On Tue, 2 Dec 2025 at 15:09, Yuao Ma <[email protected]> wrote:
>
> Hi Patrick,
>
> On Tue, Dec 2, 2025 at 3:27 AM Patrick Palka <[email protected]> wrote:
> >
> > On Sat, 22 Nov 2025, Yuao Ma wrote:
> >
> > > On Sat, Nov 22, 2025 at 12:33 AM Tomasz Kaminski <[email protected]>
> > > wrote:
> > > >
> > > >
> > > >
> > > > On Fri, Nov 21, 2025 at 5:11 PM Yuao Ma <[email protected]> wrote:
> > > >>
> > > >> On Fri, Nov 21, 2025 at 6:32 PM Tomasz Kaminski <[email protected]>
> > > >> wrote:
> > > >> >
> > > >> > There was a paper submitted before Kona: p3862r0 Postpone
> > > >> > basic_string::subview and wait for cstring_view [1],
> > > >> > that according to github papers status [2] have also NB comment. I
> > > >> > was not able to locale the comment thou.
> > > >> > So maybe it would be good to have this as patch series for
> > > >> > string_view (that would contain feature test macro) and then string,
> > > >> > so we can revert later?
> > > >> >
> > > >> >
> > > >> > [1]
> > > >> > https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2025/p3862r0.html#
> > > >> > [2] https://github.com/cplusplus/papers/issues/2466
> > > >> >
> > > >>
> > > >> Thanks for the info! Could you elaborate more on how making this a
> > > >> patch series helps with reverting? My understanding of P3862R0 is that
> > > >> it proposed two wording options: one to entirely remove the subview
> > > >> for string and string_view, and the other to remove the default
> > > >> parameter argument.
> > > >
> > > > Both wording options are changing only basic_string::subview,
> > > > and not touching subview on string_view at all.
> > >
> > > Oh yeah, my mistake. Split into two patches as below.
> >
> > > From 668363e4dbf6db6d7cd40899a480256936dcd0ae Mon Sep 17 00:00:00 2001
> > > From: Yuao Ma <[email protected]>
> > > Date: Sat, 22 Nov 2025 00:37:48 +0800
> > > Subject: [PATCH 1/2] libstdc++: implement P3044R2 - sub-string_view from
> > > string (string_view part)
> > >
> > > libstdc++-v3/ChangeLog:
> > >
> > > * include/bits/version.def: Add string_subview FTM.
> > > * include/bits/version.h: Regenerate.
> > > * include/std/string_view: Add subview.
> > > *
> > > testsuite/21_strings/basic_string_view/operations/subview/char.cc: New
> > > test.
> > > *
> > > testsuite/21_strings/basic_string_view/operations/subview/wchar_t.cc: New
> > > test.
> > > ---
> > > libstdc++-v3/include/bits/version.def | 9 ++++
> > > libstdc++-v3/include/bits/version.h | 10 ++++
> > > libstdc++-v3/include/std/string_view | 10 +++-
> > > .../operations/subview/char.cc | 52 +++++++++++++++++++
> > > .../operations/subview/wchar_t.cc | 52 +++++++++++++++++++
> > > 5 files changed, 132 insertions(+), 1 deletion(-)
> > > create mode 100644
> > > libstdc++-v3/testsuite/21_strings/basic_string_view/operations/subview/char.cc
> > > create mode 100644
> > > libstdc++-v3/testsuite/21_strings/basic_string_view/operations/subview/wchar_t.cc
> > >
> > > diff --git a/libstdc++-v3/include/bits/version.def
> > > b/libstdc++-v3/include/bits/version.def
> > > index 29ecf15c7e3..b5575d2399f 100644
> > > --- a/libstdc++-v3/include/bits/version.def
> > > +++ b/libstdc++-v3/include/bits/version.def
> > > @@ -1934,6 +1934,15 @@ ftms = {
> > > };
> > > };
> > >
> > > +ftms = {
> > > + name = string_subview;
> > > + values = {
> > > + v = 202506;
> > > + cxxmin = 26;
> > > + hosted = yes;
> >
> > I think we need to add cxx11abi = yes if we don't support this with the
> > COW string. Wouldn't adding COW string support amount to a few extra
> > lines in cow_string.h though?
> >
>
> Yes I added cxx11abi = yes in the latest version of the patch:
> https://gcc.gnu.org/pipermail/libstdc++/2025-November/064522.html
Why not just support it for the old ABI though?
Making it cxx11abi=yes means that basic_string_view::subview is not
available, and that makes no sense.
>
> > > + };
> > > +};
> > > +
> > > ftms = {
> > > name = to_underlying;
> > > values = {
> > > diff --git a/libstdc++-v3/include/bits/version.h
> > > b/libstdc++-v3/include/bits/version.h
> > > index 5901d27113d..413da56b088 100644
> > > --- a/libstdc++-v3/include/bits/version.h
> > > +++ b/libstdc++-v3/include/bits/version.h
> > > @@ -2161,6 +2161,16 @@
> > > #endif /* !defined(__cpp_lib_string_resize_and_overwrite) */
> > > #undef __glibcxx_want_string_resize_and_overwrite
> > >
> > > +#if !defined(__cpp_lib_string_subview)
> > > +# if (__cplusplus > 202302L) && _GLIBCXX_HOSTED
> > > +# define __glibcxx_string_subview 202506L
> > > +# if defined(__glibcxx_want_all) ||
> > > defined(__glibcxx_want_string_subview)
> > > +# define __cpp_lib_string_subview 202506L
> > > +# endif
> > > +# endif
> > > +#endif /* !defined(__cpp_lib_string_subview) */
> > > +#undef __glibcxx_want_string_subview
> > > +
> > > #if !defined(__cpp_lib_to_underlying)
> > > # if (__cplusplus >= 202100L)
> > > # define __glibcxx_to_underlying 202102L
> > > diff --git a/libstdc++-v3/include/std/string_view
> > > b/libstdc++-v3/include/std/string_view
> > > index 842f6ad89af..b226544fa6f 100644
> > > --- a/libstdc++-v3/include/std/string_view
> > > +++ b/libstdc++-v3/include/std/string_view
> > > @@ -40,9 +40,10 @@
> > > #define __glibcxx_want_constexpr_char_traits
> > > #define __glibcxx_want_constexpr_string_view
> > > #define __glibcxx_want_freestanding_string_view
> > > -#define __glibcxx_want_string_view
> > > #define __glibcxx_want_starts_ends_with
> > > #define __glibcxx_want_string_contains
> > > +#define __glibcxx_want_string_subview
> > > +#define __glibcxx_want_string_view
> > > #include <bits/version.h>
> > >
> > > #if __cplusplus >= 201703L
> > > @@ -342,6 +343,13 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
> > > return basic_string_view{_M_str + __pos, __rlen};
> > > }
> > >
> > > +#ifdef __glibcxx_string_subview // >= C++26
> > > + [[nodiscard]]
> > > + constexpr basic_string_view
> > > + subview(size_type __pos = 0, size_type __n = npos) const
> > > + { return substr(__pos, __n); }
> > > +#endif
> >
> > IIUC since subview is specified as freestanding-deleted, we need to
> > define it as deleted if we don't support it in freestanding mode.
> > So perhaps why don't we just always define it in C++26 mode regardless
> > of whether the FTM is defined? Maybe I'm not understanding the
> > intent of freestanding-deleted though.
> >
>
> IMHO __glibcxx_string_subview is guarded under HOSTED. Should I
> explicitly mark this as deleted in freestanding mode?
We should just support it for freestanding. It's only marked
freestanding deleted because it might throw an exception, and in the
standard everything that can throw an exception is not included in
freestanding. But we can just abort instead of throwing. We already
support basic_string_view::substr, basic_string_view::compare etc.
So I think we don't want the macro to depend on cxx11abi or on hosted.
Then the basic_string_view::subview member will always be defined for
C++26.
