Fix#13266.
There is a little bit of acrobatics in the definition of `crate_paths`
to avoid calling `clone()` on the dylib/rlib unless we actually are
going to need them.
The other oddity is that I have replaced the `root_ident: Option<&str>`
parameter with a `root: &Option<CratePaths>`, which may surprise one
who was expecting to see something like: `root: Option<&CratePaths>`.
I went with the approach here because I could not come up with code for
the alternative that was acceptable to the borrow checker.
This is a continuation of the work done in #13184 to make struct fields private
by default. This commit finishes RFC 4 by making all tuple structs have private
fields by default. Note that enum variants are not affected.
A tuple struct having a private field means that it cannot be matched on in a
pattern match (both refutable and irrefutable), and it also cannot have a value
specified to be constructed. Similarly to private fields, switching the type of
a private field in a tuple struct should be able to be done in a backwards
compatible way.
The one snag that I ran into which wasn't mentioned in the RFC is that this
commit also forbids taking the value of a tuple struct constructor. For example,
this code now fails to compile:
mod a {
pub struct A(int);
}
let a: fn(int) -> a::A = a::A; //~ ERROR: first field is private
Although no fields are bound in this example, it exposes implementation details
through the type itself. For this reason, taking the value of a struct
constructor with private fields is forbidden (outside the containing module).
RFC: 0004-private-fields
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
A variety of stuff here, mostly aimed at making `make install` work correctly with `--libdir` and `--mandir`. `make install` again goes through `install.sh`.
This change is in preparation for #8122. Nothing is currently done with these
visibility qualifiers, they are just parsed and accepted by the compiler.
RFC: 0004-private-fields
This is not sufficient for finding the library directory for binary
installs, but it does make the build more complex by requiring
env vars be set to build rustc.
This needs to be removed as part of removing `~[T]`. Partial type hints
are now allowed, and will remove the need to add a version of this
method for `Vec<T>`. For now, this involves a few workarounds for
partial type hints not completely working.
These are superfluous now that we have fixed rvalue lifetimes and Deref.
I'd also like to kill off `get` and `set`, but that'll be a large change so I want to make sure that we actually want to do that first.
This is the first step to replacing OptVec with a new representation:
remove all mutability. Any mutations have to go via `Vec` and then make
to `OptVec`.
Many of the uses of OptVec are unnecessary now that Vec has no-alloc
emptiness (and have been converted to Vec): the only ones that really
need it are the AST and sty's (and so on) where there are a *lot* of
instances of them, and they're (mostly) immutable.
Previously, any library of the pattern `lib<name>-<hash>-<version>.so` was
>considered a candidate (rightly so) for loading a crate. Sets are generated for
each unique `<hash>`, and then from these sets a candidate is selected. If a set
contained more than one element, then it immediately generated an error saying
that multiple copies of the same dylib were found.
This is incorrect because each candidate needs to be validated to actually
contain a rust library (valid metadata). This commit alters the logic to filter
each set of candidates for a hash to only libraries which are actually rust
libraries. This means that if multiple false positives are found with the right
name pattern, they're all ignored.
Closes#13010
`Share` implies that all *reachable* content is *threadsafe*.
Threadsafe is defined as "exposing no operation that permits a data race if multiple threads have access to a &T pointer simultaneously". (NB: the type system should guarantee that if you have access to memory via a &T pointer, the only other way to gain access to that memory is through another &T pointer)...
Fixes#11781
cc #12577
What this PR will do
================
- [x] Add Share kind and
- [x] Replace usages of Freeze with Share in bounds.
- [x] Add Unsafe<T> #12577
- [x] Forbid taking the address of a immutable static item with `Unsafe<T>` interior
What's left to do in a separate PR (after the snapshot)?
===========================================
- Remove `Freeze` completely
Previously, any library of the pattern `lib<name>-<hash>-<version>.so` was
>considered a candidate (rightly so) for loading a crate. Sets are generated for
each unique `<hash>`, and then from these sets a candidate is selected. If a set
contained more than one element, then it immediately generated an error saying
that multiple copies of the same dylib were found.
This is incorrect because each candidate needs to be validated to actually
contain a rust library (valid metadata). This commit alters the logic to filter
each set of candidates for a hash to only libraries which are actually rust
libraries. This means that if multiple false positives are found with the right
name pattern, they're all ignored.
Closes#13010
This reduces the size of sty from 112 to 96; like with the ty_trait
variant, this variant of sty occurs rarely (~1%) so the benefits are
large and the costs small.
This reduces ty::sty from 160 bytes to just 112, and some measurements
eddyb made suggest that the ty_trait variant occurs very
rarely (e.g. ~1% of all sty instances) hence this will result in a large
memory saving, and the cost of the indirection is unlikely to be an
issue.
This commit switches over the backtrace infrastructure from piggy-backing off
the RUST_LOG environment variable to using the RUST_BACKTRACE environment
variable (logging is now disabled in libstd).
This commit starts to topographically sort rust dependencies on the linker
command line. The reason for this is that linkers use right-hand libraries to
resolve left-hand libraries symbols, which is especially crucial for us because
we're using --as-needed on linux.
This commit shreds all remnants of libextra from the compiler and standard
distribution. Two modules, c_vec/tempfile, were moved into libstd after some
cleanup, and the other modules were moved to separate crates as seen fit.
Closes#8784Closes#12413Closes#12576
Refactored get_metadata_section to return a Result<MetadataBlob,~str> instead of a Option<MetadataBlob>. This provides more clarity to the user through the debug output when using --ls.
This is kind of a continuation of my original closed pull request 2 months ago (#11544), but I think the time-span constitutes a new pull request.
This leverages the new hashing framework and hashmap implementation to provide a
much speedier hashing algorithm for node ids and def ids. The hash algorithm
used is currentl FNV hashing, but it's quite easy to swap out.
I originally implemented hashing as the identity function, but this actually
ended up in slowing down rustc compiling libstd from 8s to 13s. I would suspect
that this is a result of a large number of collisions.
With FNV hashing, we get these timings (compiling with --no-trans, in seconds):
| | before | after |
|-----------|---------:|--------:|
| libstd | 8.324 | 6.703 |
| stdtest | 47.674 | 46.857 |
| libsyntax | 9.918 | 8.400 |
Linker argument order with respect to libraries is important enough that we
shouldn't be attempting to filter out libraries getting passed through to the
linker. When linking with a native library that has multiple dependant native
libraries, it's useful to have control over the link argument order.
When the metadata format changes, old libraries often cause librustc to abort
when reading their metadata. This should all change with the introduction of SVH
markers, but the loader for crates should gracefully handle libraries without
SVH markers still.
This commit adds support for tripping fewer assertions when loading libraries by
using maybe_get_doc when initially parsing metadata. It's still possible for
some libraries to fall through the cracks, but this should deal with a fairly
large number of them up front.
This new SVH is used to uniquely identify all crates as a snapshot in time of
their ABI/API/publicly reachable state. This current calculation is just a hash
of the entire crate's AST. This is obviously incorrect, but it is currently the
reality for today.
This change threads through the new Svh structure which originates from crate
dependencies. The concept of crate id hash is preserved to provide efficient
matching on filenames for crate loading. The inspected hash once crate metadata
is opened has been changed to use the new Svh.
The goal of this hash is to identify when upstream crates have changed but
downstream crates have not been recompiled. This will prevent the def-id drift
problem where upstream crates were recompiled, thereby changing their metadata,
but downstream crates were not recompiled.
In the future this hash can be expanded to exclude contents of the AST like doc
comments, but limitations in the compiler prevent this change from being made at
this time.
Closes#10207
The previous code passed around a {name,version} pair everywhere, but this is
better expressed as a CrateId. This patch changes these paths to store and pass
around crate ids instead of these pairs of name/version. This also prepares the
code to change the type of hash that is stored in crates.
This commit implements a layman's version of realpath() for metadata::loader to
use in order to not error on symlinks pointing to the same file.
Closes#12459
Commits for details. Highlights:
- `flate` returns `CVec<u8>` to save reallocating a whole new `&[u8]`
- a lot of `transmute`s removed outright or replaced with `as` (etc.)
This trades an O(n) allocation + memcpy for a O(1) proc allocation (for
the destructor). Most users only need &[u8] anyway (all of the users in
the main repo), and so this offers large gains.
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.
This PR merges `IterBytes` and `Hash` into a trait that allows for generic non-stream-based hashing. It makes use of @eddyb's default type parameter support in order to have a similar usage to the old `Hash` framework.
Fixes#8038.
Todo:
- [x] Better documentation
- [ ] Benchmark
- [ ] Parameterize `HashMap` on a `Hasher`.
Closes#12366.
Parentheses around assignment statements such as
let mut a = (0);
a = (1);
a += (2);
are not necessary and therefore an unnecessary_parens warning is raised when
statements like this occur.
The warning mechanism was refactored along the way to allow for code reuse
between the routines for checking expressions and statements.
Code had to be adopted throughout the compiler and standard libraries to comply
with this modification of the lint.
This patch merges IterBytes and Hash traits, which clears up the
confusion of using `#[deriving(IterBytes)]` to support hashing.
Instead, it now is much easier to use the new `#[deriving(Hash)]`
for making a type hashable with a stream hash.
Furthermore, it supports custom non-stream-based hashers, such as
if a value's hash was cached in a database.
This does not yet replace the old IterBytes-hash with this new
version.
Added allow(non_camel_case_types) to librustc where necesary
Tried to fix problems with non_camel_case_types outside rustc
fixed failing tests
Docs updated
Moved #[allow(non_camel_case_types)] a level higher.
markdown.rs reverted
Fixed timer that was failing tests
Fixed another timer
The first commit improves error messages during linking, and the second commit improves error messages during crate-loading time.
Closes#12297Closes#12377
This commit rewrites crate loading internally in attempt to look at less
metadata and provide nicer errors. The loading is now split up into a few
stages:
1. Collect a mapping of (hash => ~[Path]) for a set of candidate libraries for a
given search. The hash is the hash in the filename and the Path is the
location of the library in question. All candidates are filtered based on
their prefix/suffix (dylib/rlib appropriate) and then the hash/version are
split up and are compared (if necessary).
This means that if you're looking for an exact hash of library you don't have
to open up the metadata of all libraries named the same, but also in your
path.
2. Once this mapping is constructed, each (hash, ~[Path]) pair is filtered down
to just a Path. This is necessary because the same rlib could show up twice
in the path in multiple locations. Right now the filenames are based on just
the crate id, so this could be indicative of multiple version of a crate
during one crate_id lifetime in the path. If multiple duplicate crates are
found, an error is generated.
3. Now that we have a mapping of (hash => Path), we error on multiple versions
saying that multiple versions were found. Only if there's one (hash => Path)
pair do we actually return that Path and its metadata.
With this restructuring, it restructures code so errors which were assertions
previously are now first-class errors. Additionally, this should read much less
metadata with lots of crates of the same name or same version in a path.
Closes#11908
This "bubble up an error" macro was originally named if_ok! in order to get it
landed, but after the fact it was discovered that this name is not exactly
desirable.
The name `if_ok!` isn't immediately clear that is has much to do with error
handling, and it doesn't look fantastic in all contexts (if if_ok!(...) {}). In
general, the agreed opinion about `if_ok!` is that is came in as subpar.
The name `try!` is more invocative of error handling, it's shorter by 2 letters,
and it looks fitting in almost all circumstances. One concern about the word
`try!` is that it's too invocative of exceptions, but the belief is that this
will be overcome with documentation and examples.
Close#12037
The first setp for #9880 is to add a new `crate` keyword. This PR does exactly that. I took a chance to refactor `parse_item_foreign_mod` and I broke it down into 2 separate methods to isolate each feature.
The next step will be to push a new stage0 snapshot and then get rid of all `extern mod` around the code.
This patch replaces all `crate` usage with `krate` before introducing the
new keyword. This ensures that after introducing the keyword, there
won't be any compilation errors.
krate might not be the most expressive substitution for crate but it's a
very close abbreviation for it. `module` was already used in several
places already.
Error messages cleaned in librustc/middle
Error messages cleaned in libsyntax
Error messages cleaned in libsyntax more agressively
Error messages cleaned in librustc more aggressively
Fixed affected tests
Fixed other failing tests
Last failing tests fixed
- `extra::json` didn't make the cut, because of `extra::json` required
dep on `extra::TreeMap`. If/when `extra::TreeMap` moves out of `extra`,
then `extra::json` could move into `serialize`
- `libextra`, `libsyntax` and `librustc` depend on the newly created
`libserialize`
- The extensions to various `extra` types like `DList`, `RingBuf`, `TreeMap`
and `TreeSet` for `Encodable`/`Decodable` were moved into the respective
modules in `extra`
- There is some trickery, evident in `src/libextra/lib.rs` where a stub
of `extra::serialize` is set up (in `src/libextra/serialize.rs`) for
use in the stage0 build, where the snapshot rustc is still making
deriving for `Encodable` and `Decodable` point at extra. Big props to
@huonw for help working out the re-export solution for this
extra: inline extra::serialize stub
fix stuff clobbered in rebase + don't reexport serialize::serialize
no more globs in libserialize
syntax: fix import of libserialize traits
librustc: fix bad imports in encoder/decoder
add serialize dep to librustdoc
fix failing run-pass tests w/ serialize dep
adjust uuid dep
more rebase de-clobbering for libserialize
fixing tests, pushing libextra dep into cfg(test)
fix doc code in extra::json
adjust index.md links to serialize and uuid library
This removes @[] from the parser as well as much of the handling of it (and `@str`) from the compiler as I can find.
I've just rebased @pcwalton's (already reviewed) `@str` removal (and fixed the problems in a separate commit); the only new work is the trailing commits with my authorship.
Closes#11967
It was decided a long, long time ago that libextra should not exist, but rather its modules should be split out into smaller independent libraries maintained outside of the compiler itself. The theory was to use `rustpkg` to manage dependencies in order to move everything out of the compiler, but maintain an ease of usability.
Sadly, the work on `rustpkg` isn't making progress as quickly as expected, but the need for dissolving libextra is becoming more and more pressing. Because of this, we've thought that a good interim solution would be to simply package more libraries with the rust distribution itself. Instead of dissolving libextra into libraries outside of the mozilla/rust repo, we can dissolve libraries into the mozilla/rust repo for now.
Work on this has been excruciatingly painful in the past because the makefiles are completely opaque to all but a few. Adding a new library involved adding about 100 lines spread out across 8 files (incredibly error prone). The first commit of this pull request targets this pain point. It does not rewrite the build system, but rather refactors large portions of it. Afterwards, adding a new library is as simple as modifying 2 lines (easy, right?). The build system automatically keeps track of dependencies between crates (rust *and* native), promotes binaries between stages, tracks dependencies of installed tools, etc, etc.
With this newfound buildsystem power, I chose the `extra::flate` module as the first candidate for removal from libextra. While a small module, this module is relative complex in that is has a C dependency and the compiler requires it (messing with the dependency graph a bit). Albeit I modified more than 2 lines of makefiles to accomodate libflate (the native dependency required 2 extra lines of modifications), but the removal process was easy to do and straightforward.
---
Testing-wise, I've cross-compiled, run tests, built some docs, installed, uninstalled, etc. I'm still working out a few kinks, and I'm sure that there's gonna be built system issues after this, but it should be working well for basic use!
cc #8784
This is hopefully the beginning of the long-awaited dissolution of libextra.
Using the newly created build infrastructure for building libraries, I decided
to move the first module out of libextra.
While not being a particularly meaty module in and of itself, the flate module
is required by rustc and additionally has a native C dependency. I was able to
very easily split out the C dependency from rustrt, update librustc, and
magically everything gets installed to the right locations and built
automatically.
This is meant to be a proof-of-concept commit to how easy it is to remove
modules from libextra now. I didn't put any effort into modernizing the
interface of libflate or updating it other than to remove the one glob import it
had.
Before this commit, rustc looked in `dirname $0`/../lib for libraries
but that doesn't work when rustc is invoked through a symlink.
This commit makes rustc look in `dirname $(readlink $0)`/../lib, i.e.
it first canonicalizes the symlink before walking up the directory tree.
Fixes#3632.
The old method of serializing the AST gives totally bogus spans if the
expansion of an imported macro causes compilation errors. The best
solution seems to be to serialize the actual textual macro definition
and load it the same way the std-macros are. I'm not totally confident
that getting the source from the CodeMap will always do the right thing,
but it seems to work in simple cases.
The old method of serializing the AST gives totally bogus spans if the
expansion of an imported macro causes compilation errors. The best
solution seems to be to serialize the actual textual macro definition
and load it the same way the std-macros are. I'm not totally confident
that getting the source from the CodeMap will always do the right thing,
but it seems to work in simple cases.
Before this commit, rustc looked in `dirname $0`/../lib for libraries
but that doesn't work when rustc is invoked through a symlink.
This commit makes rustc look in `dirname $(readlink $0)`/../lib, i.e.
it first canonicalizes the symlink before walking up the directory tree.
Fixes#3632.
NodeIds are sequential integers starting at zero, so we can achieve some
memory savings by just storing the items all in a line in a vector.
The occupancy for typical crates seems to be 75-80%, so we're already
more efficient than a HashMap (maximum occupancy 75%), not even counting
the extra book-keeping that HashMap does.
* Reexport io::mem and io::buffered structs directly under io, make mem/buffered
private modules
* Remove with_mem_writer
* Remove DEFAULT_CAPACITY and use DEFAULT_BUF_SIZE (in io::buffered)
cc #11119
* Reexport io::mem and io::buffered structs directly under io, make mem/buffered
private modules
* Remove with_mem_writer
* Remove DEFAULT_CAPACITY and use DEFAULT_BUF_SIZE (in io::buffered)
The `print!` and `println!` macros are now the preferred method of printing, and so there is no reason to export the `stdio` functions in the prelude. The functions have also been replaced by their macro counterparts in the tutorial and other documentation so that newcomers don't get confused about what they should be using.
This is just an unnecessary trait that no one's ever going to parameterize over
and it's more useful to just define the methods directly on the types
themselves. The implementors of this type almost always don't want
inner_mut_ref() but they're forced to define it as well.
The resulting symbol names aren't very pretty at all:
trait Trait { fn method(&self); }
impl<'a> Trait for ~[(&'a int, fn())] { fn method(&self) {} }
gives
Trait$$UP$$VEC$$TUP_2$$BP$int$$FN$$::method::...hash...::v0.0
However, at least it contain some reference to the Self type, unlike
`Trait$__extensions__::method:...`, which is what the symbol name used
to be for anything other than `impl Trait for foo::bar::Baz` (which
became, and still becomes, `Trait$Baz::method`).
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.
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).
