The structure of the old translator as well as MIR assumed that drop glue cannot possibly panic and
translated the drops accordingly. However, in presence of `Drop::drop` this assumption can be
trivially shown to be untrue. As such, the Rust code like the following would never print number 2:
```rust
struct Droppable(u32);
impl Drop for Droppable {
fn drop(&mut self) {
if self.0 == 1 { panic!("Droppable(1)") } else { println!("{}", self.0) }
}
}
fn main() {
let x = Droppable(2);
let y = Droppable(1);
}
```
While the behaviour is allowed according to the language rules (we allow drops to not run), that’s
a very counter-intuitive behaviour. We fix this in MIR by allowing `Drop` to have a target to take
on divergence and connect the drops in such a way so the leftover drops are executed when some drop
unwinds.
Note, that this commit still does not implement the translator part of changes necessary for the
grand scheme of things to fully work, so the actual observed behaviour does not change yet. Coming
soon™.
See #14875.
We used to have CallKind only because there was a requirement to have all successors in a
contiguous memory block. Now that the requirement is gone, remove the CallKind and instead just
have the necessary information inline.
Awesome!
This commit transitions the compiler to using the new exception handling
instructions in LLVM for implementing unwinding for MSVC. This affects both 32
and 64-bit MSVC as they're both now using SEH-based strategies. In terms of
standard library support, lots more details about how SEH unwinding is
implemented can be found in the commits.
In terms of trans, this change necessitated a few modifications:
* Branches were added to detect when the old landingpad instruction is used or
the new cleanuppad instruction is used to `trans::cleanup`.
* The return value from `cleanuppad` is not stored in an `alloca` (because it
cannot be).
* Each block in trans now has an `Option<LandingPad>` instead of `is_lpad: bool`
for indicating whether it's in a landing pad or not. The new exception
handling intrinsics require that on MSVC each `call` inside of a landing pad
is annotated with which landing pad that it's in. This change to the basic
block means that whenever a `call` or `invoke` instruction is generated we
know whether to annotate it as part of a cleanuppad or not.
* Lots of modifications were made to the instruction builders to construct the
new instructions as well as pass the tagging information for the call/invoke
instructions.
* The translation of the `try` intrinsics for MSVC has been overhauled to use
the new `catchpad` instruction. The filter function is now also a
rustc-generated function instead of a purely libstd-defined function. The
libstd definition still exists, it just has a stable ABI across architectures
and leaves some of the really weird implementation details to the compiler
(e.g. the `localescape` and `localrecover` intrinsics).
This brings some routine upgrades to the bundled LLVM that we're using, the most
notable of which is a bug fix to the way we handle range asserts when loading
the discriminant of an enum. This fix ended up being very similar to f9d4149c
where we basically can't have a range assert when loading a discriminant due to
filling drop, and appropriate flags were added to communicate this to
`trans::adt`.
Previously it was returning a value, mostly for the two reasons:
* Cloning Lvalue is very cheap most of the time (i.e. when Lvalue is not a Projection);
* There’s users who want &mut lvalue and there’s users who want &lvalue. Returning a value allows
to make either one easier when pattern matching (i.e. Some(ref dest) or Some(ref mut dest)).
However, I’m now convinced this is an invalid approach. Namely the users which want a mutable
reference may modify the Lvalue in-place, but the changes won’t be reflected in the final MIR,
since the Lvalue modified is merely a clone.
Instead, we have two accessors `destination` and `destination_mut` which return a reference to the
destination in desired mode.
This considerably simplifies code around calling functions and translation of Resume itself. This
removes requirement that a block containing Resume terminator is always translated after something
which creates a landing pad, thus allowing us to actually translate some valid MIRs we could not
translate before.
However, an assumption is added that translator is correct (in regards to landing pad generation)
and code will never reach the Resume terminator without going through a landing pad first. Breaking
these assumptions would pass an `undef` value into the personality functions.
This merges two separate Call terminators and uses a separate CallKind sub-enum instead.
A little bit unrelatedly, copying into destination value for a certain kind of invoke, is also
implemented here. See the associated comment in code for various details that arise with this
implementation.
DivergingCall is different enough from the regular converging Call to warrant the split. This also
inlines CallData struct and creates a new CallTargets enum in order to have a way to differentiate
between calls that do not have an associated cleanup block.
Note, that this patch still does not produce DivergingCall terminator anywhere. Look for that in
the next patches.
