Fix nested_format_specs buffer context construction

[Tyagiquamar]

Problem

In nested_format_specs::write, buffer_ctx was constructed directly as FormatContext(basic_appender<Char>(buf), ctx.args(), ctx.locale()). When FormatContext is not parameterized with basic_appender<Char> (such as when custom contexts or different output iterators are passed, e.g. during compilation/custom formatters with nested formatted ranges like std::array or tuples), attempting to construct FormatContext directly with basic_appender<Char> leads to template instantiation and iterator mismatch errors (#4949).

Solution

• Explicitly instantiate basic_format_context<basic_appender<Char>, Char> for the temporary buffer context in nested_format_specs::write().

Validation

• Validated in Linux container with GCC 14:
  • Built and passed 100% of tests in ranges-test (42/42 tests) and compile-test (19/19 tests).

[vitaut]

This seems to move the mismatch from the output iterator to the argument pack. ctx.args() has type basic_format_args<FormatContext>, while the new buffer context expects basic_format_args<basic_format_context<basic_appender<Char>, Char>>, and there is no conversion between those when FormatContext is a generic_context.

Could you add the exact #4949 reproducer to compile-test.cc? I believe this still fails to instantiate.

[Tyagiquamar]

Thanks, addressed the review feedback in 8f23dec.

• Added the FMT_COMPILE fails to format std::array #4949 reproducer to compile-test.cc, including the width and dynamic element-spec cases. On the previous head it failed to instantiate: basic_format_args<generic_context<back_insert_iterator<string>, char>> does not convert to basic_format_args<context>.
• Nested width formatting now buffers through a context that keeps basic_format_args of the caller. A context that already writes with basic_appender is reused, so a nested range does not wrap that argument pack again.

Validation:

• Before the fix, compile-test failed to compile with GCC 13 (no known conversion for the argument pack).
• After the fix, compile-test 20/20 pass and ranges-test 42/42 pass (Ubuntu 24.04, GCC 13.3, Debug).

Ready for another look.

[vitaut]

Please don't copy other PRs. This just wastes everyone's time.

[Tyagiquamar]

Hi Victor, I'd like to clarify the timeline, as I think there may have been a misunderstanding.

I originally opened #4963 on October 1, before #4974 was opened on October 5. My October 7 commit addressed the argument-pack mismatch you pointed out in your earlier review.

I understand that the final implementations overlap significantly, and I respect your decision to consolidate the work into one PR. However, I didn't intentionally copy another contributor's work, and I wouldn't want to leave that impression.

Thanks for your time reviewing this.

Verified chronology

PR | Author | Opened (UTC) | Status -- | -- | -- | -- #4951 | linsen458-spec | September 18, 2026 | Closed, unmerged #4963 | Tyagiquamar (mine) | October 1, 2026 | Closed, unmerged #4974 | mayankjaiswal12 | October 5, 2026 | Open

Verified chronology PR Author Opened (UTC) Status [#4951](https://github.com//pull/4951) linsen458-spec September 18, 2026 Closed, unmerged [#4963](https://github.com//pull/4963) Tyagiquamar (mine) October 1, 2026 Closed, unmerged [#4974](https://github.com//pull/4974) mayankjaiswal12 October 5, 2026 Open