Summary:
So far, we've used the term POD "Plain Old Data" to refer to types that
can be safely copied. However, this term is not consistent with the
other built-in bounds that use verbs instead. This patch renames the Pod
kind into Copy.
RFC: 0003-opt-in-builtin-traits
Test Plan: make check
Reviewers: cmr
Differential Revision: http://phabricator.octayn.net/D3
std: remove the `equals` method from `TotalEq`.
`TotalEq` is now just an assertion about the `Eq` impl of a
type (i.e. `==` is a total equality if a type implements `TotalEq`) so
the extra method is just confusing.
Also, a new method magically appeared as a hack to allow deriving to
assert that the contents of a struct/enum are also TotalEq, because the
deriving infrastructure makes it very hard to do anything but create a
trait method. (You didn't hear about this horrible work-around from me
:(.)
`TotalEq` is now just an assertion about the `Eq` impl of a
type (i.e. `==` is a total equality if a type implements `TotalEq`) so
the extra method is just confusing.
Also, a new method magically appeared as a hack to allow deriving to
assert that the contents of a struct/enum are also TotalEq, because the
deriving infrastructure makes it very hard to do anything but create a
trait method. (You didn't hear about this horrible work-around from me
:(.)
This commit moves all logging out of the standard library into an external
crate. This crate is the new crate which is responsible for all logging macros
and logging implementation. A few reasons for this change are:
* The crate map has always been a bit of a code smell among rust programs. It
has difficulty being loaded on almost all platforms, and it's used almost
exclusively for logging and only logging. Removing the crate map is one of the
end goals of this movement.
* The compiler has a fair bit of special support for logging. It has the
__log_level() expression as well as generating a global word per module
specifying the log level. This is unfairly favoring the built-in logging
system, and is much better done purely in libraries instead of the compiler
itself.
* Initialization of logging is much easier to do if there is no reliance on a
magical crate map being available to set module log levels.
* If the logging library can be written outside of the standard library, there's
no reason that it shouldn't be. It's likely that we're not going to build the
highest quality logging library of all time, so third-party libraries should
be able to provide just as high-quality logging systems as the default one
provided in the rust distribution.
With a migration such as this, the change does not come for free. There are some
subtle changes in the behavior of liblog vs the previous logging macros:
* The core change of this migration is that there is no longer a physical
log-level per module. This concept is still emulated (it is quite useful), but
there is now only a global log level, not a local one. This global log level
is a reflection of the maximum of all log levels specified. The previously
generated logging code looked like:
if specified_level <= __module_log_level() {
println!(...)
}
The newly generated code looks like:
if specified_level <= ::log::LOG_LEVEL {
if ::log::module_enabled(module_path!()) {
println!(...)
}
}
Notably, the first layer of checking is still intended to be "super fast" in
that it's just a load of a global word and a compare. The second layer of
checking is executed to determine if the current module does indeed have
logging turned on.
This means that if any module has a debug log level turned on, all modules
with debug log levels get a little bit slower (they all do more expensive
dynamic checks to determine if they're turned on or not).
Semantically, this migration brings no change in this respect, but
runtime-wise, this will have a perf impact on some code.
* A `RUST_LOG=::help` directive will no longer print out a list of all modules
that can be logged. This is because the crate map will no longer specify the
log levels of all modules, so the list of modules is not known. Additionally,
warnings can no longer be provided if a malformed logging directive was
supplied.
The new "hello world" for logging looks like:
#[phase(syntax, link)]
extern crate log;
fn main() {
debug!("Hello, world!");
}
Partially addresses #11783.
Previously, rust's hashtable was totally unoptimized. It used an Option
per key-value pair, and used very naive open allocation.
The old hashtable had very high variance in lookup time. For an example,
see the 'find_nonexisting' benchmark below. This is fixed by keys in
'lucky' spots with a low probe sequence length getting their good spots
stolen by keys with long probe sequence lengths. This reduces hashtable
probe length variance, while maintaining the same mean.
Also, other optimization liberties were taken. Everything is as cache
aware as possible, and this hashtable should perform extremely well for
both large and small keys and values.
Benchmarks:
```
comprehensive_old_hashmap 378 ns/iter (+/- 8)
comprehensive_new_hashmap 206 ns/iter (+/- 4)
1.8x faster
old_hashmap_as_queue 238 ns/iter (+/- 8)
new_hashmap_as_queue 119 ns/iter (+/- 2)
2x faster
old_hashmap_insert 172 ns/iter (+/- 8)
new_hashmap_insert 146 ns/iter (+/- 11)
1.17x faster
old_hashmap_find_existing 50 ns/iter (+/- 12)
new_hashmap_find_existing 35 ns/iter (+/- 6)
1.43x faster
old_hashmap_find_notexisting 49 ns/iter (+/- 49)
new_hashmap_find_notexisting 34 ns/iter (+/- 4)
1.44x faster
Memory usage of old hashtable (64-bit assumed):
aligned(8+sizeof(Option)+sizeof(K)+sizeof(V))/0.75 + 48ish bytes
Memory usage of new hashtable:
(aligned(sizeof(K))
+ aligned(sizeof(V))
+ 8)/0.9 + 112ish bytes
Timing of building librustc:
compile_and_link: x86_64-unknown-linux-gnu/stage0/lib/rustlib/x86_64-unknown-linux-gnu/lib/librustc
time: 0.457 s parsing
time: 0.028 s gated feature checking
time: 0.000 s crate injection
time: 0.108 s configuration 1
time: 1.049 s expansion
time: 0.219 s configuration 2
time: 0.222 s maybe building test harness
time: 0.223 s prelude injection
time: 0.268 s assinging node ids and indexing ast
time: 0.075 s external crate/lib resolution
time: 0.026 s language item collection
time: 1.016 s resolution
time: 0.038 s lifetime resolution
time: 0.000 s looking for entry point
time: 0.030 s looking for macro registrar
time: 0.061 s freevar finding
time: 0.138 s region resolution
time: 0.110 s type collecting
time: 0.072 s variance inference
time: 0.126 s coherence checking
time: 9.110 s type checking
time: 0.186 s const marking
time: 0.049 s const checking
time: 0.418 s privacy checking
time: 0.057 s effect checking
time: 0.033 s loop checking
time: 1.293 s compute moves
time: 0.182 s match checking
time: 0.242 s liveness checking
time: 0.866 s borrow checking
time: 0.150 s kind checking
time: 0.013 s reachability checking
time: 0.175 s death checking
time: 0.461 s lint checking
time: 13.112 s translation
time: 4.352 s llvm function passes
time: 96.702 s llvm module passes
time: 50.574 s codegen passes
time: 154.611 s LLVM passes
time: 2.821 s running linker
time: 15.750 s linking
compile_and_link: x86_64-unknown-linux-gnu/stage1/lib/rustlib/x86_64-unknown-linux-gnu/lib/librustc
time: 0.422 s parsing
time: 0.031 s gated feature checking
time: 0.000 s crate injection
time: 0.126 s configuration 1
time: 1.014 s expansion
time: 0.251 s configuration 2
time: 0.249 s maybe building test harness
time: 0.273 s prelude injection
time: 0.279 s assinging node ids and indexing ast
time: 0.076 s external crate/lib resolution
time: 0.033 s language item collection
time: 1.028 s resolution
time: 0.036 s lifetime resolution
time: 0.000 s looking for entry point
time: 0.029 s looking for macro registrar
time: 0.063 s freevar finding
time: 0.133 s region resolution
time: 0.111 s type collecting
time: 0.077 s variance inference
time: 0.565 s coherence checking
time: 8.953 s type checking
time: 0.176 s const marking
time: 0.050 s const checking
time: 0.401 s privacy checking
time: 0.063 s effect checking
time: 0.032 s loop checking
time: 1.291 s compute moves
time: 0.172 s match checking
time: 0.249 s liveness checking
time: 0.831 s borrow checking
time: 0.121 s kind checking
time: 0.013 s reachability checking
time: 0.179 s death checking
time: 0.503 s lint checking
time: 14.385 s translation
time: 4.495 s llvm function passes
time: 92.234 s llvm module passes
time: 51.172 s codegen passes
time: 150.809 s LLVM passes
time: 2.542 s running linker
time: 15.109 s linking
```
BUT accesses are much more cache friendly. In fact, if the probe
sequence length is below 8, only two cache lines worth of hashes will be
pulled into cache. This is unlike the old version which would have to
stride over the stoerd keys and values, and would be more cache
unfriendly the bigger the stored values got.
And did you notice the higher load factor? We can now reasonably get a
load factor of 0.9 with very good performance.
Please review this very closely. This is my first major contribution to Rust. Sorry for the ugly diff!
Previously, rust's hashtable was totally unoptimized. It used an Option
per key-value pair, and used very naive open allocation.
The old hashtable had very high variance in lookup time. For an example,
see the 'find_nonexisting' benchmark below. This is fixed by keys in
'lucky' spots with a low probe sequence length getting their good spots
stolen by keys with long probe sequence lengths. This reduces hashtable
probe length variance, while maintaining the same mean.
Also, other optimization liberties were taken. Everything is as cache
aware as possible, and this hashtable should perform extremely well for
both large and small keys and values.
Benchmarks:
comprehensive_old_hashmap 378 ns/iter (+/- 8)
comprehensive_new_hashmap 206 ns/iter (+/- 4)
1.8x faster
old_hashmap_as_queue 238 ns/iter (+/- 8)
new_hashmap_as_queue 119 ns/iter (+/- 2)
2x faster
old_hashmap_insert 172 ns/iter (+/- 8)
new_hashmap_insert 146 ns/iter (+/- 11)
1.17x faster
old_hashmap_find_existing 50 ns/iter (+/- 12)
new_hashmap_find_existing 35 ns/iter (+/- 6)
1.43x faster
old_hashmap_find_notexisting 49 ns/iter (+/- 49)
new_hashmap_find_notexisting 34 ns/iter (+/- 4)
1.44x faster
Memory usage of old hashtable (64-bit assumed):
aligned(8+sizeof(K)+sizeof(V))/0.75 + 6 words
Memory usage of new hashtable:
(aligned(sizeof(K))
+ aligned(sizeof(V))
+ 8)/0.9 + 6.5 words
BUT accesses are much more cache friendly. In fact, if the probe
sequence length is below 8, only two cache lines worth of hashes will be
pulled into cache. This is unlike the old version which would have to
stride over the stoerd keys and values, and would be more cache
unfriendly the bigger the stored values got.
And did you notice the higher load factor? We can now reasonably get a
load factor of 0.9 with very good performance.
* `Ord` inherits from `Eq`
* `TotalOrd` inherits from `TotalEq`
* `TotalOrd` inherits from `Ord`
* `TotalEq` inherits from `Eq`
This is a partial implementation of #12517.
Formatting via reflection has been a little questionable for some time now, and
it's a little unfortunate that one of the standard macros will silently use
reflection when you weren't expecting it. This adds small bits of code bloat to
libraries, as well as not always being necessary. In light of this information,
this commit switches assert_eq!() to using {} in the error message instead of
{:?}.
In updating existing code, there were a few error cases that I encountered:
* It's impossible to define Show for [T, ..N]. I think DST will alleviate this
because we can define Show for [T].
* A few types here and there just needed a #[deriving(Show)]
* Type parameters needed a Show bound, I often moved this to `assert!(a == b)`
* `Path` doesn't implement `Show`, so assert_eq!() cannot be used on two paths.
I don't think this is much of a regression though because {:?} on paths looks
awful (it's a byte array).
Concretely speaking, this shaved 10K off a 656K binary. Not a lot, but sometime
significant for smaller binaries.
This PR allows `HashMap`s to work with custom hashers. Also with this patch are:
* a couple generic implementations of `Hash` for a variety of types.
* added `Default`, `Clone` impls to the hashers.
* added a `HashMap::with_hasher()` constructor.
This patch series does a couple things:
* replaces manual `Hash` implementations with `#[deriving(Hash)]`
* adds `Hash` back to `std::prelude`
* minor cleanup of whitespace and variable names.
This commit changes the ToStr trait to:
impl<T: fmt::Show> ToStr for T {
fn to_str(&self) -> ~str { format!("{}", *self) }
}
The ToStr trait has been on the chopping block for quite awhile now, and this is
the final nail in its coffin. The trait and the corresponding method are not
being removed as part of this commit, but rather any implementations of the
`ToStr` trait are being forbidden because of the generic impl. The new way to
get the `to_str()` method to work is to implement `fmt::Show`.
Formatting into a `&mut Writer` (as `format!` does) is much more efficient than
`ToStr` when building up large strings. The `ToStr` trait forces many
intermediate allocations to be made while the `fmt::Show` trait allows
incremental buildup in the same heap allocated buffer. Additionally, the
`fmt::Show` trait is much more extensible in terms of interoperation with other
`Writer` instances and in more situations. By design the `ToStr` trait requires
at least one allocation whereas the `fmt::Show` trait does not require any
allocations.
Closes#8242Closes#9806
These two containers are indeed collections, so their place is in
libcollections, not in libstd. There will always be a hash map as part of the
standard distribution of Rust, but by moving it out of the standard library it
makes libstd that much more portable to more platforms and environments.
This conveniently also removes the stuttering of 'std::hashmap::HashMap',
although 'collections::HashMap' is only one character shorter.
