Currently we don't emit lifetime end markers when translating the
unwinding code. I omitted that when I added the support for lifetime
intrinsics, because I initially made the mistake of just returning true
in clean_on_unwind(). That caused almost all calls to be translated as
invokes, leading to quite awful results.
To correctly emit the lifetime end markers, we must differentiate
between cleanup that requires unwinding and such cleanup that just wants
to emit code during unwinding.
Currently we don't emit lifetime end markers when translating the
unwinding code. I omitted that when I added the support for lifetime
intrinsics, because I initially made the mistake of just returning true
in clean_on_unwind(). That caused almost all calls to be translated as
invokes, leading to quite awful results.
To correctly emit the lifetime end markers, we must differentiate
between cleanup that requires unwinding and such cleanup that just wants
to emit code during unwinding.
This makes edge cases in which the `Iterator` trait was not in scope
and/or `Option` or its variants were not in scope work properly.
This breaks code that looks like:
struct MyStruct { ... }
impl MyStruct {
fn next(&mut self) -> Option<int> { ... }
}
for x in MyStruct { ... } { ... }
Change ad-hoc `next` methods like the above to implementations of the
`Iterator` trait. For example:
impl Iterator<int> for MyStruct {
fn next(&mut self) -> Option<int> { ... }
}
Closes#15392.
[breaking-change]
The allocas used in match expression currently don't get good lifetime
markers, in fact they only get lifetime start markers, because their
lifetimes don't match to cleanup scopes.
While the bindings themselves are bog standard and just need a matching
pair of start and end markers, they might need them twice, once for a
guard clause and once for the match body.
The __llmatch alloca OTOH needs a single lifetime start marker, but
when there's a guard clause, it needs two end markers, because its
lifetime ends either when the guard doesn't match or after the match
body.
With these intrinsics in place, LLVM can now, for example, optimize
code like this:
````rust
enum E {
A1(int),
A2(int),
A3(int),
A4(int),
}
pub fn variants(x: E) {
match x {
A1(m) => bar(&m),
A2(m) => bar(&m),
A3(m) => bar(&m),
A4(m) => bar(&m),
}
}
````
To a single call to bar, using only a single stack slot. It still fails
to eliminate some of checks.
````gas
.Ltmp5:
.cfi_def_cfa_offset 16
movb (%rdi), %al
testb %al, %al
je .LBB3_5
movzbl %al, %eax
cmpl $1, %eax
je .LBB3_5
cmpl $2, %eax
.LBB3_5:
movq 8(%rdi), %rax
movq %rax, (%rsp)
leaq (%rsp), %rdi
callq _ZN3bar20hcb7a0d8be8e17e37daaE@PLT
popq %rax
retq
````
Lifetime intrinsics help to reduce stack usage, because LLVM can apply
stack coloring to reuse the stack slots of dead allocas for new ones.
For example these functions now both use the same amount of stack, while
previous `bar()` used five times as much as `foo()`:
````rust
fn foo() {
println("{}", 5);
}
fn bar() {
println("{}", 5);
println("{}", 5);
println("{}", 5);
println("{}", 5);
println("{}", 5);
}
````
On top of that, LLVM can also optimize out certain operations when it
knows that memory is dead after a certain point. For example, it can
sometimes remove the zeroing used to cancel the drop glue. This is
possible when the glue drop itself was already removed because the
zeroing dominated the drop glue call. For example in:
````rust
pub fn bar(x: (Box<int>, int)) -> (Box<int>, int) {
x
}
````
With optimizations, this currently results in:
````llvm
define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 {
"_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit":
%2 = bitcast { i64*, i64 }* %1 to i8*
%3 = bitcast { i64*, i64 }* %0 to i8*
tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false)
tail call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 16, i32 8, i1 false)
ret void
}
````
But with lifetime intrinsics we get:
````llvm
define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 {
"_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit":
%2 = bitcast { i64*, i64 }* %1 to i8*
%3 = bitcast { i64*, i64 }* %0 to i8*
tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false)
tail call void @llvm.lifetime.end(i64 16, i8* %2)
ret void
}
````
Fixes#15665
Lifetime intrinsics help to reduce stack usage, because LLVM can apply
stack coloring to reuse the stack slots of dead allocas for new ones.
For example these functions now both use the same amount of stack, while
previous `bar()` used five times as much as `foo()`:
````rust
fn foo() {
println("{}", 5);
}
fn bar() {
println("{}", 5);
println("{}", 5);
println("{}", 5);
println("{}", 5);
println("{}", 5);
}
````
On top of that, LLVM can also optimize out certain operations when it
knows that memory is dead after a certain point. For example, it can
sometimes remove the zeroing used to cancel the drop glue. This is
possible when the glue drop itself was already removed because the
zeroing dominated the drop glue call. For example in:
````rust
pub fn bar(x: (Box<int>, int)) -> (Box<int>, int) {
x
}
````
With optimizations, this currently results in:
````llvm
define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 {
"_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit":
%2 = bitcast { i64*, i64 }* %1 to i8*
%3 = bitcast { i64*, i64 }* %0 to i8*
tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false)
tail call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 16, i32 8, i1 false)
ret void
}
````
But with lifetime intrinsics we get:
````llvm
define void @_ZN3bar20h330fa42547df8179niaE({ i64*, i64 }* noalias nocapture nonnull sret, { i64*, i64 }* noalias nocapture nonnull) unnamed_addr #0 {
"_ZN29_$LP$Box$LT$int$GT$$C$int$RP$39glue_drop.$x22glue_drop$x22$LP$1347$RP$17h88cf42702e5a322aE.exit":
%2 = bitcast { i64*, i64 }* %1 to i8*
%3 = bitcast { i64*, i64 }* %0 to i8*
tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %3, i8* %2, i64 16, i32 8, i1 false)
tail call void @llvm.lifetime.end(i64 16, i8* %2)
ret void
}
````
Fixes#15665
`call_visit_glue` is only ever called from trans_intrinsic, and the
block won't be unreachable there. Also, the comment doesn't make sense
anymore. When the code was introduced in 38fee9526a the function was
also responsible for the cleanup glue, which is no longer the case.
While we're at it, also fixed the debug message to output the right
function name.
This implements RFC 39. Omitted lifetimes in return values will now be
inferred to more useful defaults, and an error is reported if a lifetime
in a return type is omitted and one of the two lifetime elision rules
does not specify what it should be.
This primarily breaks two uncommon code patterns. The first is this:
unsafe fn get_foo_out_of_thin_air() -> &Foo {
...
}
This should be changed to:
unsafe fn get_foo_out_of_thin_air() -> &'static Foo {
...
}
The second pattern that needs to be changed is this:
enum MaybeBorrowed<'a> {
Borrowed(&'a str),
Owned(String),
}
fn foo() -> MaybeBorrowed {
Owned(format!("hello world"))
}
Change code like this to:
enum MaybeBorrowed<'a> {
Borrowed(&'a str),
Owned(String),
}
fn foo() -> MaybeBorrowed<'static> {
Owned(format!("hello world"))
}
Closes#15552.
[breaking-change]
r? @nick29581
This is accomplished by rewriting static expressions into equivalent patterns.
This way, patterns referencing static variables can both participate
in exhaustiveness analysis as well as be compiled down into the appropriate
branch of the decision trees that match expressions are codegened to.
Fixes#6533.
Fixes#13626.
Fixes#13731.
Fixes#14576.
Fixes#15393.
This implements RFC 39. Omitted lifetimes in return values will now be
inferred to more useful defaults, and an error is reported if a lifetime
in a return type is omitted and one of the two lifetime elision rules
does not specify what it should be.
This primarily breaks two uncommon code patterns. The first is this:
unsafe fn get_foo_out_of_thin_air() -> &Foo {
...
}
This should be changed to:
unsafe fn get_foo_out_of_thin_air() -> &'static Foo {
...
}
The second pattern that needs to be changed is this:
enum MaybeBorrowed<'a> {
Borrowed(&'a str),
Owned(String),
}
fn foo() -> MaybeBorrowed {
Owned(format!("hello world"))
}
Change code like this to:
enum MaybeBorrowed<'a> {
Borrowed(&'a str),
Owned(String),
}
fn foo() -> MaybeBorrowed<'static> {
Owned(format!("hello world"))
}
Closes#15552.
[breaking-change]
This is accomplished by rewriting static expressions into equivalent patterns.
This way, patterns referencing static variables can both participate
in exhaustiveness analysis as well as be compiled down into the appropriate
branch of the decision trees that match expressions are codegened to.
Fixes#6533.
Fixes#13626.
Fixes#13731.
Fixes#14576.
Fixes#15393.
`call_visit_glue` is only ever called from trans_intrinsic, and the
block won't be unreachable there. Also, the comment doesn't make sense
anymore. When the code was introduced in 38fee9526a the function was
also responsible for the cleanup glue, which is no longer the case.
While we're at it, also fixed the debug message to output the right
function name.
This should fix issue #15541. It would be good to have an test case for this would also be nice but I haven't had the time to write one. The change is very small though and it doesn't break anything in the existing test suite, so I guess we can add it without test for now.
except where trait objects are involved.
Part of issue #15349, though I'm leaving it open for trait objects.
Cross borrowing for trait objects remains because it is needed until we
have DST.
This will break code like:
fn foo(x: &int) { ... }
let a = box 3i;
foo(a);
Change this code to:
fn foo(x: &int) { ... }
let a = box 3i;
foo(&*a);
[breaking-change]
This makes two changes to region inference: (1) it allows region
inference to relate early-bound regions; and (2) it allows regions to be
related before variance runs. The former is needed because there is no
relation between the two regions before region substitution happens,
while the latter is needed because type collection has to run before
variance. We assume that, before variance is inferred, that lifetimes
are invariant. This is a conservative overapproximation.
This relates to #13885. This does not remove `~self` from the language
yet, however.
[breaking-change]
Disabling the redzone is required in x86-64's kernel mode to avoid interrupts trashing the stack.
I'm not sure if decl_fn is the right place to tag all functions with noredzone. It might have interactions with external functions when linking with bitcode built without -C no-redzone although I see no reason to do that.
I'm not sure how to write a test inspecting the bitcode output for noredzone attributes on all functions either.
Per @pnkfelix 's suggestion, using a trait to make these
field accesses more readable (and vastly more similar
to the original code.
oops fix new ast_map fix
This change propagates to many locations, but because of the
Macro Exterminator (or, more properly, the invariant that it
protects), macro invocations can't occur downstream of expansion.
This means that in librustc and librustdoc, extracting the
desired field can simply assume that it can't be a macro
invocation. Functions in ast_util abstract over this check.
This removes a bunch of token types. Tokens now store the original, unaltered
numeric literal (that is still checked for correctness), which is parsed into
an actual number later, as needed, when creating the AST.
This can change how syntax extensions work, but otherwise poses no visible
changes.
[breaking-change]
formerly, the self identifier was being discarded during parsing, which
stymies hygiene. The best fix here seems to be to attach a self identifier
to ExplicitSelf_, a change that rippled through the rest of the compiler,
but without any obvious damage.
Extend the null ptr optimization to work with slices, closures, procs, & trait objects by using the internal pointers as the discriminant.
This decreases the size of `Option<&[int]>` (and similar) by one word.
