Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Clarify the relationships between various kinds of structs and variants #1506

Merged
merged 2 commits into from
Aug 12, 2016
Merged
Changes from 1 commit
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
169 changes: 169 additions & 0 deletions text/0000-adt-kinds.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,169 @@
- Feature Name: clarified_adt_kinds
- Start Date: 2016-02-07
- RFC PR: (leave this empty)
- Rust Issue: (leave this empty)

# Summary
[summary]: #summary

Provide a simple model describing three kinds of structs and variants and their relationships.
Provide a way to match on structs/variants in patterns regardless of their kind (`S{..}`).
Permit tuple structs and tuple variants with zero fields (`TS()`).

# Motivation
[motivation]: #motivation

There's some mental model lying under the current implementation of ADTs, but it is not written
out explicitly and not implemented completely consistently.
Writing this model out helps to identify its missing parts.
Some of this missing parts turn out to be practically useful.
This RFC can also serve as a piece of documentation.

# Detailed design
[design]: #detailed-design

The text below mostly talks about structures, but almost everything is equally applicable to
variants.

## Braced structs

Braced structs are declared with braces (unsurprisingly).

```
struct S {
field1: Type1,
field2: Type2,
field3: Type3,
}
```

Braced structs are the basic struct kind, other kinds are built on top of them.
Braced structs have 0 or more user-named fields and are defined only in type namespace.

Braced structs can be used in struct expressions `S{field1: expr, field2: expr}`, including
functional record update (FRU) `S{field1: expr, ..s}`/`S{..s}` and with struct patterns
`S{field1: pat, field2: pat}`/`S{field1: pat, ..}`/`S{..}`.
In all cases the path `S` of the expression or pattern is looked up in the type namespace.
Fields of a braced struct can be accessed with dot syntax `s.field1`.

Note: struct *variants* are currently defined in the value namespace in addition to type namespace,
there are no particular reasons for this and this is probably temporary.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do these have any use?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No. (Not now at least.)
See rust-lang/rust#30882 (comment)

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I guess this is because regular variants are in both namespaces and braced variants are a mistake. However, if we one day get named fields for functions, then I guess we could use struct names as ctor functions, so it might be useful. Too bad that ship has sailed for regular structs.


## Unit structs

Unit structs are defined without any fields or brackets.

```
struct US;
```

Unit structs can be thought of as a single declaration for two things: a basic struct

```
struct US {}
```

and a constant with the same name

```
const US: US = US{};
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

c.f., an actual braced struct with zero fields, which is not in the value namespace

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think that's the point. An empty struct-defined-with-without-braces is like an empty struct-defined-with-braces, plus a const like this.

```

Unit structs have 0 fields and are defined in both type (the type `US`) and value (the
constant `US`) namespaces.

As a basic struct, a unit struct can participate in struct expressions `US{}`, including FRU
`US{..s}` and in struct patterns `US{}`/`US{..}`. In both cases the path `US` of the expression
or pattern is looked up in the type namespace.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A unit struct/variant pattern is a struct/variant pattern, not a constant pattern, and therefore is exhaustive.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There's a "Note 1" below (I could expand it though).

Fields of a unit struct could also be accessed with dot syntax, but it doesn't have any fields.

As a constant, a unit struct can participate in unit struct expressions `US` and unit struct
patterns `US`, both of these are looked up in the value namespace in which the constant `US` is
defined.

Note 1: the constant is not exactly a `const` item, there are subtle differences, but it's a close
approximation.
Note 2: the constant is pretty weirdly namespaced in case of unit *variants*, constants can't be
defined in "enum modules" manually.

## Tuple structs

Tuple structs are declared with parentheses.
```
struct TS(Type0, Type1, Type2);
```

Tuple structs can be thought of as a single declaration for two things: a basic struct

```
struct TS {
0: Type0,
1: Type1,
2: Type2,
}
```

and a constructor function with the same name

```
fn TS(arg0: Type0, arg1: Type1, arg2: Type2) -> TS {
TS{0: arg0, 1: arg1, 2: arg2}
}
```

Tuple structs have 0 or more automatically-named fields and are defined in both type (the type `TS`)
and the value (the constructor function `TS`) namespaces.

As a basic struct, a tuple struct can participate in struct expressions `TS{0: expr, 1: expr}`,
including FRU `TS{0: expr, ..ts}`/`TS{..ts}` and in struct patterns
`TS{0: pat, 1: pat}`/`TS{0: pat, ..}`/`TS{..}`.
In both cases the path `TS` of the expression or pattern is looked up in the type namespace.
Fields of a braced tuple can be accessed with dot syntax `ts.0`.

As a constructor, a tuple struct can participate in tuple struct expressions `TS(expr, expr)` and
tuple struct patterns `TS(pat, pat)`/`TS(..)`, both of these are looked up in the value namespace
in which the constructor `TS` is defined. Tuple struct expressions `TS(expr, expr)` are usual
function calls, but the compiler reserves the right to make observable improvements to them based
on the additional knowledge, that `TS` is a constructor.

Note: the automatically assigned field names are quite interesting, they are not identifiers
lexically (they are integer literals), so such fields can't be defined manually.

## Summary of the changes.

Everything related to braced structs and unit structs is already implemented.

New: Permit tuple structs and tuple variants with 0 fields. This restriction is artificial and can
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could you expand on the exact semantics here, I assume there is no interoperability between the different kinds of zero-sized structs. Whereas, a unit struct can be thought of as a single constant object, I would think of zero-field tuple structs generating a potentially infinite number of unique instances, but I can't think of how this makes an actual difference in code.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a unit struct can be thought of as a single constant object, I would think of zero-field tuple structs generating a potentially infinite number of unique instances

Can you expand on your intuition here?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

struct TS(); is roughly equivalent to

struct TS {
}

 fn TS() -> TS {
     TS{}
 }

, it's described in the section about tuple structs above.

be lifted trivially. Macro writers dealing with tuple structs/variants will be happy to get rid of
this one special case.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

examples of crates that had to implement workarounds because of this limitation?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I have an anecdotal evidence from @DanielKeep but no concrete examples so far.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A more subtle question might be crates that are broken, but haven't noticed yet, because nobody tried to make it generate a struct with no fields.


New: Permit using tuple structs and tuple variants in braced struct patterns and expressions not
requiring naming their fields - `TS{..ts}`/`TS{}`/`TS{..}`. This doesn't require much effort to
implement as well.
This also means that `S{..}` patterns can be used to match structures and variants of any kind.
The desire to have such "match everything" patterns is sometimes expressed given
that number of fields in structures and variants can change from zero to non-zero and back during
development.

New: Permit using tuple structs and tuple variants in braced struct patterns and expressions
requiring naming their fields - `TS{0: expr}`/`TS{0: pat}`/etc. This change looks a bit worse from
the cost/benefit point of view. There's not much motivation for it besides consistency and probably
shortening patterns like `ItemFn(name, _, _, _, _, _)` into something like `ItemFn{0: name, ..}`.
Automatic code generators (e.g. syntax extensions like `derive`) can probably benefit from the
ability to generate uniform code for all structure kinds as well.
The author of the RFC is ready to postpone or drop this particular extension at any moment.

# Drawbacks
[drawbacks]: #drawbacks

None.

# Alternatives
[alternatives]: #alternatives

None.

# Unresolved questions
[unresolved]: #unresolved-questions

None.