Fix format_as for tuple-like types with ranges.h

[dajiaohuang]

Fixes #4944

A tuple-like type can also provide an ADL format_as function. The tuple formattability probe should defer to that formatter instead of recursively checking tuple elements.

This change detects format_as before probing tuple elements, so the existing format_as formatter takes precedence.

Tests:

• Reproducer compile with MSVC /std:c++20
• Full CMake/CTest suite (22 tests)

[vitaut]

Thanks for the PR. I think format_as should indeed take precedence over tuple formatting, but I’d prefer not to change the semantics of is_tuple_formattable.

fmt::is_tuple_formattable is an exported trait and currently answers whether the tuple elements are formattable. With this change it becomes false for every tuple-like type that has format_as, even if all of its elements are formattable. Could we instead keep is_tuple_formattable_ unchanged and exclude format_as when selecting the tuple formatter, while still making sure the tuple formattability check is not instantiated in that case?

Nit: could you also replace the \n escapes in the PR description with actual newlines? The current description is a bit odd to read/edit.

[dajiaohuang]

Updated in 4d5286b (amended after CI showed the public tuple-formattability value is intentionally unchanged for this recursive tuple-like test type).

• Restored fmt::is_tuple_formattable to its existing tuple-element semantics.
• Added a formatter-only trait that skips the tuple probe when ADL format_as is available, so the dedicated format_as formatter still takes precedence without instantiating the recursive check.
• Kept the existing runtime regression test for a tuple-like format_as type.
• Rewrote the PR description with actual newlines.

Validation: git diff --check passed. A local CMake/compiler toolchain is not installed in this workspace; the repository CI will provide the full build/test validation.

[dajiaohuang]

Implemented and pushed commit 57dc090. This preserves the public fmt::is_tuple_formattable trait semantics while keeping formatter dispatch excluded for format_as tuple-like types, and adds a regression assertion covering the distinction. git diff --check passes. Local CMake/CTest are unavailable in this environment, so please rely on the remote CI for the full test suite.

[dajiaohuang]

Follow-up: corrected the regression assertion to pass the required char parameter to the exported fmt::is_tuple_formattable<T, C> trait (commit 13e3aa7). The preceding CI failure was only this test compile error; the fix is now pushed. Remote CI should rerun.

[dajiaohuang]

Follow-up: the exported trait assertion now uses static_assert (commit c462af2), avoiding C++11/14 ODR-use linker errors while still verifying fmt::is_tuple_formattable<format_as_tuple_like, char>::value remains true. The prior CI failures were limited to that assertion form.

[vitaut]

Mostly looks good, just one more inline comment.

[vitaut]

This can be folded into is_tuple_formattable_for_formatter.

[dajiaohuang]

Folded the format_as detection into is_tuple_formattable_for_formatter so the helper directly defaults to the exported tuple trait and specializes to false only when format_as_result<T> is valid. Pushed as 83c869a; git diff --check passes.

[vitaut]

Merged, thanks