4 ms·

I'm a bit surprised nobody has tried to start up the "what are tuples really for in Python" argument in this thread. The definition given -- "Python tuples act

I'm a bit surprised nobody has tried to start up the "what are tuples really for in Python" argument in this thread. The definition given -- "Python tuples act like immutable lists" -- is correct as a description of behavior, but the decision of when and whether to use a tuple versus a list or some other structure can be a pretty contentious one in some Python circles.

cup-of-tea 9y ago
Lists are for iterating, tuples are for unpacking.
- kzrdude 9y ago
  Wouldn't you say that lists are for homogenous collections and tuples for short heterogeneous ones? Of course we use the Python "duck typing" definition for homogenous.
  - cup-of-tea 9y ago
    You could, but I think mine is better. (firstname, lastname) and (x,y,z) are homogeneous collections. But they probably should be tuples. In a tuple the position describes the structure, in a list the position merely describes the order. Hence why tuples are best if you plan to unpack them (ie. firstname, lastname = name). Trouble is Python doesn't completely separate the concepts like some languages do (e.g. C array/struct, R array/list). You can unpack lists and store heterogeneous items in them. Many libraries do this (e.g. SQL libraries will give a list per row by default). And if you follow the bash read idiom in Python using str.split you'll unpack a list as well. For me the distinction is about documenting my code. The only real practical advantage I'm aware of is tuples are hashable. I don't think they are implemented anything like structs underneath.
    - lmm 9y ago
      > (firstname, lastname) and (x,y,z) are homogeneous collections. Disagree. In a language with a decent type system you would give them different types, because using a first name as a last name or an x as a z is always an error.
      - shoo 9y ago
        Yeah. Another rule of thumb is that a lot of the time you can make code that uses tuples clearer by rewriting it to use namedtuple-defined struct values. If the language supported it, it'd probably also make sense most of the time to define a type for each tuple field, and have a type-checker look for errors. Occasionally you see tuples used where they want immutable and hashable lists, but most of the time they're used as shorthand struct/"pod" tours types.
      - cup-of-tea 9y ago
        In what language would you make firstname and lastname or x and y different types?
        lmm 9y ago
        Well, I would in Rust for starters.
        cup-of-tea 9y ago
        Interesting. I should probably try one of these typed languages. When I was thinking of types I was thinking about "physical" types like in C, not abstract types.
        dllthomas 9y ago
        Yeah, it's pretty common to think of types as "representational", and that really doesn't get at most of what they are good for. Even in C you can get some good help from the type checker if you know how to use it. For instance, see https://dlthomas.github.io/using-c-types-talk/slides/slides.html#26.0 https://dlthomas.github.io/using-c-types-talk/slides/slides.... for a technique that can turn a flakey segfault from misuse of an API into a compile time error (later slides do just that for a simple SDL program, with a modified header file).

cup-of-tea 9y ago

Lists are for iterating, tuples are for unpacking.

kzrdude 9y ago
Wouldn't you say that lists are for homogenous collections and tuples for short heterogeneous ones? Of course we use the Python "duck typing" definition for homogenous.
- cup-of-tea 9y ago
  You could, but I think mine is better. (firstname, lastname) and (x,y,z) are homogeneous collections. But they probably should be tuples. In a tuple the position describes the structure, in a list the position merely describes the order. Hence why tuples are best if you plan to unpack them (ie. firstname, lastname = name). Trouble is Python doesn't completely separate the concepts like some languages do (e.g. C array/struct, R array/list). You can unpack lists and store heterogeneous items in them. Many libraries do this (e.g. SQL libraries will give a list per row by default). And if you follow the bash read idiom in Python using str.split you'll unpack a list as well. For me the distinction is about documenting my code. The only real practical advantage I'm aware of is tuples are hashable. I don't think they are implemented anything like structs underneath.
  - lmm 9y ago
    > (firstname, lastname) and (x,y,z) are homogeneous collections. Disagree. In a language with a decent type system you would give them different types, because using a first name as a last name or an x as a z is always an error.
    - shoo 9y ago
      Yeah. Another rule of thumb is that a lot of the time you can make code that uses tuples clearer by rewriting it to use namedtuple-defined struct values. If the language supported it, it'd probably also make sense most of the time to define a type for each tuple field, and have a type-checker look for errors. Occasionally you see tuples used where they want immutable and hashable lists, but most of the time they're used as shorthand struct/"pod" tours types.
    - cup-of-tea 9y ago
      In what language would you make firstname and lastname or x and y different types?
      - lmm 9y ago
        Well, I would in Rust for starters.
        cup-of-tea 9y ago
        Interesting. I should probably try one of these typed languages. When I was thinking of types I was thinking about "physical" types like in C, not abstract types.
        dllthomas 9y ago
        Yeah, it's pretty common to think of types as "representational", and that really doesn't get at most of what they are good for. Even in C you can get some good help from the type checker if you know how to use it. For instance, see https://dlthomas.github.io/using-c-types-talk/slides/slides.html#26.0 https://dlthomas.github.io/using-c-types-talk/slides/slides.... for a technique that can turn a flakey segfault from misuse of an API into a compile time error (later slides do just that for a simple SDL program, with a modified header file).