This commit stabilizes the `std::time` module and the `Duration` type.
`Duration::span` remains unstable, and the `Display` implementation for
`Duration` has been removed as it is still being reworked and all trait
implementations for stable types are de facto stable.
This is a [breaking-change] to those using `Duration`'s `Display`
implementation.
I'm opening this PR as a platform for discussion - there may be some method renaming to do as part of the stabilization process.
These two commits are aimed at "fixing" our usage of `dllexport` in the compiler. Currently we blanket apply this attribute to *everything* public in a crate, but this ends up with a few downsides:
* Executables are larger than the should be as a result of thinking they should export everything
* Native libraries aren't handled correctly because technically a statically included native library should be exported from a DLL in some cases.
* Symbols don't actually need to be exported if they never end up in a DLL.
The first commit adds a new unstable attribute, `#[linked_from]`, which is a way to tell the compiler what native library a block of symbols comes from. This is used to inform the compiler what set of native libraries are statically included in the rlib (or other output). This information is later used to export them from a DLL if necessary. Currently this is only used in a few places (such as the LLVM bindings) to get the compiler to link correctly.
The second commit stops adding `dllexport` to all items in LLVM and instead explicitly telling the linker what symbols should be exported. We only need to do this when building a dynamic library, and otherwise we can avoid adding `dllexport` or telling the linker about exported symbols.
As a testament to this change, the size of "Hello World" on MSVC drops from 1.2MB to 67KB as a result of this patch. This is because the linker can much more aggressively remove unused code.
These commits do not yet attempt to fix our story with `dllimport`, and I'll leave that to a future PR and issue, for now though I'm going to say that this
Closes#7196
Rust's current compilation model makes it impossible on Windows to generate one
object file with a complete and final set of dllexport annotations. This is
because when an object is generated the compiler doesn't actually know if it
will later be included in a dynamic library or not. The compiler works around
this today by flagging *everything* as dllexport, but this has the drawback of
exposing too much.
Thankfully there are alternate methods of specifying the exported surface area
of a dll on Windows, one of which is passing a `*.def` file to the linker which
lists all public symbols of the dynamic library. This commit removes all
locations that add `dllexport` to LLVM variables and instead dynamically
generates a `*.def` file which is passed to the linker. This file will include
all the public symbols of the current object file as well as all upstream
libraries, and the crucial aspect is that it's only used when generating a
dynamic library. When generating an executable this file isn't generated, so all
the symbols aren't exported from an executable.
To ensure that statically included native libraries are reexported correctly,
the previously added support for the `#[linked_from]` attribute is used to
determine the set of FFI symbols that are exported from a dynamic library, and
this is required to get the compiler to link correctly.
To correctly reexport statically included libraries from a DLL on Windows, the
compiler will soon need to have knowledge about what symbols are statically
included and which are not. To solve this problem a new unstable
`#[linked_from]` attribute is being added and recognized on `extern` blocks to
indicate which native library the symbols are coming from.
The compiler then keeps track of what the set of FFI symbols are that are
included statically. This information will be used in a future commit to
configure how we invoke the linker on Windows.
This means that we no longer need to ship the `llvm-ar.exe` binary in the MSVC
distribution, and after a snapshot we can remove a good bit of logic from the
makefiles!
This commit stabilizes the `std::time` module and the `Duration` type.
`Duration::span` remains unstable, and the `Display` implementation for
`Duration` has been removed as it is still being reworked and all trait
implementations for stable types are de facto stable.
This is a [breaking-change] to those using `Duration`'s `Display`
implementation.
This is a temporary workaround for the bugs that have been found in
the implementation of PR #26173.
* pnkfelix is unavailable in the short-term (i.e. for the next week) to fix them.
* When the bugs are fixed, we will turn this back on by default.
(If you want to play with the known-to-be-buggy optimization in the
meantime, you can opt-back in via the debugging option that this
commit is toggling.)
This commit removes all morestack support from the compiler which entails:
* Segmented stacks are no longer emitted in codegen.
* We no longer build or distribute libmorestack.a
* The `stack_exhausted` lang item is no longer required
The only current use of the segmented stack support in LLVM is to detect stack
overflow. This is no longer really required, however, because we already have
guard pages for all threads and registered signal handlers watching for a
segfault on those pages (to print out a stack overflow message). Additionally,
major platforms (aka Windows) already don't use morestack.
This means that Rust is by default less likely to catch stack overflows because
if a function takes up more than one page of stack space it won't hit the guard
page. This is what the purpose of morestack was (to catch this case), but it's
better served with stack probes which have more cross platform support and no
runtime support necessary. Until LLVM supports this for all platform it looks
like morestack isn't really buying us much.
cc #16012 (still need stack probes)
Closes#26458 (a drive-by fix to help diagnostics on stack overflow)
r? @brson
This commit removes all morestack support from the compiler which entails:
* Segmented stacks are no longer emitted in codegen.
* We no longer build or distribute libmorestack.a
* The `stack_exhausted` lang item is no longer required
The only current use of the segmented stack support in LLVM is to detect stack
overflow. This is no longer really required, however, because we already have
guard pages for all threads and registered signal handlers watching for a
segfault on those pages (to print out a stack overflow message). Additionally,
major platforms (aka Windows) already don't use morestack.
This means that Rust is by default less likely to catch stack overflows because
if a function takes up more than one page of stack space it won't hit the guard
page. This is what the purpose of morestack was (to catch this case), but it's
better served with stack probes which have more cross platform support and no
runtime support necessary. Until LLVM supports this for all platform it looks
like morestack isn't really buying us much.
cc #16012 (still need stack probes)
Closes#26458 (a drive-by fix to help diagnostics on stack overflow)
LLVM might perform tail merging on the calls that initiate the unwinding
process which breaks debuginfo and therefore this test. Since tail
merging is guaranteed to break debuginfo, it should be disabled for this
test.
This allows us to restore a testcase that I had to remove earlier
because of the same problem, because back then I didn't realize that
disabling tail merging was an option.
cc #27619
The replacements are functions that usually use a single `mem::transmute` in their body and restrict input and output via more concrete types than `T` and `U`. Worth noting are the `transmute` functions for slices and the `from_utf8*` family for mutable slices. Additionally, `mem::transmute` was often used for casting raw pointers, when you can already cast raw pointers just fine with `as`.
This builds upon #27233.
The investigation into #14232 discovered that it's possible that signal delivery
to a newly spawned process is racy on OSX. This test has been failing spuriously
on the OSX bots for some time now, so ignore it as we don't currently know a
solution and it looks like it may be out of our control.
As noted in my previous PR #27439 , the import resolution algorithm has two cases where it bails out:
- The algorithm will delay an import if the module containing the target of the import still has unresolved glob imports
- The algorithm will delay a glob import of the target module still has unresolved imports
This PR alters the behaviour to only bail out when the above described unresolved imports are `pub`, as non-pub imports don't affect the result anyway.
It is still possible to fail the algorithm with examples like
```rust
pub mod a {
pub use b::*;
}
pub mod b {
pub use a::*;
}
```
but such configurations cannot be resolved in any meaningful way, as these are cyclic imports.
Closes#4865
See line 181: The lookup should start with the random index and iterate from there.
Also locked stdout (which makes it a bit faster on my machine).
Perhaps the multi-thread output from the fasta benchmark could be used to speed it up even more.
New enough find on Linux doesn't support "-perm +..." and suggests
using "-perm /..." instead, but that doesn't work on Windows.
Hopefully all platforms are happy with this expanded version.
The replacements are functions that usually use a single `mem::transmute` in
their body and restrict input and output via more concrete types than `T` and
`U`. Worth noting are the `transmute` functions for slices and the `from_utf8*`
family for mutable slices. Additionally, `mem::transmute` was often used for
casting raw pointers, when you can already cast raw pointers just fine with
`as`.
