Summary:
Adds the coroutine `std::experimental::task<T>` type described in proposal P1056R0.
See https://wg21.link/P1056R0.
This implementation allows customization of the allocator used to allocate the
coroutine frame by passing std::allocator_arg as the first argument, followed by
the allocator to use.
This supports co_awaiting the same task multiple times. The second and
subsequent times it returns a reference to the already-computed value.
This diff also adds some implementations of other utilities that have potential for
standardization as helpers within the test/... area:
- `sync_wait(awaitable)` - See P1171R0
- `manual_reset_event`
Move the definition of the __aligned_allocation_size helper function
from <experimental/memory_resource> to <experimental/__memory>
so it can be more widely used without pulling in memory_resource.
Outstanding work:
- Use C++14 keywords directly rather than macro versions
eg. use `noexcept` instead of `_NOEXCEPT`).
- Add support for overaligned coroutine frames.
This may need wording in the Coroutines TS to support passing the extra `std::align_val_t`.
- Eliminate use of `if constexpr` if we want it to compile under C++14.
Patch by @lewissbaker (Lewis Baker).
llvm-svn: 357010
to reflect the new license. These used slightly different spellings that
defeated my regular expressions.
We understand that people may be surprised that we're moving the header
entirely to discuss the new license. We checked this carefully with the
Foundation's lawyer and we believe this is the correct approach.
Essentially, all code in the project is now made available by the LLVM
project under our new license, so you will see that the license headers
include that license only. Some of our contributors have contributed
code under our old license, and accordingly, we have retained a copy of
our old license notice in the top-level files in each project and
repository.
llvm-svn: 351648
to reflect the new license.
We understand that people may be surprised that we're moving the header
entirely to discuss the new license. We checked this carefully with the
Foundation's lawyer and we believe this is the correct approach.
Essentially, all code in the project is now made available by the LLVM
project under our new license, so you will see that the license headers
include that license only. Some of our contributors have contributed
code under our old license, and accordingly, we have retained a copy of
our old license notice in the top-level files in each project and
repository.
llvm-svn: 351636
Summary:
Starting in Clang 8.0 and GCC 8.0, `alignof` and `__alignof` return different values in same cases. Specifically `alignof` and `_Alignof` return the minimum alignment for a type, where as `__alignof` returns the preferred alignment. libc++ currently uses `__alignof` but means to use `alignof`. See llvm.org/PR39713
This patch introduces the macro `_LIBCPP_ALIGNOF` so we can control which spelling gets used.
This patch does not introduce any ABI guard to provide the old behavior with newer compilers. However, if we decide that is needed, this patch makes it trivial to implement.
I think we should commit this change immediately, and decide what we want to do about the ABI afterwards.
Reviewers: ldionne, EricWF
Reviewed By: ldionne, EricWF
Subscribers: jyknight, christof, libcxx-commits
Differential Revision: https://reviews.llvm.org/D54814
llvm-svn: 351289
This patch changes <experimental/foo> to use #warning instead of
is harmful to common feature detection idioms.
We should also consider only emitting the warning when __DEPRECATED is
defined, like we do in the <ext/foo> headers. Users may want to specify
"-Werror=-W#warnings" while still ignoring the libc++ warnings.
llvm-svn: 350485
This reverts commit 087f065cb0c7463f521a62599884493aaee2ea12.
The tests were failing on 32 bit builds, and I don't have time
to clean them up right now. I'll recommit tomorrow with fixed tests.
llvm-svn: 347816
Summary:
Starting in Clang 8.0 and GCC 8.0, `alignof` and `__alignof` return different values in same cases. Specifically `alignof` and `_Alignof` return the minimum alignment for a type, where as `__alignof` returns the preferred alignment. libc++ currently uses `__alignof` but means to use `alignof`. See llvm.org/PR39713
This patch introduces the macro `_LIBCPP_ALIGNOF` so we can control which spelling gets used.
This patch does not introduce any ABI guard to provide the old behavior with newer compilers. However, if we decide that is needed, this patch makes it trivial to implement.
I think we should commit this change immediately, and decide what we want to do about the ABI afterwards.
Reviewers: ldionne, EricWF
Reviewed By: EricWF
Subscribers: christof, libcxx-commits
Differential Revision: https://reviews.llvm.org/D54814
llvm-svn: 347787
Summary:
std::dynarray had been proposed for C++14, but it was pulled out from C++14
and there are no plans to standardize it anymore.
Reviewers: mclow.lists, EricWF
Subscribers: mgorny, christof, jkorous, dexonsmith, arphaman, libcxx-commits
Differential Revision: https://reviews.llvm.org/D54801
llvm-svn: 347783
Summary:
C++14 sized deallocation is disabled by default due to ABI concerns. However, when a user manually enables it then libc++ should take advantage of it since sized deallocation can provide a significant performance win depending on the underlying malloc implementation. (Note that libc++'s definitions of sized delete don't do anything special yet, but users are free to provide their own).
This patch updates __libcpp_deallocate to selectively call sized operator delete when it's available. `__libcpp_deallocate_unsized` should be used when the size of the allocation is unknown.
On Apple this patch makes no attempt to determine if the sized operator delete is unavailable, only that the language feature is enabled. This could cause a compile error when using `std::allocator`, but the same compile error would occur whenever the user calls `new`, so I don't think it's a problem.
Reviewers: ldionne, mclow.lists
Reviewed By: ldionne
Subscribers: rsmith, ckennelly, libcxx-commits, christof
Differential Revision: https://reviews.llvm.org/D53120
llvm-svn: 345281
Summary:
This patch adds a new macro _LIBCPP_HAS_NO_VECTOR_EXTENSION for detecting
whether a vector extension (\_\_attribute\_\_((vector_size(num_bytes)))) is
available.
On the top of that, this patch implements the following API:
* all constructors
* operator[]
* copy_from
* copy_to
It also defines simd_abi::native to use vector extension, if available.
In GCC and Clang, certain values with vector extension are passed by registers,
instead of memory.
Based on D41148.
Reviewers: mclow.lists, EricWF
Subscribers: cfe-commits, MaskRay, lichray, sanjoy
Differential Revision: https://reviews.llvm.org/D41376
llvm-svn: 338309
This patch implements the <filesystem> header and uses that
to provide <experimental/filesystem>.
Unlike other standard headers, the symbols needed for <filesystem>
have not yet been placed in libc++.so. Instead they live in the
new libc++fs.a library. Users of filesystem are required to link this
library. (Also note that libc++experimental no longer contains the
definition of <experimental/filesystem>, which now requires linking libc++fs).
The reason for keeping <filesystem> out of the dylib for now is that
it's still somewhat experimental, and the possibility of requiring an
ABI breaking change is very real. In the future the symbols will likely
be moved into the dylib, or the dylib will be made to link libc++fs automagically).
Note that moving the symbols out of libc++experimental may break user builds
until they update to -lc++fs. This should be OK, because the experimental
library provides no stability guarantees. However, I plan on looking into
ways we can force libc++experimental to automagically link libc++fs.
In order to use a single implementation and set of tests for <filesystem>, it
has been placed in a special `__fs` namespace. This namespace is inline in
C++17 onward, but not before that. As such implementation is available
in C++11 onward, but no filesystem namespace is present "directly", and
as such name conflicts shouldn't occur in C++11 or C++14.
llvm-svn: 338093
Summary:
The ``file_time_type`` time point is used to represent the write times for files.
Its job is to act as part of a C++ wrapper for less ideal system interfaces. The
underlying filesystem uses the ``timespec`` struct for the same purpose.
However, the initial implementation of ``file_time_type`` could not represent
either the range or resolution of ``timespec``, making it unsuitable. Fixing
this requires an implementation which uses more than 64 bits to store the
time point.
I primarily considered two solutions: Using ``__int128_t`` and using a
arithmetic emulation of ``timespec``. Each has its pros and cons, and both
come with more than one complication.
However, after a lot of consideration, I decided on using `__int128_t`. This patch implements that change.
Please see the [FileTimeType Design Document](http://libcxx.llvm.org/docs/DesignDocs/FileTimeType.html) for more information.
Reviewers: mclow.lists, ldionne, joerg, arthur.j.odwyer, EricWF
Reviewed By: EricWF
Subscribers: christof, K-ballo, cfe-commits, BillyONeal
Differential Revision: https://reviews.llvm.org/D49774
llvm-svn: 337960
Previously the <experimental/filesystem> didn't guard its
contents in any dialect. However, the implementation implicitly
requires at least C++11, and the tests have always been marked
unsupported in C++03. This patch puts a header guard around the
contents to avoid exposing them before C++11.
Additionally, it replaces all of the usages of _NOEXCEPT or
_LIBCPP_CONSTEXPR with the keyword directly, since we can
expect the compiler to implement those by now.
llvm-svn: 337884
To avoid exposing implementation details, path::iterator and PathParser
both implicitly used the same set of values to represent the state,
but they were defined twice. This could have lead to a mismatch
occuring.
This patch moves all of the parser state values into the filesystem
header and changes PathParser to use those value to avoid this.
llvm-svn: 337883
The initial patch didn't correctly handle systems when the dirent struct
didn't provide the d_type member. Specifically it set the cache to the incorrect state,
and claimed it was partially populated.
The updated version of this change correctly handles setting up the
cache when the file type is not known (aka file_type::none).
llvm-svn: 337765
This patch implements the `what()` for filesystem errors. The message
includes the 'what_arg', any paths that were specified, and the
error code message.
Additionally this patch refactors how errors are created, making it easier
to report them correctly.
llvm-svn: 337664
First, <experimental/filesystem> didn't correctly guard
against min/max macros. This adds the proper push/pop macro guards.
Second, an internal time helper had been renamed but the test for
it hadn't been updated. This patch updates those tests.
llvm-svn: 337520
Summary:
This patch implements directory_entry caching *almost* as specified in P0317r1. However, I explicitly chose to deviate from the standard as I'll explain below.
The approach I decided to take is a fully caching one. When `refresh()` is called, the cache is populated by calls to `stat` and `lstat` as needed.
During directory iteration the cache is only populated with the `file_type` as reported by `readdir`.
The cache can be in the following states:
* `_Empty`: There is nothing in the cache (likely due to an error)
* `_IterSymlink`: Created by directory iteration when we walk onto a symlink only the symlink file type is known.
* `_IterNonSymlink`: Created by directory iteration when we walk onto a non-symlink. Both the regular file type and symlink file type are known.
* `_RefreshSymlink` and `_RefreshNonSymlink`: A full cache created by `refresh()`. This case includes dead symlinks.
* `_RefreshSymlinkUnresolved`: A partial cache created by refresh when we fail to resolve the file pointed to by a symlink (likely due to permissions). Symlink attributes are cached, but attributes about the linked entity are not.
As mentioned, this implementation purposefully deviates from the standard. According to some readings of the specification, and the Windows filesystem implementation, the constructors and modifiers which don't pass an `error_code` must throw when the `directory_entry` points to a entity which doesn't exist. or when attribute resolution fails for another reason.
@BillyONeal has proposed a more reasonable set of requirements, where modifiers other than refresh ignore errors. This is the behavior libc++ currently implements, with the expectation some form of the new language will be accepted into the standard.
Some additional semantics which differ from the Windows implementation:
1. `refresh` will not throw when the entry doesn't exist. In this case we can still meet the functions specification, so we don't treat it as an error.
2. We don't clear the path name when a constructor fails via refresh (this will hopefully be changed in the standard as well).
It should be noted that libstdc++'s current implementation has the same behavior as libc++, except for point (2).
If the changes to the specification don't get accepted, we'll be able to make the changes later.
[1] http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0317r1.html
Reviewers: mclow.lists, gromer, ldionne, aaron.ballman
Subscribers: BillyONeal, christof, cfe-commits
Differential Revision: https://reviews.llvm.org/D49530
llvm-svn: 337516
Summary:
We never actually mean to always inline a function -- all the uses of
the macro I could find are actually attempts to control the visibility
of symbols. This is better described by _LIBCPP_INLINE_VISIBILITY, which
is actually always defined the same.
This change is orthogonal to the decision of what we're actually going
to do with _LIBCPP_INLINE_VISIBILITY -- it just simplifies things by
having one canonical way of doing things.
Note that this commit had originally been applied in r336369 and then
reverted in r336382 because of unforeseen problems. Both of these problems
have now been fixed.
Reviewers: EricWF, mclow.lists
Subscribers: christof, dexonsmith, erikvanderpoel
Differential Revision: https://reviews.llvm.org/D48892
llvm-svn: 336866
Summary:
It was defined with the right visibility, but declared without any visibility.
This function was left out of a prior revision that did the same to several
functions in <compare> (r336665) because the compiler I used didn't support
coroutines. This reinforces the need for automated checks -- there might
still be several cases of this throughout the library.
Reviewers: EricWF
Subscribers: modocache, christof, dexonsmith, llvm-commits
Differential Revision: https://reviews.llvm.org/D49145
llvm-svn: 336709
This reverts commit r336369. The commit had two problems:
1. __pbump was marked as _LIBCPP_EXTERN_TEMPLATE_INLINE_VISIBILITY instead of
_LIBCPP_INLINE_VISIBILITY, which lead to two symbols being added in the
dylib and the check-cxx-abilist failing.
2. The LLDB tests started failing because they undefine
`_LIBCPP_INLINE_VISIBILITY`. I need to figure out why they do that and
fix the tests before we can go forward with this change.
llvm-svn: 336382
Summary:
We never actually mean to always inline a function -- all the uses of
the macro I could find are actually attempts to control the visibility
of symbols. This is better described by _LIBCPP_INLINE_VISIBILITY, which
is actually always defined the same.
This change is orthogonal to the decision of what we're actually going
to do with _LIBCPP_INLINE_VISIBILITY -- it just simplifies things by
having one canonical way of doing things.
Reviewers: EricWF
Subscribers: christof, llvm-commits, dexonsmith, erikvanderpoel, mclow.lists
Differential Revision: https://reviews.llvm.org/D48892
llvm-svn: 336369
Patch from Arthur O'Dwyer.
`__user_alloc_construct_impl` is used by <experimental/memory_resource>, but
this `__user_alloc_construct` is never used.
Also, `<experimental/memory_resource>` doesn't need a full definition of
`std::tuple`; just the forward declaration in `<__tuple>` will suffice.
Reviewed as https://reviews.llvm.org/D46806
llvm-svn: 334069
Patch from Arthur O'Dwyer.
In the TS, `uses_allocator` construction for `pair` tried to use an allocator
type of `memory_resource*`, which is incorrect because `memory_resource*` is
not an allocator type. LWG 2969 fixed it to use `polymorphic_allocator` as the
allocator type instead.
https://wg21.link/lwg2969
(D47090 included this in `<memory_resource>`; at Eric's request, I've split
this out into its own patch applied to the existing
`<experimental/memory_resource>` instead.)
Reviewed as https://reviews.llvm.org/D47109
llvm-svn: 333384
There are 3 changes:
* Renamed genertor.pass.cpp to generator.pass.cpp
* Removed nothing_to_do.pass.cpp
* Mark GCC 4.9 as UNSUPPORTED for the test files that have negative
narrowing conversion SFINAE test (see GCC PR63723).
llvm-svn: 330655
Summary:
The patch includes all declarations, and also implements the following features:
* ABI.
* narrowing-conversion related SFIANE, including simd<> ctors and (static_)simd_cast.
Reviewers: mclow.lists, EricWF
Subscribers: lichray, sanjoy, MaskRay, cfe-commits
Differential Revision: https://reviews.llvm.org/D41148
llvm-svn: 330627
This patch implements P0430R2, who's largest change is adding the path::format
enumeration for supporting path format conversions in path constructors.
However, since libc++'s filesystem only really supports POSIX like systems,
there are no real changes needed. This patch simply adds the format enum
and then ignores it when it's passed to constructors.
llvm-svn: 329031
This is a fairly large patch that implements all of the filesystem NB comments
and the relative paths changes (ex. adding weakly_canonical). These issues
and papers are all interrelated so their implementation couldn't be split up
nicely.
This patch upgrades <experimental/filesystem> to match the C++17 spec and not
the published experimental TS spec. Some of the changes in this patch are both
API and ABI breaking, however libc++ makes no guarantee about stability for
experimental implementations.
The major changes in this patch are:
* Implement NB comments for filesystem (P0492R2), including:
* Implement `perm_options` enum as part of NB comments, and update the
`permissions` function to match.
* Implement changes to `remove_filename` and `replace_filename`
* Implement changes to `path::stem()` and `path::extension()` which support
splitting examples like `.profile`.
* Change path iteration to return an empty path instead of '.' for trailing
separators.
* Change `operator/=` to handle absolute paths on the RHS.
* Change `absolute` to no longer accept a current path argument.
* Implement relative paths according to NB comments (P0219r1)
* Combine `path.cpp` and `operations.cpp` since some path functions require
access to the operations internals, and some fs operations require access
to the path parser.
llvm-svn: 329028
The NB comments for filesystem changed permissions and added
a new enum `perm_options` which control how the permissions
are applied.
This implements than NB resolution
llvm-svn: 328476
This patch removes the noexcept declaration from filesystem
operations which require creating temporary paths or
creating a directory iterator. Either of these operations
can throw.
llvm-svn: 324192
Because path can be constructed from a ton of different types, including string
and wide strings, this caused it's streaming operators to suck up all sorts
of silly types via silly conversions. For example:
using namespace std::experimental::filesystem::v1;
std::wstring w(L"wide");
std::cout << w; // converts to path.
This patch tentatively adopts the resolution to LWG2989 and fixes the issue
by making the streaming operators friends of path.
llvm-svn: 324189
Summary:
In the CHERI clang compiler __output and __input are keywords and therefore
we can't compile libc++ with our compiler.
Reviewers: mclow.lists, EricWF, theraven
Reviewed By: EricWF
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D39537
llvm-svn: 318144
LWG 3013 points out that the constructors and increment members
of the directory iterators need to allocate, and therefore cannot
be marked noexcept.
It also points out that `is_empty` and `copy` likely need to allocate
as well, and as such can also not be noexcept.
This patch speculatively implements the resolution removing noexcept,
because libc++ does indeed have the possibility of throwing on allocation
failure.
llvm-svn: 316941
It seems conceivable that a user would need to get a coroutine handle
having only a const reference to the promise_type, for example from
within a const member function of the promise.
This patch allows that use case. A coroutine_handle<const T> can be used
in essentially the same way a coroutine_handle<T>, ie to start and destroy
the coroutine. The constness of the promise doesn't/shouldn't propagate
to the handle.
llvm-svn: 305536
Summary:
This patch improves how libc++ handles min/max macros within the headers. Previously libc++ would undef them and emit a warning.
This patch changes libc++ to use `#pragma push_macro` to save the macro before undefining it, and `#pragma pop_macro` to restore the macros and the end of the header.
Reviewers: mclow.lists, bcraig, compnerd, EricWF
Reviewed By: EricWF
Subscribers: cfe-commits, krytarowski
Differential Revision: https://reviews.llvm.org/D33080
llvm-svn: 304357
from_address requires that the provided pointer refer to the suspended coroutine,
which doesn't have a type, or at least not one knowable by the user. Therefore
every use of `from_address` with a typed pointer is almost certainly a bug.
This behavior is a part of the TS specification, but hopefully it will be
in the future.
llvm-svn: 304172
More tests to come. I think that from_address overload should be deleted
or ill-formed, except for the 'void*' one; The user cannot possibly
have a typed pointer to the coroutine state.
llvm-svn: 304131
Clang supports coroutines in all dialects; Therefore libc++ should too,
otherwise the Clang extension is unusable.
I'm not convinced extending support to C++03 is a feasible long term
plan, since as the library grows to offer things like generators it
will be come increasingly difficult to limit the implementation to C++03.
However for the time being supporting C++03 isn't a big deal.
llvm-svn: 303963
Clang started providing -fcoroutines and defining __cpp_coroutines
way before it implemented the __builtin_coro_foo functions. This
means that simply checking if __cpp_coroutines is not a sufficient
way of detecting the actual feature.
This patch implements _LIBCPP_HAS_NO_COROUTINES which implements
a slightly more complex feature check. Specifically it requires
__cpp_coroutines >= 201703L, which only holds for Clang 5.0 built
after 2017/05/24.
llvm-svn: 303956
The original issues were caused because <experimental/coroutine>
didn't correctly #ifdef out enough of the header, which caused incomplete
types to be used.
This patch fixes the `#if defined(__cpp_coroutines)` guard and re-adds
the headers to the module map.
It also uglifies some incorrectly non-reserved names.
llvm-svn: 303936
This patch updates the promise() member to match the current spec.
Specifically it removes the non-const overload and make the return
type of the const overload non-const.
This patch also makes the ASSERT_NOT_NOEXCEPT tests libc++ specific,
since other implementations may be free to strengthen the specification.
llvm-svn: 303895
This patch fixes the test failures and unexpected passes that occur
when testing against GCC 7. Specifically:
* don't mark __gcd as always inline because it's a recursive function. GCC diagnoses this.
* don't XFAIL the aligned allocation tests. GCC 7 supports them but not the -faligned-allocation option.
* Work around gcc.gnu.org/PR78489 in variants constructors.
llvm-svn: 302488
Libc++ is used as a system library on macOS and iOS (amongst others). In order
for users to be able to compile a binary that is intended to be deployed to an
older version of the platform, clang provides the
availability attribute <https://clang.llvm.org/docs/AttributeReference.html#availability>_
that can be placed on declarations to describe the lifecycle of a symbol in the
library.
See docs/DesignDocs/AvailabilityMarkup.rst for more information.
Differential Revision: https://reviews.llvm.org/D31739
llvm-svn: 302172
path::iterator isn't a strictly conforming iterator. Specifically
it stashes the current element inside the iterator. This leads to
UB when used with reverse_iterator since it requires the element
to outlive the lifetime of the iterator.
This patch adds a static_assert inside reverse_iterator to disallow
"stashing iterator types", and it tags path::iterator as such a type.
Additionally this patch removes all uses of reverse_iterator<path::iterator>
within the tests.
llvm-svn: 300164
These member functions were decorated with `_LIBCPP_FUNC_VIS` when the
class is also decorated with external visibility. This breaks down when
building for PE/COFF, where the member function cannot be decorated if
it is within a decorated class. The class attribute will propagate to
the member. Remove the extraneous decoration.
llvm-svn: 293454
Microsoft's SAL has a `__deref` macro which results in a compilation
failure when building the filesystem module on Windows. Rename the
member function internally to avoid the conflict.
llvm-svn: 293449
Adding `path::operator=(string_type&&)` made the expression `p = {}`
ambiguous. This path fixes that ambiguity by making the `string&&`
overload a template so it ranks lower during overload resolution.
llvm-svn: 292345
Windows is greedy and it defines the identifier `__out` as a macro.
This patch renames all conflicting libc++ identifiers in order
to correctly work on Windows.
llvm-svn: 291345
Summary:
On Windows the identifier `__deallocate` is defined as a macro by one of the Windows system headers. Previously libc++ worked around this by `#undef __deallocate` and generating a warning. However this causes the WIN32 version of `__threading_support` to always generate a warning on Windows. This is not OK.
This patch renames all usages of `__deallocate` internally as to not conflict with the macro.
Reviewers: mclow.lists, majnemer, rnk, rsmith, smeenai, compnerd
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D28426
llvm-svn: 291332
The name _LIBCPP_TYPE_VIS_ONLY is no longer accurate because both
_LIBCPP_TYPE_VIS and _LIBCPP_TYPE_VIS_ONLY expand to
__attribute__((__type_visibility__)) with Clang. The only remaining difference
is that _LIBCPP_TYPE_VIS_ONLY can be applied to templates whereas
_LIBCPP_TYPE_VIS cannot (due to dllimport/dllexport not being allowed on
templates).
This patch renames _LIBCPP_TYPE_VIS_ONLY to _LIBCPP_TEMPLATE_VIS.
llvm-svn: 291035
In the previous fix I used a PMF type as a semi-safe bool type in C++03.
However immediately after committing I realized clang offered explicit
conversion operators as an extension. This patch removes the old fix and
enables _LIBCPP_EXPLICIT using __has_extension instead.
This change also affects the following other classes, which have
'_LIBCPP_EXPLICIT operator bool()'.
* shared_ptr
* unique_ptr
* error_condition
* basic_ios
* function (already C++11 only)
* istream::sentry
* experimental::string_view.
In all of the above cases I believe it is safe to enable the extension, except
in the experimental::string_view case. There seem to be some Clang bugs
affecting the experimental::string_view conversion to std::basic_string. To
work around that I manually disabled _LIBCPP_EXPLICIT in that case.
llvm-svn: 290831
path uses string::append to construct, append, and concatenate paths. Unfortunatly
string::append has a strong exception safety guaranteed and if it can't prove
that the iterator operations don't throw then it will allocate a temporary
string copy to append to. However this extra allocation and copy is very
undesirable for path which doesn't have the same exception guarantees.
To work around this this patch adds string::__append_forward_unsafe which exposes
the std::string::append interface for forward iterators without enforcing
that the iterator is noexcept.
llvm-svn: 285532
This patch fixes a performance bug when constructing or appending to a path
from a string or c-string. Previously we called 'push_back' to append every
single character. This caused multiple re-allocation and copies when at most
one reallocation is necessary. The new behavior is to simply call
`string::append` so it can correctly handle reallocation.
For large strings this change is a ~4x improvement. This also makes our path
faster to construct than libstdc++'s.
llvm-svn: 285530
This patch entirely rewrites the parsing logic for paths. Unlike the previous
implementation this one stores information about the current state; For example
if we are in a trailing separator or a root separator. This avoids the need for
extra lookahead (and extra work) when incrementing or decrementing an iterator.
Roughly this gives us a 15% speedup over the previous implementation.
Unfortunately this implementation is still a lot slower than libstdc++'s.
Because libstdc++ pre-parses and splits the path upon construction their
iterators are trivial to increment/decrement. This makes libc++ lazy parsing
100x slower than libstdc++. However the pre-parsing libstdc++ causes a ton
of extra and unneeded allocations when constructing the string. For example
`path("/foo/bar/")` would require at least 5 allocations with libstdc++
whereas libc++ uses only one. The non-allocating behavior is much preferable
when you consider filesystem usages like 'exists("/foo/bar/")'.
Even then libc++'s path seems to be twice as slow to simply construct compared
to libstdc++. More investigation is needed about this.
llvm-svn: 285526
Marshall Clow