Improve UI/UX of post-monomorphization errors #1268
Conversation
src/macros.rs
Outdated
| /// Asserts at compile time that `$condition` is true for `Self` or the given | ||
| /// `$tyvar`s. | ||
| macro_rules! ct_assert { |
There was a problem hiding this comment.
Is there a way to name this (and phrase the doc comment) so that it's clear why it's different from const_assert (immediately above in this file)?
There was a problem hiding this comment.
Note that this name will appear in compiler errors, so we want it to be both externally intelligible and succinct.
I can definitely improve the documentation, but would welcome suggestions of alternative names.
There was a problem hiding this comment.
Oh, I suppose static_assert might be fitting.
src/macros.rs
Outdated
| (Self $(: $(? $optbound:ident $(+)?)* $($bound:ident $(+)?)* )? => $condition:expr) => { | ||
| trait ConstAssert { | ||
| const ASSERT: bool; | ||
| } | ||
|
|
||
| impl<T $(: $(? $optbound +)* $($bound +)*)?> ConstAssert for T { | ||
| const ASSERT: bool = { | ||
| const_assert!($condition); | ||
| $condition | ||
| }; | ||
| } | ||
|
|
||
| const_assert!(<Self as ConstAssert>::ASSERT); | ||
| }; | ||
| ($($tyvar:ident $(: $(? $optbound:ident $(+)?)* $($bound:ident $(+)?)* )?),* => $condition:expr) => { | ||
| trait ConstAssert { | ||
| const ASSERT: bool; | ||
| } | ||
|
|
||
| impl<$($tyvar $(: $(? $optbound +)* $($bound +)*)?,)*> ConstAssert for ($($tyvar,)*) { | ||
| const ASSERT: bool = { | ||
| const_assert!($condition); | ||
| $condition | ||
| }; | ||
| } | ||
|
|
||
| const_assert!(<($($tyvar,)*) as ConstAssert>::ASSERT); | ||
| }; |
There was a problem hiding this comment.
Wrap these bodies in an extra set of curly braces? Otherwise ConstAssert is introduced into the caller's namespace.
src/macros.rs
Outdated
| ($tyvar:ident) => { | ||
| use crate::KnownLayout; | ||
| ct_assert!($tyvar: ?Sized + KnownLayout => { | ||
| let dst_is_zst = match $tyvar::LAYOUT.size_info { | ||
| crate::SizeInfo::Sized { .. } => false, | ||
| crate::SizeInfo::SliceDst(TrailingSliceLayout { elem_size, .. }) => { | ||
| elem_size == 0 | ||
| } | ||
| }; | ||
| !dst_is_zst | ||
| }); | ||
| } |
There was a problem hiding this comment.
Wrap this in an extra set of curly braces? Otherwise, use crate::KnownLayout; is introduced into the caller's namespace.
|
We cannot, because |
Oh look, not the first time we've run into this issue. Bummer. |
jswrenn
force-pushed
the
zsty-dsty-ui-ux
branch
2 times, most recently
from
2a433d4
to
b2cb6b7
Compare
|
Still need a resolution on https://github.com/google/zerocopy/pull/1268/files#r1601997671. |
jswrenn
force-pushed
the
zsty-dsty-ui-ux
branch
2 times, most recently
from
bdba67b
to
6cfde91
Compare
src/macros.rs
Outdated
|
|
||
| /// Assert at compile time that `tyvar` does not have a zero-sized DST | ||
| /// component. | ||
| macro_rules! ct_assert_dst_is_not_zst { |
src/macros.rs
Outdated
| /// Asserts at compile time that `$condition` is true for `Self` or the given | ||
| /// `$tyvar`s. Unlike `const_assert`, this is *strictly* a compile-time check. | ||
| /// The condition is checked after monomorphization and, upon failure, emits a | ||
| /// compile error. |
There was a problem hiding this comment.
const_assert! is also strictly a compile-time check so long as it's always called from const code; its fanciness has to do with its behavior on older toolchains that don't support explicitly panicking in const code. Maybe I should update the docs to be more clear about that.
The distinction is actually that static_assert uses post-monomorphization errors only, and so its errors are generated during a different compiler phase. Arguably we could just inline static_assert into ct_assert_dst_is_not_zst below.
There was a problem hiding this comment.
I made additional clarifications both in the docs of const_assert and static_assert.
By asserting in a macro, rather than a helper function, rustc's diagnostics point to the user's callsite, rather than our own.
By asserting in a macro, rather than a helper function, rustc's diagnostics point to the user's callsite, rather than our own.