This pull request extracts all scheduling functionality from libstd, moving it into its own separate crates. The new libnative and libgreen will be the new way in which 1:1 and M:N scheduling is implemented. The standard library still requires an interface to the runtime, however, (think of things like `std::comm` and `io::println`). The interface is now defined by the `Runtime` trait inside of `std::rt`.
The booting process is now that libgreen defines the start lang-item and that's it. I want to extend this soon to have libnative also have a "start lang item" but also allow libgreen and libnative to be linked together in the same process. For now though, only libgreen can be used to start a program (unless you define the start lang item yourself). Again though, I want to change this soon, I just figured that this pull request is large enough as-is.
This certainly wasn't a smooth transition, certain functionality has no equivalent in this new separation, and some functionality is now better enabled through this new system. I did my best to separate all of the commits by topic and keep things fairly bite-sized, although are indeed larger than others.
As a note, this is currently rebased on top of my `std::comm` rewrite (or at least an old copy of it), but none of those commits need reviewing (that will all happen in another pull request).
* vec::raw::to_ptr is gone
* Pausible => Pausable
* Removing @
* Calling the main task "<main>"
* Removing unused imports
* Removing unused mut
* Bringing some libextra tests up to date
* Allowing compiletest to work at stage0
* Fixing the bootstrap-from-c rmake tests
* assert => rtassert in a few cases
* printing to stderr instead of stdout in fail!()
This measure is simply to allow programs to continue compiling as they once did.
In the future, this needs a more robust solution to choose how to start with
libgreen or libnative.
For now, this moves the following modules to std::sync
* UnsafeArc (also removed unwrap method)
* mpsc_queue
* spsc_queue
* atomics
* mpmc_bounded_queue
* deque
We may want to remove some of the queues, but for now this moves things out of
std::rt into std::sync
This uses quite a bit of unsafe code for speed and failure safety, and allocates `2*n` temporary storage.
[Performance](https://gist.github.com/huonw/5547f2478380288a28c2):
| n | new | priority_queue | quick3 |
|-------:|---------:|---------------:|---------:|
| 5 | 200 | 155 | 106 |
| 100 | 6490 | 8750 | 5810 |
| 10000 | 1300000 | 1790000 | 1060000 |
| 100000 | 16700000 | 23600000 | 12700000 |
| sorted | 520000 | 1380000 | 53900000 |
| trend | 1310000 | 1690000 | 1100000 |
(The times are in nanoseconds, having subtracted the set-up time (i.e. the `just_generate` bench target).)
I imagine that there is still significant room for improvement, particularly because both priority_queue and quick3 are doing a static call via `Ord` or `TotalOrd` for the comparisons, while this is using a (boxed) closure.
Also, this code does not `clone`, unlike `quick_sort3`; and is stable, unlike both of the others.
Trap the io_error condition so that a more informative error message is
displayed when the linker program cannot be started, such as when the
name of the linker binary is accidentally mistyped.
closes#10755
Right now the --crate-id and related flags are all process *after* the entire
crate is parsed. This is less than desirable when used with makefiles because it
means that just to learn the output name of the crate you have to parse the
entire crate (unnecessary).
This commit changes the behavior to lift the handling of these flags much sooner
in the compilation process. This allows us to not have to parse the entire crate
and only have to worry about parsing the crate attributes themselves. The
related methods have all been updated to take an array of attributes rather than
a crate.
Additionally, this ceases duplication of the "what output are we producing"
logic in order to correctly handle things in the case of --test.
Finally, this adds tests for all of this functionality to ensure that it does
not regress.
Right now the --crate-id and related flags are all process *after* the entire
crate is parsed. This is less than desirable when used with makefiles because it
means that just to learn the output name of the crate you have to parse the
entire crate (unnecessary).
This commit changes the behavior to lift the handling of these flags much sooner
in the compilation process. This allows us to not have to parse the entire crate
and only have to worry about parsing the crate attributes themselves. The
related methods have all been updated to take an array of attributes rather than
a crate.
Additionally, this ceases duplication of the "what output are we producing"
logic in order to correctly handle things in the case of --test.
Finally, this adds tests for all of this functionality to ensure that it does
not regress.
We were previously reading metadata via `ar p`, but as learned from rustdoc
awhile back, spawning a process to do something is pretty slow. Turns out LLVM
has an Archive class to read archives, but it cannot write archives.
This commits adds bindings to the read-only version of the LLVM archive class
(with a new type that only has a read() method), and then it uses this class
when reading the metadata out of rlibs. When you put this in tandem of not
compressing the metadata, reading the metadata is 4x faster than it used to be
The timings I got for reading metadata from the respective libraries was:
libstd-04ff901e-0.9-pre.dylib => 100ms
libstd-04ff901e-0.9-pre.rlib => 23ms
librustuv-7945354c-0.9-pre.dylib => 4ms
librustuv-7945354c-0.9-pre.rlib => 1ms
librustc-5b94a16f-0.9-pre.dylib => 87ms
librustc-5b94a16f-0.9-pre.rlib => 35ms
libextra-a6ebb16f-0.9-pre.dylib => 63ms
libextra-a6ebb16f-0.9-pre.rlib => 15ms
libsyntax-2e4c0458-0.9-pre.dylib => 86ms
libsyntax-2e4c0458-0.9-pre.rlib => 22ms
In order to always take advantage of these faster metadata read-times, I sort
the files in filesearch based on whether they have an rlib extension or not
(prefer all rlib files first).
Overall, this halved the compile time for a `fn main() {}` crate from 0.185s to
0.095s on my system (when preferring dynamic linking). Reading metadata is still
the slowest pass of the compiler at 0.035s, but it's getting pretty close to
linking at 0.021s! The next best optimization is to just not copy the metadata
from LLVM because that's the most expensive part of reading metadata right now.
We were previously reading metadata via `ar p`, but as learned from rustdoc
awhile back, spawning a process to do something is pretty slow. Turns out LLVM
has an Archive class to read archives, but it cannot write archives.
This commits adds bindings to the read-only version of the LLVM archive class
(with a new type that only has a read() method), and then it uses this class
when reading the metadata out of rlibs. When you put this in tandem of not
compressing the metadata, reading the metadata is 4x faster than it used to be
The timings I got for reading metadata from the respective libraries was:
libstd-04ff901e-0.9-pre.dylib => 100ms
libstd-04ff901e-0.9-pre.rlib => 23ms
librustuv-7945354c-0.9-pre.dylib => 4ms
librustuv-7945354c-0.9-pre.rlib => 1ms
librustc-5b94a16f-0.9-pre.dylib => 87ms
librustc-5b94a16f-0.9-pre.rlib => 35ms
libextra-a6ebb16f-0.9-pre.dylib => 63ms
libextra-a6ebb16f-0.9-pre.rlib => 15ms
libsyntax-2e4c0458-0.9-pre.dylib => 86ms
libsyntax-2e4c0458-0.9-pre.rlib => 22ms
In order to always take advantage of these faster metadata read-times, I sort
the files in filesearch based on whether they have an rlib extension or not
(prefer all rlib files first).
Overall, this halved the compile time for a `fn main() {}` crate from 0.185s to
0.095s on my system (when preferring dynamic linking). Reading metadata is still
the slowest pass of the compiler at 0.035s, but it's getting pretty close to
linking at 0.021s! The next best optimization is to just not copy the metadata
from LLVM because that's the most expensive part of reading metadata right now.
Now that the metadata is an owned value with a lifetime of a borrowed byte
slice, it's possible to have future optimizations where the metadata doesn't
need to be copied around (very expensive operation).
Now that the metadata is an owned value with a lifetime of a borrowed byte
slice, it's possible to have future optimizations where the metadata doesn't
need to be copied around (very expensive operation).
For `str.as_mut_buf`, un-closure-ification is achieved by outright removal (see commit message). The others are replaced by `.as_ptr`, `.as_mut_ptr` and `.len`
Trap the io_error condition so that a more informative error message is
displayed when the linker program cannot be started, such as when the
name of the linker binary is accidentally mistyped.
closes#10755
As the title says. The trans changes will lead to an auxiliary alloca being created that allows debug info to track the `self` argument. This alloca is only created in debug builds however. Otherwise very little had to be done after I managed to navigate to some degree the jungle that is self-argument handling `:P`
Closes#10549
When a borrow occurs twice illegally, Rust will label the other borrow
as the "second borrow". This is quite confusing, as the "second borrow"
usually happened before the flagged barrow (e.g. as far as dataflow
is concerned, the first borrow is OK, the second borrow is illegal.)
This patch renames "second borrow" to "previous borrow", to make the
spatial relationship between the two borrows clearer.
Signed-off-by: Edward Z. Yang <ezyang@cs.stanford.edu>
This code in resolve accidentally forced all types with an impl to become
public. This fixes it by default inheriting the privacy of what was previously
there and then becoming `true` if nothing else exits.
Closes#10545
By performing this logic very late in the build process, it ended up leading to
bugs like those found in #10973 where certain stages of the build process
expected a particular output format which didn't end up being the case. In order
to fix this, the build output generation is moved very early in the build
process to the absolute first thing in phase 2.
Closes#10973
This code in resolve accidentally forced all types with an impl to become
public. This fixes it by default inheriting the privacy of what was previously
there and then becoming `true` if nothing else exits.
Closes#10545
When a borrow occurs twice illegally, Rust will label the other borrow
as the "second borrow". This is quite confusing, as the "second borrow"
usually happened before the flagged borrow (e.g. as far as dataflow
is concerned, the first borrow is OK, the second borrow is illegal.)
This patch renames "second borrow" to "previous borrow", to make the
spatial relationship between the two borrows clearer.
Signed-off-by: Edward Z. Yang <ezyang@cs.stanford.edu>
If it's a trait method, this checks the stability attribute of the
method inside the trait definition. Otherwise, it checks the method
implementation itself.
Close#8961.
This pull request completely rewrites std::comm and all associated users. Some major bullet points
* Everything now works natively
* oneshots have been removed
* shared ports have been removed
* try_recv no longer blocks (recv_opt blocks)
* constructors are now Chan::new and SharedChan::new
* failure is propagated on send
* stream channels are 3x faster
I have acquired the following measurements on this patch. I compared against Go, but remember that Go's channels are fundamentally different than ours in that sends are by-default blocking. This means that it's not really a totally fair comparison, but it's good to see ballpark numbers for anyway
```
oneshot stream shared1
std 2.111 3.073 1.730
my 6.639 1.037 1.238
native 5.748 1.017 1.250
go8 1.774 3.575 2.948
go8-inf slow 0.837 1.376
go8-128 4.832 1.430 1.504
go1 1.528 1.439 1.251
go2 1.753 3.845 3.166
```
I had three benchmarks:
* oneshot - N times, create a "oneshot channel", send on it, then receive on it (no task spawning)
* stream - N times, send from one task to another task, wait for both to complete
* shared1 - create N threads, each of which sends M times, and a port receives N*M times.
The rows are as follows:
* `std` - the current libstd implementation (before this pull request)
* `my` - this pull request's implementation (in M:N mode)
* `native` - this pull request's implementation (in 1:1 mode)
* `goN` - go's implementation with GOMAXPROCS=N. The only relevant value is 8 (I had 8 cores on this machine)
* `goN-X` - go's implementation where the channels in question were created with buffers of size `X` to behave more similarly to rust's channels.
If it's a trait method, this checks the stability attribute of the
method inside the trait definition. Otherwise, it checks the method
implementation itself.
This PR improves the stepping experience in GDB. It contains some fine tuning of line information and makes *rustc* produce nearly the same IR/DWARF as Clang. The focus of the changes is function prologue handling which has caused some problems in the past (https://github.com/mozilla/rust/issues/9641).
It seems that GDB does not properly handle function prologues when the function uses segmented stacks, i.e. it does not recognize that the `__morestack` check is part of the prologue. When setting a breakpoint like `break foo` it will set the break point before the arguments of `foo()` have been loaded and still contain bogus values. For function with the #[no_split_stack] attribute this problem has never occurred for me so I'm pretty sure that segmented stacks are the cause of the problem. @jdm mentioned that segmented stack won't be completely abandoned after all. I'd be grateful if you could tell me about what the future might bring in this regard (@brson, @cmr).
Anyway, this PR should alleviate this problem at least in the case when setting breakpoints using line numbers and also make it less confusing when setting them via function names because then GDB will break *before* the first statement where one could conceivably argue that arguments need not be initialized yet.
Also, a koala: 🐨
Cheers,
Michael
