Re-amalgamate single_include

[nlohmann]

single_include/nlohmann/json.hpp on develop is stale: forcing make amalgamate on e5a89d6 changes 259 lines. All of it comes from #5737, which changed 13 headers under include/ and merged without the matching amalgamation. The regenerated diff equals git diff b54ed188e e5a89d671 -- include/ (+164/−95), and every spot-checked hunk traces back to #5737 with git log -S. The PRs merged just before it (#5548, #5729, #5731, #5732) updated single_include correctly. json_fwd.hpp and json_literals.hpp were already up to date.

What single_include was missing from #5737:

• removal of the GCC C++20 -Wignored-attributes pragma block from Fix GCC C++20 modules compilation #5103 #5164. This was intentional (Lint and static-analysis config: clang-tidy checks disabled "to get the CI going", no-op scan-build job, stale NOLINTs, library-wide warning suppressions in tests, stale GCC flags and frozen clang-tidy image #5725 item 9): the pragma could not suppress the hard errors in Build issue when compiling with gcc (any modules-capable version) in a C++ modules project #5103. It was removed outright, not moved. The note in src/modules/json.cppm predates it (Fix GCC C++20 modules compilation #5103 #5164).
• removal of the global clang -Wdocumentation push/pop, and the doc-comment fixes it made possible (\\u escapes, @complexity → Complexity:, @liveexample, @tparam in @return, unescape's @param[in,out])
• from_json tuple/pair/array helpers now take const BasicJsonType& instead of forwarding. The forwards were no-ops because at() always returns an lvalue.
• output_adapter no longer names std::basic_string/std::basic_ostream for non-character CharType (Lint and static-analysis config: clang-tidy checks disabled "to get the CI going", no-op scan-build job, stale NOLINTs, library-wide warning suppressions in tests, stale GCC flags and frozen clang-tidy image #5725 item 2), plus the -Wdocumentation-deprecated-sync suppression around the deprecated sax_parse overload
• NOLINT updates (c-array aliases in the enum macros, bugprone-std-namespace-modification, misc-anonymous-namespace-in-header, portability-template-virtual-member-function, the double forward in container_input_adapter_factory), and dropping LCOV_EXCL_LINE from the json_pointer out_of_range.410 throw

Why this went unnoticed, and the second commit

#5737 merged 39 seconds after its last push. Its check_amalgamation run was still queued, and earlier runs had been cancelled by the concurrency group, so no failing check was ever reported. check_amalgamation.yml only runs on pull_request, so nothing on develop caught it afterwards either.

The second commit adds push: branches: [develop] to check_amalgamation.yml:

• in-progress runs are cancelled only for PRs, never for develop
• the save job (PR number/author) only runs for pull requests
• the checkout uses github.event.pull_request.head.sha || github.sha
• comment_check_amalgamation.yml only comments for PR-triggered runs, because push runs have no PR and no pr artifact

A failure on develop then shows up as a red check on the merge commit, with the ready-to-apply amalgamation-patch artifact. This doesn't stop the merge itself; only making the check a required status check would do that. If you'd rather keep the workflow PR-only, drop the second commit.

Public API

No breaking changes. single_include now matches include/.

This PR was written by Claude Code.

🤖 Generated with Claude Code