a == ba is bTip== calls __eq__, so each class decides what equal means. Use is only for singletons, as in value is None.
TrapCPython reuses small integers, so x is y can be true by accident. Never use is to compare numbers or strings.
Asked in interviews
AskedWhat is the difference betweenis and == with examples?list, dict, set and most objects of your own classes.str, tuple, frozenset and bytes.items.append(4)items = [0]Only the local name moves.TipPython passes by assignment, with no copy. Immutable arguments can only be rebound, so they behave as if passed by value.
TrapA function that changes a mutable argument also changes the caller's object. To protect it, pass items.copy().
copy.copy(a)a.copy()a[:]copy.deepcopy(a) Tipdeepcopy is slower and uses more memory. It also handles objects that refer to themselves.
TrapAfter a shallow copy, changing shallow[0][0] changes the original too. The difference shows only with nested, mutable objects.
Asked in interviews
AskedWhat is the difference between shallow copy and deep copy?global nnonlocal nTrapAssign to a name anywhere in a function and it's local for the whole body. Reading it first raises Unbound.
Asked in interviews
AskedWhat is the difference between local and global variables?int, float, complex, str, list, tuple, set, frozenset, dict, bool, NoneType.TipPython is dynamically typed. A variable has no fixed type, and type(value) tells you the type of the value it holds.
Asked in interviews
AskedExplain Python's built-in data types.[1, 2, 3](1, 2)TipA tuple is a little faster and smaller, because Python allocates it once at a fixed size.
TrapA list can't be a dict key or a set member. A tuple can, when all its items are hashable.
Asked in interviews
AskedWhat is the difference between list and tuple?d[key]d.get(key, default)TipA lookup takes O(1) on average. d[key] raises KeyError on a missing key. Since Python 3.7, a dict keeps insertion order.
TrapKeys must be hashable. Strings, numbers and tuples of hashable items work, but lists don't.
Asked in interviews
AskedExplain Python dictionaries and their characteristics.[x*x for x in a if x]{x for x in a}{k: v for k, v in a}(x for x in a)TrapParentheses build a generator expression, not a tuple.
Asked in interviews
AskedExplain list comprehensions in Python.sorted(a)a.sort()map(f, a)filter(f, a)zip(a, b)isinstance(x, int)hasattr(x, "y")Trapmap, filter, enumerate and zip return lazy iterators, not lists. Wrap one in list() to see its values.
Asked in interviews
AskedWhat are Python's built-in functions you should know?snake_caseCapWordsUPPER_CASE_ first, then letters, digits and _. Case matters, and it can't be a keyword.TipPEP 8 also limits lines to 79 characters. It orders imports as standard library, third-party, then your own modules.
TrapA wrong indent moves a line into another block or stops the file from running. Python raises Indentation, or TabError for mixed tabs and spaces.
if a:elif b:else:x if ok else ymatch value:for x in a:while n < 3:enumerate(a)breakcontinuepassTipA condition doesn't need to be a Boolean. Python treats 0, empty collections and None as false.
Trapbreak exits only the innermost loop. It also skips the loop's else clause, which runs only when no break happened.
print "hello"Python 3 uses print("hello").5 / 2It gave 2. Python 3 gives 2.5, and // does floor division.raw_input()Python 3 uses input(), which returns a string.xrange(n)Python 3's range() works like it.TipPython 2 reached end of life on January 1, 2020. Python 3 strings are Unicode by default, with a separate bytes type.
TrapA range is a lazy sequence, not an iterator. It supports len(), indexing and in, and you can loop over it many times.
f(1, 2)f(a=1)TipA function with no return gives back None.
Trapitems=[] creates one list when Python defines the function. Every call that relies on the default shares it.
def f(a, /):def f(*, timeout):def f(*args, **kwargs):f(*nums)f(**opts)TipPython reads the stars, not the names, so *values works too. / lets authors rename a parameter later without breaking callers.
lambda n=x: nkey=lambda s: s[1]TipA lambda is one expression, with no loops or assignments. A def gets a real name in tracebacks.
TrapClosures use late binding, so they read a variable when they run. Lambdas made in a loop all see its last value.
@loggedShort for f = logged(f).TrapWithout @functools., the wrapper hides the original function's name and docstring.
Asked in interviews
AskedWhat are decorators and how do they work?iter(nums)next(it)next raises StopIteration at the end.__iter____next__TipEvery iterator is iterable, but not every iterable is an iterator. A for loop calls iter() once, then next().
TrapAn iterator is single-use. A list can loop many times, but a generator or a file object is empty after one pass.
Asked in interviews
AskedWhat is the difference between an iterable and an iterator?yield aIt keeps its local variables.gen = fib(10)The body doesn't run yet.TipOnly one value exists at a time, so a generator walks a huge file or an endless stream in constant memory.
TrapA generator is single-pass. Once consumed it's empty, and it has no len() or indexing.
Asked in interviews
AskedExplain generators and theyield keyword.__init__(self, name)p = Person("Alice")school = "ABC"Write it in the class body.self.name = nameTipself is the object the method was called on. Python passes it in for you.
Traps.school = "x" hides the class variable for s only. A list as a class variable is shared by every instance.
def describe(self):@classmethod@staticmethodTipA class method is often an alternative constructor, like dict.fromkeys(). On a subclass, it builds that subclass.
class Dog(Animal):Redefine a method to override it.super().speak()Animal.speak(self) def add(self, *args):Python has no overloading. A later def replaces the earlier one.TipUse inheritance for a true "is a" relationship. Otherwise prefer composition, where a class holds another.
Trapsuper() calls the next class in the MRO, Dog.__mro__, not simply the parent. In a diamond, a direct parent call can run a base class twice.
self._balancePython doesn't enforce it.self.__balancePython renames it _Account__balance.@property@x.setterTrapPython has no private keyword. Name mangling isn't real privacy, because a. still works.
Asked in interviews
AskedExplain encapsulation and access modifiers.+ and ==__add____eq____str____repr__len() and for loops__len____iter__TrapDefining __eq__ without __hash__ makes instances unhashable.
Asked in interviews
AskedWhat are Python's special or magic methods?@dataclassfield(default_factory=list) frozen=Trueorder=Trueslots=TrueTip@dataclass writes __init__, __repr__ and __eq__. __post_init__ runs after it, so validate there.
TrapFields with defaults must come after fields without them.
Asked in interviews
AskedWhat are dataclasses and how do they work?class A(metaclass=M):The class of a class. The default is type.__init_subclass__ Or use a class decorator.Calc.add = loggedmock.patchpatch is for tests and restores the original.TrapMonkey patching is risky in production. The change is global, invisible at the call site, and can break when the library changes.
except ValueError as e:Handle an errorelse:Run the success pathfinally:Clean up in every caseTipfinally runs even after return. An exception no except matches triggers it, then keeps propagating.
TrapPython checks except blocks in order, so put the most specific first. A broad except Exception can hide bugs.
class E(Exception):raise E("bad age")raise E("x") from errassert n > 0, "msg"TipIn a larger project, derive every error from one base exception, so callers can catch them all with one except.
Trappython -O removes every assert. Never use one to validate input or check permissions. Raise ValueError instead.
with open(p) as f:__enter____exit__@contextmanagerTipThe teardown runs even if the block raises. If __exit__ returns a true value, Python suppresses the exception.
Asked in interviews
AskedExplain context managers and thewith statement.self.assertEqual(a, 5)with self.assertRaises(E): python -m unittestpytestunittest.mockbreakpoint()pdb.set_trace()logger.debug("x") TipRead the full traceback first. It names the exception and the line. Prints give a quick look, and logging suits production.
.py file. A package is a directory of modules.import mathimport numpy as npfrom math import sqrtfrom .mod import fnimport * brings in__all__ = ["fn"]Tip__init__.py runs on first import, so keep it light. A directory without one is a namespace package.
TrapAvoid from math import *. It hides where each name came from.
Tippython mymodule.py sets __name__ to "__main__". An import sets it to the module's name, so the block is skipped.
Asked in interviews
AskedExplain Python's__name__ == "__main__" idiom.open(p, "r")open(p, "w")open(p, "a")open(p, "x")TipAdd b for bytes, or + to read and write, as in "r+". Open files in a with block so Python closes them.
Trap"w" truncates an existing file to empty. Use "x" to protect existing data.
Asked in interviews
AskedExplain Python's file handling modes.python -m venv .venvsource .venv/bin/activate .venv\Scripts\activatepip freeze > requirements.txt pip install -r requirements.txt TipTreat it as disposable. Don't commit it or copy it to another machine. Recreate it from the requirements file.
Asked in interviews
AskedWhat are Python virtual environments and why are they important?dis.dis(func)TipPyPy instead uses a just-in-time (JIT) compiler.
TrapA .pyc file in __pycache__ only makes imports load faster, not run faster. The script you run directly compiles again each time.
Asked in interviews
AskedWhat happens when you run a Python program?del aAt zero references, CPython frees the object at once.gc.collect()The collector also runs by itself.__dict____slots__ = ["x", "y"]You then can't add other attributes.sys.intern(s)-5 to 256 up to Python 3.14, and -5 to 1024 from Python 3.15.Trapdel removes a name, not the object. Objects in a cycle keep each other's count above zero until the collector runs.
python -m cProfile app.pyx in my_setA set takes O(1), and a list takes O(n).@functools.lru_cache TipMeasure first, then fix only the hot spots. Built-ins and NumPy run in compiled code.
Asked in interviews
AskedHow can you improve Python performance?CPU-bound work. Threads take turns holding the GIL. Each process has its own GIL, so processes run at once and finish sooner.
ThreadPoolExecutor(4) A waiting thread releases the GIL.ProcessPoolExecutor(4) It costs more memory, and data between processes gets copied.TrapThreads don't speed up CPU-bound Python code, because only one runs at a time. The free-threaded build runs single-threaded code somewhat slower.
One thread. Solid bars run on the event loop one at a time, and dashed bars wait on I/O. The waits overlap, so the total is close to the slowest.
async def f():await f()asyncio.run(main()) asyncio.gather(a(), b())asyncio.TaskGroup() asyncio.to_thread(f) TipCalling a coroutine function only returns a coroutine object. Nothing runs until you await or schedule it.
TrapA coroutine gives up control only at an await. A blocking call or a long CPU loop freezes every other task.
Asked in interviews
AskedExplain Python's async/await and asynchronous programming.Nothing matches. Try a shorter word, or clear the search.
With Pro, you can:
Already have Pro? Sign in