }
+
+ if (!repr_string) {
+ repr_string = PyUnicode_InternFromString("repr");
+ if (!repr_string)
+ return 0;
+ }
+ if (!ascii_string) {
+ ascii_string = PyUnicode_InternFromString("ascii");
+ if (!ascii_string)
+ return 0;
+ }
+
+ ADDOP_NAME(c, LOAD_GLOBAL, format_string, names);
+
+ /* If needed, convert via str, repr, or ascii. */
+ if (e->v.FormattedValue.conversion != -1) {
+ switch (e->v.FormattedValue.conversion) {
+ case 's':
+ conversion_name = str_string;
+ break;
+ case 'r':
+ conversion_name = repr_string;
+ break;
+ case 'a':
+ conversion_name = ascii_string;
+ break;
+ default:
+ PyErr_SetString(PyExc_SystemError,
+ "Unrecognized conversion character");
+ return 0;
+ }
+ ADDOP_NAME(c, LOAD_GLOBAL, conversion_name, names);
+ }
+
+ /* Evaluate the value. */
+ VISIT(c, expr, e->v.FormattedValue.value);
+
+ /* If needed, convert via str, repr, or ascii. */
+ if (conversion_name) {
+ /* Call the function we previously pushed. */
+ ADDOP_I(c, CALL_FUNCTION, 1);
+ }
+
+ /* If we have a format spec, use format(value, format_spec). Otherwise,
+ use the single argument form. */
+ if (e->v.FormattedValue.format_spec) {
+ VISIT(c, expr, e->v.FormattedValue.format_spec);
+ ADDOP_I(c, CALL_FUNCTION, 2);
+ } else {
+ /* No format spec specified, call format(value). */
+ ADDOP_I(c, CALL_FUNCTION, 1);
+ }
+
+ return 1;
+}
+
/* shared code between compiler_call and compiler_class */
static int
compiler_call_helper(struct compiler *c,
@@ -3878,6 +3990,10 @@ compiler_visit_expr(struct compiler *c, expr_ty e)
case Str_kind:
ADDOP_O(c, LOAD_CONST, e->v.Str.s, consts);
break;
+ case JoinedStr_kind:
+ return compiler_joined_str(c, e);
+ case FormattedValue_kind:
+ return compiler_formatted_value(c, e);
case Bytes_kind:
ADDOP_O(c, LOAD_CONST, e->v.Bytes.s, consts);
break;
@@ -4784,4 +4900,3 @@ PyAST_Compile(mod_ty mod, const char *filename, PyCompilerFlags *flags,
{
return PyAST_CompileEx(mod, filename, flags, -1, arena);
}
-
diff --git a/Python/symtable.c b/Python/symtable.c
index 64910d8..8431d51 100644
--- a/Python/symtable.c
+++ b/Python/symtable.c
@@ -1439,6 +1439,14 @@ symtable_visit_expr(struct symtable *st, expr_ty e)
VISIT_SEQ(st, expr, e->v.Call.args);
VISIT_SEQ_WITH_NULL(st, keyword, e->v.Call.keywords);
break;
+ case FormattedValue_kind:
+ VISIT(st, expr, e->v.FormattedValue.value);
+ if (e->v.FormattedValue.format_spec)
+ VISIT(st, expr, e->v.FormattedValue.format_spec);
+ break;
+ case JoinedStr_kind:
+ VISIT_SEQ(st, expr, e->v.JoinedStr.values);
+ break;
case Num_kind:
case Str_kind:
case Bytes_kind: