with_end_to_cap is enormously expensive now that it's initializing
memory since it involves 64k allocation + memset on every call. This is
most noticable when calling read_to_end on very small readers, where the
new version if **4 orders of magnitude** faster.
BufReader also depended on with_end_to_cap so I've rewritten it in its
original form.
As a bonus, converted the buffered IO struct Debug impls to use the
debug builders.
I first came across this in sfackler/rust-postgres#106 where a user reported a 10x performance regression. A call to read_to_end turned out to be the culprit: 9cd413d42c.
The new version differs from the old in a couple of ways. The buffer size used is now adaptive. It starts at 32 bytes and doubles each time EOF hasn't been reached up to a limit of 64k. In addition, the buffer is only truncated when EOF or an error has been reached, rather than after every call to read as was the case for the old implementation.
I wrote up a benchmark to compare the old version and new version: https://gist.github.com/sfackler/e979711b0ee2f2063462
It tests a couple of different cases: a high bandwidth reader, a low bandwidth reader, and a low bandwidth reader that won't return more than 10k per call to `read`. The high bandwidth reader should be analagous to use cases when reading from e.g. a `BufReader` or `Vec`, and the low bandwidth readers should be analogous to reading from something like a `TcpStream`.
Of special note, reads from a high bandwith reader containing 4 bytes are now *4,495 times faster*.
```
~/foo ❯ cargo bench
Compiling foo v0.0.1 (file:///home/sfackler/foo)
Running target/release/foo-7498d7dd7faecf5c
running 13 tests
test test_new ... ignored
test new_delay_4 ... bench: 230768 ns/iter (+/- 14812)
test new_delay_4_cap ... bench: 231421 ns/iter (+/- 7211)
test new_delay_5m ... bench: 14495370 ns/iter (+/- 4008648)
test new_delay_5m_cap ... bench: 73127954 ns/iter (+/- 59908587)
test new_nodelay_4 ... bench: 83 ns/iter (+/- 2)
test new_nodelay_5m ... bench: 12527237 ns/iter (+/- 335243)
test std_delay_4 ... bench: 373095 ns/iter (+/- 12613)
test std_delay_4_cap ... bench: 374190 ns/iter (+/- 19611)
test std_delay_5m ... bench: 17356012 ns/iter (+/- 15906588)
test std_delay_5m_cap ... bench: 883555035 ns/iter (+/- 205559857)
test std_nodelay_4 ... bench: 144937 ns/iter (+/- 2448)
test std_nodelay_5m ... bench: 16095893 ns/iter (+/- 3315116)
test result: ok. 0 passed; 0 failed; 1 ignored; 12 measured
```
r? @alexcrichton
with_end_to_cap is enormously expensive now that it's initializing
memory since it involves 64k allocation + memset on every call. This is
most noticable when calling read_to_end on very small readers, where the
new version if **4 orders of magnitude** faster.
BufReader also depended on with_end_to_cap so I've rewritten it in its
original form.
As a bonus, converted the buffered IO struct Debug impls to use the
debug builders.
Fixes#23815
Previously a panic was generated for recursive prints due to a double-borrow of
a `RefCell`. This was solved by the second borrow's output being directed
towards the global stdout instead of the per-thread stdout (still experimental
functionality).
After this functionality was altered, however, recursive prints still deadlocked
due to the overridden `write_fmt` method which locked itself first and then
wrote all the data. This was fixed by removing the override of the `write_fmt`
method. This means that unlocked usage of `write!` on a `Stdout`/`Stderr` may be
slower due to acquiring more locks, but it's easy to make more performant with a
call to `.lock()`.
Closes#23781
Now that `<[_]>::split` is an inherent method, it will trump `BufRead::split`
when `BufRead` is in scope, so there is no longer a conflict. As a result,
calling `slice.split()` will probably always give you precisely what you want!
This attribute has been deprecated in favor of #[should_panic]. This also
updates rustdoc to no longer accept the `should_fail` directive and instead
renames it to `should_panic`.
Now that `<[_]>::split` is an inherent method, it will trump `BufRead::split`
when `BufRead` is in scope, so there is no longer a conflict. As a result,
calling `slice.split()` will probably always give you precisely what you want!
This attribute has been deprecated in favor of #[should_panic]. This also
updates rustdoc to no longer accept the `should_fail` directive and instead
renames it to `should_panic`.
Main motivation was to update docs for the removal or "demotion" of certain extension traits. The update to the slice docs was larger, since the text was largely outdated.
Reject specialized Drop impls.
See Issue #8142 for discussion.
This makes it illegal for a Drop impl to be more specialized than the original item.
So for example, all of the following are now rejected (when they would have been blindly accepted before):
```rust
struct S<A> { ... };
impl Drop for S<i8> { ... } // error: specialized to concrete type
struct T<'a> { ... };
impl Drop for T<'static> { ... } // error: specialized to concrete region
struct U<A> { ... };
impl<A:Clone> Drop for U<A> { ... } // error: added extra type requirement
struct V<'a,'b>;
impl<'a,'b:a> Drop for V<'a,'b> { ... } // error: added extra region requirement
```
Due to examples like the above, this is a [breaking-change].
(The fix is to either remove the specialization from the `Drop` impl, or to transcribe the requirements into the struct/enum definition; examples of both are shown in the PR's fixed to `libstd`.)
----
This is likely to be the last thing blocking the removal of the `#[unsafe_destructor]` attribute.
Fix#8142Fix#23584
This commit alters the behavior of the `Read::read_to_end()` method to zero all
memory instead of passing an uninitialized buffer to `read`. This change is
motivated by the [discussion on the internals forum][discuss] where the
conclusion has been that the standard library will not expose uninitialized
memory.
[discuss]: http://internals.rust-lang.org/t/uninitialized-memory/1652Closes#20314
There have been some recent panics on the bots and this commit is an attempt to
appease them. Previously it was considered invalid to run `rt::at_exit` after
the handlers had already started running. Due to the multithreaded nature of
applications, however, it is not always possible to guarantee this. For example
[this program][ex] will show off the abort.
[ex]: https://gist.github.com/alexcrichton/56300b87af6fa554e52d
The semantics of the `rt::at_exit` function have been modified as such:
* It is now legal to call `rt::at_exit` at any time. The return value now
indicates whether the closure was successfully registered or not. Callers must
now decide what to do with this information.
* The `rt::at_exit` handlers will now be run for a fixed number of iterations.
Common cases (such as the example shown) may end up registering a new handler
while others are running perhaps once or twice, so this common condition is
covered by re-running the handlers a fixed number of times, after which new
registrations are forbidden.
Some usage of `rt::at_exit` was updated to handle these new semantics, but
deprecated or unstable libraries calling `rt::at_exit` were not updated.
This commit alters the behavior of the `Read::read_to_end()` method to zero all
memory instead of passing an uninitialized buffer to `read`. This change is
motivated by the [discussion on the internals forum][discuss] where the
conclusion has been that the standard library will not expose uninitialized
memory.
[discuss]: http://internals.rust-lang.org/t/uninitialized-memory/1652Closes#20314
This small commit stabilizes the `Error` trait as-is, except that `Send`
and `Debug` are added as constraints. The `Send` constraint is because
most uses of `Error` will be for trait objects, and by default we would
like these objects to be transferrable between threads. The `Debug`
constraint is to ensure that e.g. `Box<Error>` is `Debug`, and because
types that implement `Display` should certainly implement `Debug` in any case.
In the near future we expect to add `Any`-like downcasting features to
`Error`, but this is waiting on some additional
mechanisms (`Reflect`). It will be added before 1.0 via default methods.
[breaking-change]
r? @alexcrichton
Closes#21790
This small commit stabilizes the `Error` trait as-is, except that `Send`
and `Debug` are added as constraints. The `Send` constraint is because
most uses of `Error` will be for trait objects, and by default we would
like these objects to be transferrable between threads. The `Debug`
constraint is to ensure that e.g. `Box<Error>` is `Debug`, and because
types that implement `Display` should certainly implement `Debug` in any case.
In the near future we expect to add `Any`-like downcasting features to
`Error`, but this is waiting on some additional
mechanisms (`Reflect`). It will be added before 1.0 via default methods.
[breaking-change]
There have been some recent panics on the bots and this commit is an attempt to
appease them. Previously it was considered invalid to run `rt::at_exit` after
the handlers had already started running. Due to the multithreaded nature of
applications, however, it is not always possible to guarantee this. For example
[this program][ex] will show off the abort.
[ex]: https://gist.github.com/alexcrichton/56300b87af6fa554e52d
The semantics of the `rt::at_exit` function have been modified as such:
* It is now legal to call `rt::at_exit` at any time. The return value now
indicates whether the closure was successfully registered or not. Callers must
now decide what to do with this information.
* The `rt::at_exit` handlers will now be run for a fixed number of iterations.
Common cases (such as the example shown) may end up registering a new handler
while others are running perhaps once or twice, so this common condition is
covered by re-running the handlers a fixed number of times, after which new
registrations are forbidden.
Some usage of `rt::at_exit` was updated to handle these new semantics, but
deprecated or unstable libraries calling `rt::at_exit` were not updated.
This commit removes the reexports of `old_io` traits as well as `old_path` types
and traits from the prelude. This functionality is now all deprecated and needs
to be removed to make way for other functionality like `Seek` in the `std::io`
module (currently reexported as `NewSeek` in the io prelude).
Closes#23377Closes#23378
This commit removes the reexports of `old_io` traits as well as `old_path` types
and traits from the prelude. This functionality is now all deprecated and needs
to be removed to make way for other functionality like `Seek` in the `std::io`
module (currently reexported as `NewSeek` in the io prelude).
Closes#23377Closes#23378
This commit stabilizes the `ErrorKind` enumeration which is consumed by and
generated by the `io::Error` type. The purpose of this type is to serve as a
cross-platform namespace to categorize errors into. Two specific issues are
addressed as part of this stablization:
* The naming of each variant was scrutinized and some were tweaked. An example
is how `FileNotFound` was renamed to simply `NotFound`. These names should not
show either a Unix or Windows bias and the set of names is intended to grow
over time. For now the names will likely largely consist of those errors
generated by the I/O APIs in the standard library.
* The mapping of OS error codes onto kinds has been altered. Coalescing no
longer occurs (multiple error codes become one kind). It is intended that each
OS error code, if bound, corresponds to only one `ErrorKind`. The current set
of error kinds was expanded slightly to include some networking errors.
This commit also adds a `raw_os_error` function which returns an `Option<i32>`
to extract the underlying raw error code from the `Error`.
Closes#16666
[breaking-change]
This commit stabilizes the `ErrorKind` enumeration which is consumed by and
generated by the `io::Error` type. The purpose of this type is to serve as a
cross-platform namespace to categorize errors into. Two specific issues are
addressed as part of this stablization:
* The naming of each variant was scrutinized and some were tweaked. An example
is how `FileNotFound` was renamed to simply `NotFound`. These names should not
show either a Unix or Windows bias and the set of names is intended to grow
over time. For now the names will likely largely consist of those errors
generated by the I/O APIs in the standard library.
* The mapping of OS error codes onto kinds has been altered. Coalescing no
longer occurs (multiple error codes become one kind). It is intended that each
OS error code, if bound, corresponds to only one `ErrorKind`. The current set
of error kinds was expanded slightly to include some networking errors.
This commit also adds a `raw_os_error` function which returns an `Option<i32>`
to extract the underlying raw error code from the `Error`.
Nothing inside of the read/write interface itself can panic, so any
poison must have been the result of user code which the lock isn't
protecting.
This seems safe to me, but if we don't want to go this route we should update the docs to indicate that these methods can panic.
r? @alexcrichton
The [associated RFC][rfc] for possibly splitting out `flush` has been closed and
as a result there are no more blockers for stabilizing this method, so this
commit marks the method as such.
[rfc]: https://github.com/rust-lang/rfcs/pull/950
