Crimes with Python's Pattern Matching (2022)

https://www.hillelwayne.com/post/python-abc/

by agluszak • 11 months ago

255 113 11 months ago

quotemstr6• 11 months ago

I've never understood why Python's pattern-matching isn't more general.

First, "case foo.bar" is a value match, but "case foo" is a name capture. Python could have defined "case .foo" to mean "look up foo as a variable the normal way" with zero ambiguity, but chose not to.

Second, there's no need to special-case some builtin types as matching whole values. You can write "case float(m): print(m)" and print the float that matched, but you can't write "case MyObject(obj): print(obj)" and print your object. Python could allow "..." or "None" or something in __match_args__ to mean "the whole object", but didn't.

Aefiam• 11 months ago

case .foo is explicitly mentioned in https://peps.python.org/pep-0622/ :

While potentially useful, it introduces strange-looking new syntax without making the pattern syntax any more expressive. Indeed, named constants can be made to work with the existing rules by converting them to Enum types, or enclosing them in their own namespace (considered by the authors to be one honking great idea)[...] If needed, the leading-dot rule (or a similar variant) could be added back later with no backward-compatibility issues.

second: you can use case MyObject() as obj: print(obj)

zahlman• 11 months ago

I don't think I've written a match-case yet. Aside from not having a lot of use cases for it personally, I find that it's very strange-feeling syntax. It tries too hard to look right, with the consequence that it's sometimes quite hard to reason about.

quotemstr6• 11 months ago

> While potentially useful, it introduces strange-looking new syntax without making the pattern syntax any more expressive. Indeed, named constants can be made to work with the existing rules by converting them to Enum types, or enclosing them in their own namespace (considered by the authors to be one honking great idea)[...]

Yeah, and I don't buy that for a microsecond.

A leading dot is not "strange" syntax: it mirrors relative imports. There's no workaround because it lets you use variables the same way you use them in any other part of the language. Having to distort your program by adding namespaces that exist only to work around an artificial pattern matching limitation is a bug, not a feature.

Also, it takes a lot of chutzpah for this PEP author to call a leading dot strange when his match/case introduces something that looks lexically like constructor invocation but is anything but.

The "as" thing works with primitive too, so why do we need int(m)? Either get rid of the syntax or make it general. Don't hard-code support for half a dozen stdlib types for some reason and make it impossible for user code to do the equivalent.

The Python pattern matching API is full of most stdlib antipatterns:

* It's irregular: matching prohibits things that the shape of the feature would suggest are possible because the PEP authors couldn't personally see a specific use case for those things. (What's the deal with prohibiting multiple _ but allowing as many __ as you want?)

* It privileges stdlib, as I mentioned above. Language features should not grant the standard library powers it doesn't extend to user code.

* The syntax feels bolted on. I get trying to reduce parser complexity and tool breakage by making pattern matching look like object construction, but it isn't, and the false cognate thing confuses every single person who tries to read a Python program. They could have used := or some other new syntax, but didn't, probably because of the need to build "consensus"

* The whole damn thing should have been an expression, like the if/then/else ternary, not a statement useless outside many lexical contexts in which one might want to make a decision. Why is it a statement? Probably because the PEP author didn't _personally_ have a need to pattern match in expression context.

And look: you can justify any of these technical decisions. You can a way to justify anything you might want to do. The end result, however, is a language facility that feels more cumbersome than it should and is applicable to fewer places than one might think.

Here's how to do it right: https://www.gnu.org/software/emacs/manual/html_node/elisp/pc...

If needed, the leading-dot rule (or a similar variant) could be added back later with no backward-compatibility issues.

So what, after another decade of debate, consensus, and compromise, we'll end up with a .-prefix-rule but one that works only if the character after the dot is a lowercase letter that isn't a vowel.

PEP: "We decided not to do this because inspection of real-life potential use cases showed that in vast majority of cases destructuring is related to an if condition. Also many of those are grouped in a series of exclusive choices."

I find this philosophical stance off-putting. It's a good thing when users find ways to use your tools in ways you didn't imagine.

PEP: In most other languages pattern matching is represented by an expression, not statement. But making it an expression would be inconsistent with other syntactic choices in Python. All decision making logic is expressed almost exclusively in statements, so we decided to not deviate from this.

We've had conditional expressions for a long time.

rpcope12• 11 months ago

After doing Erlang and Scala pattern matching, the whole Python implementation just feels really ugly and gross. They should have cribbed a lot more of how Scala does it.

orbisvicis• 11 months ago

I've given up on matching as I'm tired of running into its limitations.

That said, I don't think OP's antics are a crime. That SyntaxError though, that might be a crime.

And a class-generating callable class would get around Python caching the results of __subclasshook__.

hwayne• 11 months ago

Now I'm mad I didn't remember the word "antics". It's so much more evocative than "crimes"!

vlade11115• 11 months ago

While the article is very entertaining, I'm not a fan of the pattern matching in Python. I wish for some linter rule that can forbid the usage of pattern matching.

siddboots• 11 months ago

Can you explain why? Genuinely curious as a lover of case/match. My only complaint is that it is not general enough.

kurtis_reed2• 11 months ago

Double indentation

smcl• 11 months ago

If you're experienced enough with Python to say "I want to eliminate pattern matching from my codebase" you can surely construct that as a pre-commit check, no?

jbmchuck• 11 months ago

Should be easily doable with a semgrep rule, e.g.:

    ~> cat semgrep.yaml
    rules:
      - id: no-pattern-matching
        pattern: |
          match ...:
        message: |
          I'm not a fan of the pattern matching in Python
        severity: ERROR
        languages:
          - python

...

    ~> cat test.py
    #!/usr/bin/env python3

    foo = 1
    match foo:
      case 1:
        print("one")

...

    ~> semgrep --config semgrep.yaml test.py   


     no-pattern-matching
          I'm not a fan of the pattern matching in Python
                                                         
            4┆ match foo:
            5┆   case 1:
            6┆     print("one")

(exits non-0)

PokestarFan2• 11 months ago

You need to make that exclude match = ... since match can also be a variable name. This is because people used to write code like match = re.search(...)

purplehat_• 11 months ago

Could someone explain just what's so bad about this?

My best guess is that it adds complexity and makes code harder to read in a goto-style way where you can't reason locally about local things, but it feels like the author has a much more negative view ("crimes", "god no", "dark beating heart", the elmo gif).

gnulinux• 11 months ago

Side effects

taeric• 11 months ago

I took the memes as largely for comedic effect, only?

I do think there is a ton of indirection going on in the code that I would not immediately think to look for. As the post stated, could be a good reason for this in some things. But it would be the opposite of aiming for boring code, at that point.

xg15• 11 months ago

Maybe I have too much of a "strongly typed language" view here, but I understood the utility of isinstance() as verifying that an object is, well, an instance of that class - so that subsequent code can safely interact with that object, call class-specific methods, rely on class-specific invariants, etc.

This also makes life directly easier for me as a programmer, because I know in what code files I have to look to understand the behavior of that object.

Even linters use it to that purpose, e.g. resolving call sites by looking at the last isinstance() statement to determine the type.

__subclasshook__ puts this at risk by letting a class lie about its instances.

As an example, consider this class:

  class Everything(ABC):

    @classmethod
    def __subclasshook__(cls, C):
      return True

    def foo(self):
      ...

You can now write code like this:

  if isinstance(x, Everything):
    x.foo()

A linter would pass this code without warnings, because it assumes that the if block is only entered if x is in fact an instance of Everything and therefore has the foo() method.

But what really happens is that the block is entered for any kind of object, and objects that don't happen to have a foo() method will throw an exception.

cauthon2• 11 months ago

You _can_ write pathological code like the Everything example, but I can see this feature being helpful if used responsibly.

It essentially allows the user to check if a class implements an interface, without explicitly inheriting ABC or Protocol. It’s up to the user to ensure the body of the case doesn’t reference any methods or attributes not guaranteed by the subclass hook, but that’s not necessarily bad, just less safe.

All things have a place and time.

danudey3• 11 months ago

TL;DR having a class that determines if some other class is a subclass of itself based off of arbitrary logic and then using that arbitrary logic to categorize other people's arbitrary classes at runtime is sociopathic.

Some of these examples are similar in effect to what you might do in other languages, where you define an 'interface' and then you check to see if this class follows that interface. For example, you could define an interface DistancePoint which has the fields x and y and a distance() method, and then say "If this object implements this interface, then go ahead and do X".

Other examples, though, are more along the lines of if you implemented an interface but instead of the interface constraints being 'this class has this method' the interface constraints are 'today is Tuesday'. That's an asinine concept, which is what makes this crimes and also hilarious.

Spivak3• 11 months ago

You better not find out about Protocols in Python then. The behavior you describe is exactly how duck typing / "structural subtyping" works. Your class will be an instance of Iterable if you implement the right methods having never known the Iterable class exists.

I don't find using __subclasshook__ to implement structural subtyping that you can't express with Protocols/ABCs alone to be that much of a crime. You can do evil with it but I can perform evil with any language feature.

pansa24• 11 months ago

The real crime is the design of Python's pattern matching in the first place:

    match status:
        case 404:
            return "Not found"

    not_found = 404
    match status:
        case not_found:
            return "Not found"

Everywhere else in the language, you can give a constant a name without changing the code's behaviour. But in this case, the two snippets are very different: the first checks for equality (`status == 404`) and the second performs an assignment (`not_found = status`).

https://x.com/brandon_rhodes/status/1360226108399099909

quotemstr6• 11 months ago

And there was a much better proposal that got rejected in favor of what we got: https://peps.python.org/pep-0642/

danudey3• 11 months ago

The very first example there shows a match/case block where almost every single case just runs "pass" and yet every single one has a side effect. It's very difficult to read at first, difficult to understand if you're new to the syntax, and is built entirely around side effects. This might be one of the worst PEPs I've ever seen just based on that example alone.

Fun fact: you can do the same thing with the current match/case, except that you have to put your logic in the body of the case so that it's obvious what's happening.

almostgotcaught5• 11 months ago

because it's not a `switch` it's a `match` ie pattern matching...

pansa24• 11 months ago

Doesn't matter what it is, it shouldn't break fundamental rules of the language.

Ruby's `case`/`in` has the same problem.

user453• 11 months ago

That's destructuring, a feature not a bug. Same as it works in any functional language - and tremendously useful once you get the hang of it

pansa24• 11 months ago

At least functional languages tend to have block scope, so the latter snippet introduces a new variable that shadows `not_found` instead of mutating it.

quotemstr6• 11 months ago

Destructuring is a feature. Making it easy to confuse value capture and value reference was an error. Add single-namespace locals and you have a calamity.

mcdeltat• 11 months ago

Destructuring yes but you can still argue it's poorly designed. Particularly unintuitive because matching on a nested name e.g. module.CONSTANT works for matching and doesn't destructure. It's just the use of an unnested name which does destructuring.

What Python needs is what Elixir has. A "pin" operator that forces the variable to be used as its value for matching, rather than destructuring.

b3orn• 11 months ago

No, at least in Erlang a variable is assigned once, you can then match against that variable as it can't be reassigned:

    NotFound = 404,
    case Status of
        NotFound -> "Not Found";
        _ -> "Other Status"
    end.

That snippet will return "Other Status" for Status = 400. The Python equivalent of that snippet is a SyntaxError as the first case is a catch all and the rest is unreachable.

xdennis• 11 months ago

It wouldn't be a problem is Python had block level variable scope. Having that destructuring be limited to the 'case' would be fine, but obliterating the variable outside of the match is very unexpected.

Y_Y• 11 months ago

Barely a misdemeanor, all of the typechecks were deterministic

danudey3• 11 months ago

Next step is to have the subclass check pack all the code up, send it to ChatGPT, ask it if it thinks subjectively that class A should be a subclass of class B, and then run sentiment analysis on the resulting text to make the determination.

charlieyu1• 11 months ago

More and more dubious things were designed in Python these days. A recent PEP purposes to use {/} as the empty set

umgefahren• 11 months ago

Idk that doesn’t sound so dubious to me. ∅ might be more approachable for the PHDs then set() ;)

rtrgrd• 11 months ago

we all love non ascii code (cough emoji variable names)

augusto-moura3• 11 months ago

Problem is, we already have a syntax for empty lists [], empty tuples (), and {} is taken for an empty dict. So having a syntax for an empty set actually makes sense to me

kurtis_reed2• 11 months ago

{:} should have been the empty dict, now there's no good solution

rand_r2• 11 months ago

You can use “set()”. Introducing more weird special cases into the language is a bad direction for Python.

slightwinder• 11 months ago

Making sense, and being good, is not necessary the same.

Yes, having a solution for this makes sense, but the proposed solutions are just not good. Sometimes one has to admit that not everything can be solved gracefully and just stop, hunting the whale.

btown3• 11 months ago

My favorite Python "crime" is that a class that defines __rrshift__, instantiated and used as a right-hand-side, lets you have a pipe operator, regardless of the left-hand-side (as long as it doesn't define __rshift__).

It's reasonably type-safe, and there's no need to "close" your chain - every outputted value as you write the chain can have a primitive type.

    x = Model.objects.all() >> by_pk >> call(dict.values) >> to_list >> get('name') >> _map(str.title) >> log_debug >> to_json

It shines in notebooks and live coding, where you might want to type stream-of-thought in the same order of operations that you want to take place. Need to log where something might be going wrong? Tee it like you're on a command line!

Idiomatic? Absolutely not. Something to push to production? Not unless you like being stabbed with pitchforks. Actually useful for prototyping? 1000%.

divbzero• 11 months ago

I suppose you could use the same trick with __ror__ and | (as long as the left-hand side doesn’t define __or__).

  x = Model.objects.all() | by_pk | call(dict.values) | to_list | get('name') | _map(str.title) | log_debug | to_json

btown3• 11 months ago

Sadly many things define the __or__ operator, including dicts and sets which are common to find in pipelines. (https://peps.python.org/pep-0584/ was a happy day for everyone but me!)

In practice, rshift gives a lot more flexibility! And you’d rarely chain after a numeric value.

ck452• 11 months ago

Or overloading division like https://scapy.net/

wredcoll• 11 months ago

Oh god, it's c++ all over again!

quotemstr6• 11 months ago

Is that supposed to be a bad thing?

BitsAndObjects• 11 months ago

I spy a functional programmer lurking in this abuse of Python ;)

Looks a lot like function composition with the arguments flipped, which in Haskell is `>>>`. Neat!

But since you’re writing imperative code and binding the result to a variable, you could also compare to `>>=`.

(https://downloads.haskell.org/~ghc/7.6.2/docs/html/libraries...)

btown3• 11 months ago

Having spent a lot of time lurking on the frustratingly-slow-moving bikeshedding thread for the Javascript pipe operator [0], there's a great irony that a lot of people want a pipe operator because they don't want to deal with function composition in any way, other than just applying a series of operations to their data!

I think there's a big gap pedagogically here. Once a person understands functional programming, these kinds of composition shorthands make for very straightforward and intuitive code.

But, if you're just understanding basic Haskell/Clojure syntax, or stuck in the rabbit hole of "monad is a monoid" style introductions, a beginner could easily start to think: "This functional stuff is really making me need to think in reverse, to need to know my full pipeline before I type a single character, and even perhaps to need to write Lisp-like (g (f x)) style constructs that are quite the opposite of the way my data is flowing."

I'm quite partial to tutorials like Railway Oriented Programming [1] which start from a practical imperative problem, embrace the idea that data and code should feel like they flow in the same direction, and gradually guide the reader to understanding the power of the functional tools they can bring to bear!

If anything, I hope this hack sparks good conversations :)

[0] https://github.com/tc39/proposal-pipeline-operator/issues/91 - 6 years and 793 comments!

[1] https://fsharpforfunandprofit.com/rop/

ck452• 11 months ago

Apache beam does it: https://beam.apache.org/documentation/sdks/python-streaming/

I wanted to wash my eyes the first time I saw it.

kragen2• 11 months ago

Wow, this is awesome! Your example is especially fantastic.

kragen2• 11 months ago

Witness the chaos you have unloosed upon the world; not even little Lua is safe:

  > =dd-{"Todo","es","bueno"} /call("gsub", "[aeiou]", "i") /call("rep", 2) /call "lower"                  
  {"tiditidi", "isis", "biinibiini"}
  > ps={{x=5, y=7}, {x=5, y=4}, {x=3, y=4}}
  > =dd-ps/get"x"                                       
  {5, 5, 3}
  > =dd-ps/where {y=4}
  {[2] = {y=4, x=5}, [3] = {y=4, x=3}}                                       
  > =dd-ps/group "x"                                         
  {[5] = {{x=5, y=7}, {x=5, y=4}}, [3] = {{x=3, y=4}}}       
  > =dd-ps/group "x"/get(1)                                  
  {[5] = {x=5, y=7}, [3] = {x=3, y=4}}                       
  > =dd-ps/group "x"/map(get"y")                             
  {[5] = {7, 4}, [3] = {4}                                   
  > =dd-ps/group "x"/map(get"y")/map(max)                    
  {[5] = 7, [3] = 4}                                         
  > =dd-ps/group "x"/map(get"y")/map(max)/min                
  4
  > =dd-ps/group "x"/map(get"y")/map(max)/sum                
  11
  > =dd-ps/where(function(p) return 9 < p.x+p.y end)
  {[2] = {y=6, x=8}}                                         
  > function bytes(s) return {s:byte(1,#s)} end
  > =dd-{"hey", "you"}/map(bytes)
  {{104, 101, 121}, {121, 111, 117}}
  > =dd-{"hey", "you"}/map(bytes)/flat/freq
  {[111] = 1, [101] = 1, [104] = 1, [121] = 2, [117] = 1}

thayne4• 11 months ago

I don't think this is necessarily a "crime", I could see this being useful.

An example that uses a similar approach would be if you had a metaclass that overrode __instancecheck__ to return true if a string matched a regular expression. Then you could create (dynamically defined) classes that used that metaclass to use in match statements to march a string against multiple regexes. Unfortunately, I can't think of a good way to extract any capture groups.

VagabundoP• 11 months ago

ok, took me a bit to wrap my head around whats going on here, but actually this is neat, and probably useful.

Ive been hobbying in python for a decade plus at this point and it still ties my brain in knots sometimes with how it works at runtime. I do enjoy working with it though.

E: just to clarify if you're going to use this don't bury it deep in a multi file code structure, explicitly have it in the same file as you use it, otherwise people will get confused.

M0r13n• 11 months ago

Personally, I have never liked the PEP 634 pattern matching. I write a lot of code in Python. 99% of the time when I could use pattern matching, I am going to use simple if statements or dictionaries. Most of the time, they are more straightforward and easier to read, especially for developers who are more familiar with traditional control flow.

NeutralForest2• 11 months ago

You should use if statements if that's what you need. The match statement is for structural pattern matching mostly.

Q6T46nT668w6i3m• 11 months ago

What’s the problem using it as a switch statement if you care about typographic issues? I do this so I’d like to know if I missed something and this is a bad practice.

Balinares• 11 months ago

Dictionaries with a limited key and value type definition are fine, but dictionaries as a blind storage type are a recipe for crashing in prod with type errors or key errors. Structural pattern matching exists to support type safety.

I'll argue that code is in fact not easy to read if reading it doesn't tell you what type an item is and what a given line of code using it even does at runtime.

odyssey7• 11 months ago

Having user other languages after finishing university studies, Python does not spark joy.

JohnKemeny2• 11 months ago

Discussed 3 years ago:

Crimes with Python's pattern matching

406 points on Aug 2, 2022. 120 comments

https://news.ycombinator.com/item?id=32314368

TeeMassive• 11 months ago

I've been doing Python full time for about 4 years now. I always found that when pattern matching is used it always ends up increasing the cognitive load significantly. It shines for simple cases, and for precise object pattern sniping. But anything that acts on nested objects or more than 3 members with multiple case statements is pre-programmed for failure. What makes it worse, is that there seems to be a trend where other Python adepts will default to pattern matching over everything else, like a too good of a hammer.

Crimes with Python's Pattern Matching (2022)

Related Stories