* Add in-kernel copying support for Linux (sendfile64(2), copy_file_range(2)), FreeBSD 13 (copy_file_range(2)), MacOS (copyfile(2)), and Windows (CopyFileExA).
* Fix macro name USE_COPY_FILE_RANGE.
* Minor bugfixes for userspace fallback.
* Fix linux sendfile64 bugs.
* Remove some overzealous asserts.
* Allow transmission-test-copy to take an optional argument for an external reference file.
* Fix return value error of tr_sys_path_copy.
* Use COPYFILE_ALL for Macs without COPYFILE_CLONE.
* Add in-kernel file copying for several platforms.
Numerous operating systems now have support for copying files directly in the
kernel, which is generally more efficient than copying in a userspace read(2)/
write(2) loop. (This becomes particularly relevant for 4th gen PCI-E storage,
which approaches the latency of DRAM.) For Linux I use sendfile64(2), and, for
later kernels, copy_file_range(2). FreeBSD 13 will also support
copy_file_range(2). MacOS has copyfile(2), and Windows has CopyFileExA.
Operating systems lacking such a syscall continue to use the existing
read(2)/write(2) loop.
* Appease uncrustify.
* Appease uncrustify.
* copy-test: generate random content at run time.
* copy-test: Stylistic changes and more check()s.
* copy-test: files_are_identical should follow test idioms
* tr_sys_path_copy: numerous tweaks as requested by review.
* s/old file/source file; s/new file/destination file.
* tr_sys_path_copy: handle win32 wide characters in paths.
* Uncrustify.
* test-copy: Use non-string create_file_with_contents.
* tr_sys_path_copy: numerous fixes.
Per review: generate test file content at runtime; tidy use of check();
fix style; re-measure file sizes in the copy; define a macro when the
system does not provide it; use Unicode APIs on Windows; and fix
documentation.
* Updated as per comments.
* Rebase kernel-copy changes onto 3.0 with gtest.
* Undo irrelevant comment change.
* Fix syntax error.
* Use tr_malloc() instead of tr_valloc().
* Use EXPECT instead of TR_ASSERT in gtest.
* Add error handling.
* Acceptable coding style has changed again.
Now it's camelCase. Also use nullptr instead of NULL, etc.
* Fix east/west const.
Co-authored-by: Mike Gelfand <mikedld@users.noreply.github.com>
* refactor: const correctness
* refactor: fix some implicit conversions
* refactor: make local pointers const if their objects are not modified
* refactor: do not cast away const in torrent-cell-renderer
* refactor: remove call to deprecated gtk_icon_size_lookup_for_settings
* refactor: member functions that do not mutate their objects should be declared const
* chore: do not end comments with a semicolon
* refactor: const correctness
* refactor: use getpwuid_r instead of getpwuid
* chore: simplify dict walking loop logic
* refactor: remove dead store assignment in announcer
* refactor: use std::make_shared
Remove static assert altogether: it should have worked everywhere anyway,
otherwise there is no point.
Add a few casts here and there to ensure proper type for directory handle.
Walk up one level at a time until the directory creation succeeds, then go
back down one level at a time. This reduces the number of operations in the
most common case (when directory already exists).
This way all the qualifiers (`const`, `volatile`, `mutable`) are grouped
together, e.g. `T const* const x` vs. `const T* const x`. Also helps reading
types right-to-left, e.g. "constant pointer to constant T" vs. "constant
pointer to T which is constant".
There're places where manual intervention is still required as uncrustify
is not ideal (unfortunately), but at least one may rely on it to do the
right thing most of the time (e.g. when sending in a patch).
The style itself is quite different from what we had before but making it
uniform across all the codebase is the key. I also hope that it'll make the
code more readable (YMMV) and less sensitive to further changes.
Defer validity checks until path gets to the remote side, where they
actually make sense. Add simple checks for download directory path to
ensure it's not relative, since one cannot know what current working
directory of the remote process is.
There're too many functions and types to consider, and benefits of not
using LFS macros aren't that big (I was thinking of using fts(3) but
that may not happen soon or at all).
Additionally,
* always close file descriptor on error in cached_file_open (FD leak),
* only store file descriptor to tr_cached_file on success,
* call ftruncate after xfsctl-based preallocation so that correct size
is reported by the system.