derive(Zeroable) on fieldful enums and repr(C) enums #257

zachs18 · 2024-07-28T06:24:03Z

Closes #230.

Allows deriving Zeroable on enums where:

The enum is #[repr(Int)], #[repr(C)], or #[repr(C, Int)],
There is a variant with discriminant 0,
The variant with discriminant 0 is either fieldless, or all of its fields are Zeroable.

Also allows using the "perfect derive with additional bounds" from #196 for Zeroable on enums, where the fields for the "perfect derive" are from the variant with discriminant 0 (see the MyOption example on line 189 of derive/src/lib.rs).

Internal changes:

VariantDiscriminantIter now also gives the &'a Variant associated with the discriminant.
the comment on line 338 of derive/src/traits.rs kept getting mangled by cargo fmt, so I put it on its own line.
In derive_marker_trait_inner: let predicates = ... moved to after let fields = ..., since computing fields now requires accessing input, but predicates borrows it mutably. Does not change semantics.
generate_fields_are_trait and some other helper functions now take an Option<&Variant> to operate on if they otherwise don't support enums.
some cargo fmt
drive-by clarification/typo fixes (e.g. renaming discriminantor -> discriminant)

danielhenrymantilla

LGTM, nice! ✅

danielhenrymantilla · 2024-09-10T09:45:19Z

derive/src/traits.rs

@@ -139,24 +142,23 @@ impl Derivable for Zeroable {
    match ty {
      Data::Struct(_) => Ok(()),
      Data::Enum(DataEnum { variants, .. }) => {


Unrelated to this PR; but this logic does not seem to fit the name of the function (check_attributes), since it is also checking the variants and whatnot for enums. The simplest fix here would be to just change the function name accordingly, fn check_input(), fn validate[_input] or whatnot.

derive/src/traits.rs

danielhenrymantilla · 2024-09-10T09:53:45Z

derive/src/traits.rs

+      VariantDiscriminantIterator::new(variants.iter())
+        .map(|res| {
+          let (discriminant, _variant) = res?;
+          Ok(LitInt::new(&format!("{}", discriminant), span))


Unrelated to this PR, but we might as well smuggle this tiny improvement:

Suggested change

Ok(LitInt::new(&format!("{}", discriminant), span))

Ok(LitInt::new(&discriminant.to_string(), span))

danielhenrymantilla · 2024-09-10T09:55:24Z

derive/src/traits.rs

+    let first = &variant_discriminant_lits[0];
+    let rest = &variant_discriminant_lits[1..];


Unrelated to this PR, but it seems quite tempting to be doing the following, here:

Suggested change

let first = &variant_discriminant_lits[0];

let rest = &variant_discriminant_lits[1..];

let (first, rest) = variant_discriminant_lits.split_first().unwrap();

derive/src/traits.rs

danielhenrymantilla · 2024-09-10T10:04:34Z

derive/src/traits.rs

@@ -1215,7 +1248,7 @@ impl<'a, I: Iterator<Item = &'a Variant> + 'a> Iterator
      self.last_value += 1;


⚠️ Unrelated to this PR, but it seems rather important (somebody compiling proc-macros with a different choice of the overflow-checks knob may run into a panic with this logic):

Suggested change

self.last_value += 1;

self.last_value = self.last_value.wrapping_add(1);

Hmm, in the case of the (unstable) repr128 feature, I think this could cause unsoundness with derive(CheckedBitPattern) on an #[repr(i128)] enum X { A = i64::MAX as i128, B }, so to be safe, I'll make last_value an i128 and then do the wrapping_add (since the compiler will catch any overflows at the actual discriminant type).

derive/src/traits.rs

derive/src/lib.rs

danielhenrymantilla · 2024-09-10T10:21:35Z

derive/src/traits.rs

+  /// If this trait has a custom meaning for "perfect derive", this function
+  /// should be overridden to return `Some`.
+  ///
+  /// The default is "the fields of a struct; unions and enums not supported".


Good and important doc 👌

🤔 I wonder if we shouldn't be moving said default logic over here (by changing the output to ::syn::Result<Fields>) to ensure actual code logic and this doc comment remain in sync

1. the enum is repr(Int), repr(C), or repr(C, Int), 2. the enum has a variant with discriminant 0, 3. and all fields of the variant with discriminant 0 are Zeroable.

…on and doctests.

Co-authored-by: Daniel Henry-Mantilla <daniel.henry.mantilla@gmail.com>

fu5ha

Nice, LGTM. I'm deferring the in depth soundness checking of the macro to @danielhenrymantilla 's review whom I trust.

Lokathor requested a review from fu5ha July 30, 2024 22:16

Lokathor self-assigned this Aug 22, 2024

Lokathor added the proc-macros I don't do proc-macros, but I accepts PRs about them. label Aug 22, 2024

danielhenrymantilla approved these changes Sep 10, 2024

View reviewed changes

zachs18 added 7 commits September 13, 2024 14:08

Add support for deriving Zeroable for fieldful enums if:

d2c1197

1. the enum is repr(Int), repr(C), or repr(C, Int), 2. the enum has a variant with discriminant 0, 3. and all fields of the variant with discriminant 0 are Zeroable.

Allow using derive(Zeroable) with explicit bounds. Update documentati…

ff316f6

…on and doctests.

doc update

c8e6fca

doc update

92f77c8

remove unused

d107b2a

Factor out get_zero_variant helper function.

be6ece8

Use i128 to track disciminants instead of i64.

b5ac207

zachs18 force-pushed the zeroable-derive-fieldful-enums branch from 653200e to b5ac207 Compare September 16, 2024 19:44

zachs18 and others added 2 commits September 16, 2024 14:46

Add doc-comment for get_fields

4dc453a

Co-authored-by: Daniel Henry-Mantilla <daniel.henry.mantilla@gmail.com>

Update derive/src/traits.rs

1b02024

Co-authored-by: Daniel Henry-Mantilla <daniel.henry.mantilla@gmail.com>

fu5ha approved these changes Sep 21, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

derive(Zeroable) on fieldful enums and repr(C) enums #257

derive(Zeroable) on fieldful enums and repr(C) enums #257

zachs18 commented Jul 28, 2024

danielhenrymantilla left a comment

danielhenrymantilla Sep 10, 2024

danielhenrymantilla Sep 10, 2024

danielhenrymantilla Sep 10, 2024

danielhenrymantilla Sep 10, 2024

zachs18 Sep 16, 2024 •

edited

Loading

danielhenrymantilla Sep 10, 2024

fu5ha left a comment

	Ok(LitInt::new(&format!("{}", discriminant), span))
	Ok(LitInt::new(&discriminant.to_string(), span))

		let first = &variant_discriminant_lits[0];
		let rest = &variant_discriminant_lits[1..];

	let first = &variant_discriminant_lits[0];
	let rest = &variant_discriminant_lits[1..];
	let (first, rest) = variant_discriminant_lits.split_first().unwrap();

		@@ -1215,7 +1248,7 @@ impl<'a, I: Iterator<Item = &'a Variant> + 'a> Iterator
		self.last_value += 1;

	self.last_value += 1;
	self.last_value = self.last_value.wrapping_add(1);

derive(Zeroable) on fieldful enums and repr(C) enums #257

Are you sure you want to change the base?

derive(Zeroable) on fieldful enums and repr(C) enums #257

Conversation

zachs18 commented Jul 28, 2024

danielhenrymantilla left a comment

Choose a reason for hiding this comment

danielhenrymantilla Sep 10, 2024

Choose a reason for hiding this comment

danielhenrymantilla Sep 10, 2024

Choose a reason for hiding this comment

danielhenrymantilla Sep 10, 2024

Choose a reason for hiding this comment

danielhenrymantilla Sep 10, 2024

Choose a reason for hiding this comment

zachs18 Sep 16, 2024 • edited Loading

Choose a reason for hiding this comment

danielhenrymantilla Sep 10, 2024

Choose a reason for hiding this comment

fu5ha left a comment

Choose a reason for hiding this comment

zachs18 Sep 16, 2024 •

edited

Loading