The code was using (in the notation of Doornik 2005) `f(x_{i+1}) -
f(x_{i+2})` rather than `f(x_i) - f(x_{i+1})`. This corrects that, and
removes the F_DIFF tables which caused this problem in the first place.
They `F_DIFF` tables are a micro-optimisation (in theory, they could
easily be a micro-pessimisation): that `if` gets hit about 1% of the
time for Exp/Normal, and the rest of the condition involves RNG calls
and a floating point `exp`, so it is unlikely that saving a single FP
subtraction will be very useful (especially as more tables means more
memory reads and higher cache pressure, as well as taking up space in
the binary (although only ~2k in this case)).
Closes#10084. Notably, unlike that issue suggests, this wasn't a
problem with the Exp tables. It affected Normal too, but since it is
symmetric, there was no bias in the mean (as the bias was equal on the
positive and negative sides and so cancelled out) but it was visible as
a variance slightly lower than it should be.
New plot:
I've started writing some tests in [huonw/random-tests](https://github.com/huonw/random-tests) (not in the main repo because they can and do fail occasionally, due to randomness, but it is on Travis and Rust-CI so it will hopefully track the language), unsurprisingly, they're [currently failing](https://travis-ci.org/huonw/random-tests/builds/13313987) (note that both exp and norm are failing, the former due to both mean and variance the latter due to just variance), but pass at the 0.01 level reliably with this change.
(Currently the only test is essentially a quantitative version of the plots I've been showing, which is run on the `f64` `Rand` instance (uniform 0 to 1), and the Normal and Exp distributions.)
The previous method was unsound because you could very easily create two mutable
pointers which alias the same location (not sound behavior). This hides the
function which does so and then exports an explicit flush() function (with
documentation about how it works).
Fix the implementation of `std::rand::Rng::fill_bytes()` for
`std::rand::reseeding::ReseedingRng` to call the `fill_bytes()` method
of the underlying RNG rather than itself, which causes infinite
recursion.
Fixes#10202.
The code was using (in the notation of Doornik 2005) `f(x_{i+1}) -
f(x_{i+2})` rather than `f(x_i) - f(x_{i+1})`. This corrects that, and
removes the F_DIFF tables which caused this problem in the first place.
They `F_DIFF` tables are a micro-optimisation (in theory, they could
easily be a micro-pessimisation): that `if` gets hit about 1% of the
time for Exp/Normal, and the rest of the condition involves RNG calls
and a floating point `exp`, so it is unlikely that saving a single FP
subtraction will be very useful (especially as more tables means more
memory reads and higher cache pressure, as well as taking up space in
the binary (although only ~2k in this case)).
Closes#10084. Notably, unlike that issue suggests, this wasn't a
problem with the Exp tables. It affected Normal too, but since it is
symmetric, there was no bias in the mean (as the bias was equal on the
positive and negative sides and so cancelled out) but it was visible as
a variance slightly lower than it should be.
- `begin_unwind` and `fail!` is now generic over any `T: Any + Send`.
- Every value you fail with gets boxed as an `~Any`.
- Because of implementation issues, `&'static str` and `~str` are still
handled specially behind the scenes.
- Changed the big macro source string in libsyntax to a raw string
literal, and enabled doc comments there.
- `begin_unwind` is now generic over any `T: Any + Send`.
- Every value you fail with gets boxed as an `~Any`.
- Because of implementation details, `&'static str` and `~str` are still
handled specially behind the scenes.
- Changed the big macro source string in libsyntax to a raw string
literal, and enabled doc comments there.
Allows an enum with a discriminant to use any of the primitive integer types to store it. By default the smallest usable type is chosen, but this can be overridden with an attribute: `#[repr(int)]` etc., or `#[repr(C)]` to match the target's C ABI for the equivalent C enum.
Also adds a lint pass for using non-FFI safe enums in extern declarations, checks that specified discriminants can be stored in the specified type if any, and fixes assorted code that was assuming int.
This is one of the final steps needed to complete #9128. It still needs a little bit of polish before closing that issue, but it's in a pretty much "done" state now.
The idea here is that the entire event loop implementation using libuv is now housed in `librustuv` as a completely separate library. This library is then injected (via `extern mod rustv`) into executable builds (similarly to how libstd is injected, tunable via `#[no_uv]`) to bring in the "rust blessed event loop implementation."
Codegen-wise, there is a new `event_loop_factory` language item which is tagged on a function with 0 arguments returning `~EventLoop`. This function's symbol is then inserted into the crate map for an executable crate, and if there is no definition of the `event_loop_factory` language item then the value is null.
What this means is that embedding rust as a library in another language just got a little harder. Libraries don't have crate maps, which means that there's no way to find the event loop implementation to spin up the runtime. That being said, it's always possible to build the runtime manually. This request also makes more runtime components public which should probably be public anyway. This new public-ness should allow custom scheduler setups everywhere regardless of whether you follow the `rt::start `path.
Otherwise, run-pass/deriving-primitive.rs breaks on 32-bit platforms,
because `int::min_value` is `0xffffffff7fffffff` when evaluated for the
discriminant declaration.
Not only can discriminants be smaller than int now, but they can be
larger than int on 32-bit targets. This has obvious implications for the
reflection interface. Without this change, things fail with LLVM
assertions when we try to "extend" i64 to i32.
There are a few reasons that this is a desirable move to take:
1. Proof of concept that a third party event loop is possible
2. Clear separation of responsibility between rt::io and the uv-backend
3. Enforce in the future that the event loop is "pluggable" and replacable
Here's a quick summary of the points of this pull request which make this
possible:
* Two new lang items were introduced: event_loop, and event_loop_factory.
The idea of a "factory" is to define a function which can be called with no
arguments and will return the new event loop as a trait object. This factory
is emitted to the crate map when building an executable. The factory doesn't
have to exist, and when it doesn't then an empty slot is in the crate map and
a basic event loop with no I/O support is provided to the runtime.
* When building an executable, then the rustuv crate will be linked by default
(providing a default implementation of the event loop) via a similar method to
injecting a dependency on libstd. This is currently the only location where
the rustuv crate is ever linked.
* There is a new #[no_uv] attribute (implied by #[no_std]) which denies
implicitly linking to rustuv by default
Closes#5019
Primarily this makes the Scheduler and all of its related interfaces public. The
reason for doing this is that currently any extern event loops had no access to
the scheduler at all. This allows third-party event loops to manipulate the
scheduler, along with allowing the uv event loop to live inside of its own
crate.
This drops more of the old C++ runtime to rather be written in rust. A few
features were lost along the way, but hopefully not too many. The main loss is
that there are no longer backtraces associated with allocations (rust doesn't
have a way of acquiring those just yet). Other than that though, I believe that
the rest of the debugging utilities made their way over into rust.
Closes#8704
Some code cleanup, sorting of import blocks
Removed std::unstable::UnsafeArc's use of Either
Added run-fail tests for the new FailWithCause impls
Changed future_result and try to return Result<(), ~Any>.
- Internally, there is an enum of possible fail messages passend around.
- In case of linked failure or a string message, the ~Any gets
lazyly allocated in future_results recv method.
- For that, future result now returns a wrapper around a Port.
- Moved and renamed task::TaskResult into rt::task::UnwindResult
and made it an internal enum.
- Introduced a replacement typedef `type TaskResult = Result<(), ~Any>`.
Remove the Sha1, Sha2, MD5, and MD4 algorithms. SipHash is also cryptographically secure hash function and IsaacRng is a cryptographically secure RNG - I left those alone but removed comments that implied they were suitable for cryptographic use. I thought that MD4 was used for something by the compiler, but everything still seems to work with it removed, so, I guess not.
One thing that I'm not sure about - workcache.rs and workcache_support.rs (in librustpkg) both depend on Sha1. Without Sha1, the only hash function left is SipHash, so I switched that code over to use SipHash. The output size of SipHash is only 64-bits, however - much less than 160 for Sha1. I'm not sure this is a problem. Without other cryptographic hashes in the tree, I'm not sure what else to do. I considered moved Sha1 into librustpkg, but I don't know if that makes sense.
If merged, this closes#9300.
I'm not entirely sure why this is happening, but the server task is never seeing
the second send of the client task, and this test will very reliably fail to
complete on windows.
It was pretty much a miracle that these tests were ever passing. They would
never have passed in the single threaded case because only one sigint in the
tests is ever generated, but when run in parallel two sigints will be generated.
This drops more of the old C++ runtime to rather be written in rust. A few
features were lost along the way, but hopefully not too many. The main loss is
that there are no longer backtraces associated with allocations (rust doesn't
have a way of acquiring those just yet). Other than that though, I believe that
the rest of the debugging utilities made their way over into rust.
Closes#8704
This optimizes the `home_for_io` code path by requiring fewer scheduler
operations in some situtations.
When moving to your home scheduler, this no longer forces a context switch if
you're already on the home scheduler. Instead, the homing code now simply pins
you to your current scheduler (making it so you can't be stolen away). If you're
not on your home scheduler, then we context switch away, sending you to your
home scheduler.
When the I/O operation is done, then we also no longer forcibly trigger a
context switch. Instead, the action is cased on whether the task is homed or
not. If a task does not have a home, then the task is re-flagged as not having a
home and no context switch is performed. If a task is homed to the current
scheduler, then we don't do anything, and if the task is homed to a foreign
scheduler, then it's sent along its merry way.
I verified that there are about a third as many `write` syscalls done in print
operations now. Libuv uses write to implement async handles, and the homing
before and after each I/O operation was triggering a write on these async
handles. Additionally, using the terrible benchmark of printing 10k times in a
loop, this drives the runtime from 0.6s down to 0.3s (yay!).
This optimizes the `home_for_io` code path by requiring fewer scheduler
operations in some situtations.
When moving to your home scheduler, this no longer forces a context switch if
you're already on the home scheduler. Instead, the homing code now simply pins
you to your current scheduler (making it so you can't be stolen away). If you're
not on your home scheduler, then we context switch away, sending you to your
home scheduler.
When the I/O operation is done, then we also no longer forcibly trigger a
context switch. Instead, the action is cased on whether the task is homed or
not. If a task does not have a home, then the task is re-flagged as not having a
home and no context switch is performed. If a task is homed to the current
scheduler, then we don't do anything, and if the task is homed to a foreign
scheduler, then it's sent along its merry way.
I verified that there are about a third as many `write` syscalls done in print
operations now. Libuv uses write to implement async handles, and the homing
before and after each I/O operation was triggering a write on these async
handles. Additionally, using the terrible benchmark of printing 10k times in a
loop, this drives the runtime from 0.6s down to 0.3s (yay!).
Almost all languages provide some form of buffering of the stdout stream, and
this commit adds this feature for rust. A handle to stdout is lazily initialized
in the Task structure as a buffered owned Writer trait object. The buffer
behavior depends on where stdout is directed to. Like C, this line-buffers the
stream when the output goes to a terminal (flushes on newlines), and also like C
this uses a fixed-size buffer when output is not directed at a terminal.
We may decide the fixed-size buffering is overkill, but it certainly does reduce
write syscall counts when piping output elsewhere. This is a *huge* benefit to
any code using logging macros or the printing macros. Formatting emits calls to
`write` very frequently, and to have each of them backed by a write syscall was
very expensive.
In a local benchmark of printing 10000 lines of "what" to stdout, I got the
following timings:
when | terminal | redirected
----------|---------------|--------
before | 0.575s | 0.525s
after | 0.197s | 0.013s
C | 0.019s | 0.004s
I can also confirm that we're buffering the output appropriately in both
situtations. We're still far slower than C, but I believe much of that has to do
with the "homing" that all tasks due, we're still performing an order of
magnitude more write syscalls than C does.
Almost all languages provide some form of buffering of the stdout stream, and
this commit adds this feature for rust. A handle to stdout is lazily initialized
in the Task structure as a buffered owned Writer trait object. The buffer
behavior depends on where stdout is directed to. Like C, this line-buffers the
stream when the output goes to a terminal (flushes on newlines), and also like C
this uses a fixed-size buffer when output is not directed at a terminal.
We may decide the fixed-size buffering is overkill, but it certainly does reduce
write syscall counts when piping output elsewhere. This is a *huge* benefit to
any code using logging macros or the printing macros. Formatting emits calls to
`write` very frequently, and to have each of them backed by a write syscall was
very expensive.
In a local benchmark of printing 10000 lines of "what" to stdout, I got the
following timings:
when | terminal | redirected
----------------------------------
before | 0.575s | 0.525s
after | 0.197s | 0.013s
C | 0.019s | 0.004s
I can also confirm that we're buffering the output appropriately in both
situtations. We're still far slower than C, but I believe much of that has to do
with the "homing" that all tasks due, we're still performing an order of
magnitude more write syscalls than C does.
It's not guaranteed that there will always be an event loop to run, and this
implementation will serve as an incredibly basic one which does not provide any
I/O, but allows the scheduler to still run.
cc #9128
This is a peculiar function to require event loops to implement, and it's only
used in one spot during tests right now. Instead, a possibly more robust apis
for timers should be used rather than requiring all event loops to implement a
curious-looking function.
The PausibleIdleCallback must have some handle into the event loop, and because
struct destructors are run in order of top-to-bottom in order of fields, this
meant that the event loop was getting destroyed before the idle callback was
getting destroyed.
I can't confirm that this fixes a problem in how we use libuv, but it does
semantically fix a problem for usage with other event loops.
This adds constructors to pipe streams in the new runtime to take ownership of
file descriptors, and also fixes a few tests relating to the std::run changes
(new errors are raised on io_error and one test is xfail'd).
I was seeing a lot of weird behavior with stdin behaving as a tty, and it
doesn't really quite make sense, so instead this moves to using libuv's pipes
instead (which make more sense for stdin specifically).
This prevents piping input to rustc hanging forever.
The general idea is to remove conditions completely from I/O, so in the meantime
remove the read_error condition to mean the same thing as the io_error condition.
The isn't an ideal patch, and the comment why is in the code. Basically uvio
uses task::unkillable which touches the kill flag for a task, and if the task is
failing due to mismangement of the kill flag, then there will be serious
problems when the task tries to print that it's failing.
When uv's TTY I/O is used for the stdio streams, the file descriptors are put
into a non-blocking mode. This means that other concurrent writes to the same
stream can fail with EAGAIN or EWOULDBLOCK. By all I/O to event-loop I/O, we
avoid this error.
There is one location which cannot move, which is the runtime's dumb_println
function. This was implemented to handle the EAGAIN and EWOULDBLOCK errors and
simply retry again and again.
This involved changing a fair amount of code, rooted in how we access the local
IoFactory instance. I added a helper method to the rtio module to access the
optional local IoFactory. This is different than before in which it was assumed
that a local IoFactory was *always* present. Now, a separate io_error is raised
when an IoFactory is not present, yet I/O is requested.
This removes the PathLike trait associated with this "support module". This is
yet another "container of bytes" trait, so I didn't want to duplicate what
already exists throughout libstd. In actuality, we're going to pass of C strings
to the libuv APIs, so instead the arguments are now bound with the 'ToCStr'
trait instead.
Additionally, a layer of complexity was removed by immediately converting these
type-generic parameters into CStrings to get handed off to libuv apis.
We get a little more functionality from libuv for these kinds of streams (things
like terminal dimentions), and it also appears to more gracefully handle the
stream being a window. Beforehand, if you used stdio and hit CTRL+d on a
process, libuv would continually return 0-length successful reads instead of
interpreting that the stream was closed.
I was hoping to be able to write tests for this, but currently the testing
infrastructure doesn't allow tests with a stdin and a stdout, but this has been
manually tested! (not that it means much)
Removed unused import warning in std::mem and cleaned it up too
Removed is_true and is_false from std::bool
Removed freestanding functions in std::bool
- Adds the `Sample` and `IndependentSample` traits for generating numbers where there are parameters (e.g. a list of elements to draw from, or the mean/variance of a normal distribution). The former takes `&mut self` and the latter takes `&self` (this is the only difference).
- Adds proper `Normal` and `Exp`-onential distributions
- Adds `Range` which generates `[lo, hi)` generically & properly (via a new trait) replacing the incorrect behaviour of `Rng.gen_integer_range` (this has become `Rng.gen_range` for convenience, it's far more efficient to use `Range` itself)
- Move the `Weighted` struct from `std::rand` to `std::rand::distributions` & improve it
- optimisations and docs
- Use ["nothing up my sleeve numbers"](http://en.wikipedia.org/wiki/Nothing_up_my_sleeve_number) for the ISAAC tests.
- Replace the default implementation of `Rng.fill_bytes` with something that doesn't try to do bad things with `transmute` and vectors just for the sake of a little speed.
- Replace the transmutes used to seed the ISAAC RNGs with calls into `vec::raw`.
Slice transmutes are now (and, really, always were) dangerous, so we
avoid them and do the (only?) non-(undefined behaviour in C) pointer
cast: casting to *u8.
I'm planning on doing more updates, but the section in the tutorial stood out at me since the 'rust' tool no longer exists, this should probably be removed to lessen confusion.
This reifies the computations required for uniformity done by
(the old) `Rng.gen_integer_range` (now Rng.gen_range), so that they can
be amortised over many invocations, if it is called in a loop.
Also, it makes it correct, but using a trait + impls for each type,
rather than trying to coerce `Int` + `u64` to do the right thing. This
also makes it more extensible, e.g. big integers could & should
implement SampleRange.
Complete the implementation of Exp and Normal started by Exp1 and
StandardNormal by creating types implementing Sample & IndependentSample
with the appropriate parameters.
This should close#9468.
I removed the test stating that nested comments should not be implemented.
I had a little chicken-and-egg problem because a comment of the std contains "/*", and adding support for nested comment creates a backward incompatibility in that case, so I had to use a dirty hack to get stage1 and stage2 to compile. This part should be revert when this commit lands in a snapshot.
This is my first non-typo contribution, so I'm open to any comment.
This commit re-introduces the functionality of __morestack in a way that it was
not originally anticipated. Rust does not currently have segmented stacks,
rather just large stack segments. We do not detect when these stack segments are
overrun currently, but this commit leverages __morestack in order to check this.
This commit purges a lot of the old __morestack and stack limit C++
functionality, migrating the necessary chunks to rust. The stack limit is now
entirely maintained in rust, and the "main logic bits" of __morestack are now
also implemented in rust as well.
I put my best effort into validating that this currently builds and runs successfully on osx and linux 32/64 bit, but I was unable to get this working on windows. We never did have unwinding through __morestack frames, and although I tried poking at it for a bit, I was unable to understand why we don't get unwinding right now.
A focus of this commit is to implement as much of the logic in rust as possible. This involved some liberal usage of `no_split_stack` in various locations, along with some use of the `asm!` macro (scary). I modified a bit of C++ to stop calling `record_sp_limit` because this is no longer defined in C++, rather in rust.
Another consequence of this commit is that `thread_local_storage::{get, set}` must both be flagged with `#[rust_stack]`. I've briefly looked at the implementations on osx/linux/windows to ensure that they're pretty small stacks, and I'm pretty sure that they're definitely less than 20K stacks, so we probably don't have a lot to worry about.
Other things worthy of note:
* The default stack size is now 4MB instead of 2MB. This is so that when we request 2MB to call a C function you don't immediately overflow because you have consumed any stack at all.
* `asm!` is actually pretty cool, maybe we could actually define context switching with it?
* I wanted to add links to the internet about all this jazz of storing information in TLS, but I was only able to find a link for the windows implementation. Otherwise my suggestion is just "disassemble on that arch and see what happens"
* I put my best effort forward on arm/mips to tweak __morestack correctly, we have no ability to test this so an extra set of eyes would be useful on these spots.
* This is all really tricky stuff, so I tried to put as many comments as I thought were necessary, but if anything is still unclear (or I completely forgot to take something into account), I'm willing to write more!
This commit resumes management of the stack boundaries and limits when switching
between tasks. This additionally leverages the __morestack function to run code
on "stack overflow". The current behavior is to abort the process, but this is
probably not the best behavior in the long term (for deails, see the comment I
wrote up in the stack exhaustion routine).
d4a32386f3 broke these since slice_to() and slice_from() must get character
boundaries, and arbitrary needle lengths don't necessarily map to character
boundaries of the haystack.
This also adds new tests that would have caught this bug.
d4a32386f3 broke these since slice_to() and slice_from() must get character
boundaries, and arbitrary needle lengths don't necessarily map to character
boundaries of the haystack.
This also adds new tests that would have caught this bug.
Hello,
First time rust contributor here, please let me know if I need to sort out the contribution agreement for this.
I picked issue #9755 to dip my toe in the water, this pull request isn't quite complete though as I have not updated the documentation. The reason for this is that I haven't tracked down why this feature is gated so I don't feel I can write a justification of the same quality as the other features have been documented.
If someone would like to explain or point me at a mail thread I am happy to update with this change.
Hopefully I have understood the process of converting the old flag into a directive correctly.
Also just to call out what I am sure if a known quirk when adding feature directives, you can't build this code unless you have a snapshot of the compiler which knows about the feature directive. Chicken and the egg. I split the change into two commits, the first should be able to build a snapshot that can compile the second.
std::iter: Introduce .by_ref() adaptor
Creates a wrapper around a mutable reference to the iterator.
This is useful to allow applying iterator adaptors while still
retaining ownership of the original iterator value.
Example::
let mut xs = range(0, 10);
// sum the first five values
let partial_sum = xs.by_ref().take(5).fold(0, |a, b| a + b);
assert!(partial_sum == 10);
// xs.next() is now `5`
assert!(xs.next() == Some(5));
---
This adaptor requires the user to have good understanding of
iterators or what a particular adaptor does. There could be some
pitfalls here with the iterator protocol, it's mostly the same issues
as other places regarding what happens after the iterator
returns None for the first time.
There could also be other ways to achieve the same thing, for
example Implementing iterator on `&mut T` itself:
`impl <T: Iterator<..>> Iterator for &mut T` but that would only
lead to confusion I think.
The goal here is to avoid requiring a division or multiplication to compare against the length. The bounds check previously used an incorrect micro-optimization to replace the division by a multiplication, but now neither is necessary *for slices*. Unique/managed vectors will have to do a division to get the length until they are reworked/replaced.
Rewrite the entire `std::path` module from scratch.
`PosixPath` is now based on `~[u8]`, which fixes#7225.
Unnecessary allocation has been eliminated.
There are a lot of clients of `Path` that still assume utf-8 paths.
This is covered in #9639.
Delete the following API functions:
- set_dirname()
- with_dirname()
- set_filestem()
- with_filestem()
- add_extension()
- file_path()
Also change pop() to return a boolean instead of an owned copy of the
old filename.
Standardize the is_sep() functions to be the same in both posix and
windows, and re-export from path. Update extra::glob to use this.
Remove the usage of either, as it's going away.
Move the WindowsPath-specific methods out of WindowsPath and make them
top-level functions of path::windows instead. This way you cannot
accidentally write code that will fail to compile on non-windows
architectures without typing ::windows anywhere.
Remove GenericPath::from_c_str() and just impl BytesContainer for
CString instead.
Remove .join_path() and .push_path() and just implement BytesContainer
for Path instead.
Remove FilenameDisplay and add a boolean flag to Display instead.
Remove .each_parent(). It only had one caller, so just inline its
definition there.
...al work
This is causing really awful scheduler behavior where the main thread scheduler is
continually waking up, stealing work, discovering it can't actually run the work,
and sending it off to another scheduler.
No test cases because we don't have suitable instrumentation for it.
* Allow named parameters to specify width/precision
* Intepret the format string '0$' as "width is the 0th argument" instead of
thinking the lone '0' was the sign-aware-zero-padding flag. To get both you'd
need to put '00$' which makes more sense if you want both to happen.
Closes#9669
Rewrite these methods as methods on Display and FilenameDisplay. This
turns
do path.with_display_str |s| { ... }
into
do path.display().with_str |s| { ... }
Add a new trait BytesContainer that is implemented for both byte vectors
and strings.
Convert Path::from_vec and ::from_str to one function, Path::new().
Remove all the _str-suffixed mutation methods (push, join, with_*,
set_*) and modify the non-suffixed versions to use BytesContainer.
Remove the old path.
Rename path2 to path.
Update all clients for the new path.
Also make some miscellaneous changes to the Path APIs to help the
adoption process.
These methods return an object that can be formatted using {} to print
display strings.
Path itself does not implement fmt::Default to avoid accidental usage of
display strings in incorrect places (e.g. process arguments).
These functions are for working with a string representation of the path
even if it's not UTF-8 encoded. They replace invalid UTF-8 sequences
with the replacement char.
