[PATCH v2 0/3] libstdc++: Add _Streambuf_sink and specialize _Iter_sink for ostreambuf_iterator

[Tomasz Kaminski]

Thanks for the update.
I have left more review feedback. If anything is unclear, please feel free
to reach out for clarification
question.

On Thu, Jul 9, 2026 at 1:19 PM Anlai Lu <agicy@qq.com> wrote:

> This v2 addresses all review comments from Tomasz Kaminski on v1:
>
> - Max/counting logic moved entirely from _Streambuf_sink (base) into
>   _Iter_sink<ostreambuf_iterator> (specialization).  _Streambuf_sink
>   is now a pure I/O layer with no counters, no _M_max, and no _S_failed
>   state - failure is tracked solely by ostreambuf_iterator::_M_failed.
>
> - _M_discarding() returns false.
>   format_to_n needs the formatter to keep producing output so we can
>   count the total hypothetical output length.
>
> - _M_overflow() is overridden in the _Iter_sink specialization,
>   counting all characters and adjusting what is actually
>   committed (via _M_put or pbump) to stay within the limit,
>   without shrinking the buffer.
>
> - A protected _M_pbump helper in _Streambuf_sink avoids derived
>   classes needing direct friendship to basic_streambuf.
>
> The _Streambuf_sink itself is simplified: no _M_bump override, no
> _M_commit_put_area, no _M_do_flush.  Put-area writes accumulate
> in _M_used() and are committed in _M_overflow via a single pbump.
> The _Iter_sink specialization handles all counting and truncation.
>
> Performance (header-overlay A/B, single core, BABA interleave):
>
>   scenario                        16 B    64 B    256 B   1024 B  4096 B
>   print(ostream, "{}", x)         0.98x   1.01x   1.02x   1.02x   1.02x
>   format_to(back_inserter, ...)   1.02x   0.98x   1.00x   0.93x   0.96x
>   format_to(ostreambuf_iter, ...) 1.51x   3.16x  12.59x  51.53x 121.07x
>     with width/padding spec       1.08x   1.10x   1.09x   1.08x   1.06x
>     multi-arg ("_{}_...")         1.21x   2.18x   2.56x   4.60x  14.38x
>     append mode                    -      3.82x  11.38x  32.33x     -
>     to filebuf (/dev/null)        1.58x   4.30x  10.36x  19.65x  22.12x
>     to streambuf w/o put area       -       -    16.74x  17.12x  17.87x
>   format_to_n(ostreambuf_iter,
>     max=len/4, ...)                 -     1.40x   4.23x   6.79x     -
>     max=len*2, ...)                 -     4.33x  14.26x  52.49x     -
>
> Perf stat for the primary fmt_to path:
>
>   len    instr%  cycles%  b_IPC  a_IPC
>   16     77.2%    63.1%   2.21   2.70
>   64     39.5%    31.5%   1.61   2.01
>   256    14.6%     7.9%   1.36   2.52
>   1024    4.1%     1.8%   1.34   3.08
>   4096    1.8%     0.8%   1.33   2.86
>
> Anlai Lu (3):
>   libstdc++: Add tests for format_to with ostreambuf_iterator
>   libstdc++: Add _Streambuf_sink for direct streambuf formatting
>   libstdc++: Specialize _Iter_sink for ostreambuf_iterator
>
>  .../include/bits/streambuf_iterator.h         |   7 +
>  libstdc++-v3/include/std/format               | 184 ++++++++++++++++
>  libstdc++-v3/include/std/streambuf            |   4 +
>  .../format/functions/format_to_ostreambuf.cc  | 244 ++++++++++++++++++
>  4 files changed, 439 insertions(+)
>  create mode 100644
> libstdc++-v3/testsuite/std/format/functions/format_to_ostreambuf.cc
>
> --
> 2.34.1
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://gcc.gnu.org/pipermail/libstdc++/attachments/20260709/f3c22f5c/attachment-0001.htm>