Re: [PATCH v2 1/2] libstdc++: use copy_file_range, improve sendfile in filesystem::copy_file

[Jonathan Wakely <jwakely@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1564 bytes --]

On Mon, 20 Mar 2023 at 15:16, Jonathan Wakely wrote:

>
>
> On Wed, 15 Mar 2023 at 19:30, Jannik Glückert via Libstdc++ <
> libstdc++@gcc.gnu.org> wrote:
>
>> This iteration improves error handling for copy_file_range,
>> particularly around undocumented error codes in earlier kernel
>> versions.
>> Additionally this fixes the userspace copy fallback to handle
>> zero-length files such as in
>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108178.
>>
>> Lastly, the case "src gets resized during the copy loop" is now
>> considered and will return true once the loop hits EOF (this is the
>> only situation, aside from a zero-length src, where sendfile and
>> copy_file_range return 0).
>>
>
> I've applied this patch (with some whitespace fixes) and started testing
> it. I'm seeing some regressions:
>
> FAIL: experimental/filesystem/operations/copy.cc execution test
> FAIL: experimental/filesystem/operations/copy_file.cc execution test
>
> The failures in the log look like:
>
> terminate called after throwing an instance of
> 'std::experimental::filesystem::v1::__cxx11::filesystem_error'
>  what():  filesystem error: cannot copy: Input/output error
> [filesystem-test.copy.2900321341.ySWT77]
> [filesystem-test.copy.2900321342.vjeAar]
> FAIL: experimental/filesystem/operations/copy.cc execution test
>

Not just the Filesystem TS versions of those tests, but the std::filesystem
ones too:

FAIL: 27_io/filesystem/operations/copy.cc execution test
FAIL: 27_io/filesystem/operations/copy_file.cc execution test