[WIP] changes for next release 3.11.2

[nlohmann]

This thread collects all changes made since the 3.11.1 (#3647) release.

Bug fixes

• Fix the value function which was broken for strings, size types, and nullptr in release 3.11.0. Regression: call to member function 'value' is ambiguous #3652 regression: .value<size_t> is compilation error. #3655 Reimplement value() access functions #3663
• Fix the json_fwd.hpp header to be self-contained and add it to the single-header release. json_fwd.hpp no longer standalone #3656 Amalgamate the forward declaration header #3679 Add amalgamated json-fwd.hpp to release #3687
• Fix regression that broke using json_pointer as key in associative containers. Add operator< to , to allow use as a map key #3680 Make json_pointer usable as map key (again) #3685
• Add missing constraint to deprecated JSON Pointer overloads of contains and at. Add forgotten is_basic_json guard to functions with json_pointer<Basi… #3658 Properly constrain non-string json_pointer overloads #3681
• Fix comparison between json_pointer and strings with == and !=. These comparisons worked in 3.10.5, but were broken in 3.11.0 and 3.11.1. Regression: no match for 'operator!=' comparing json_pointer and const char */string_t #3654 Add json_pointer/string_t equality comparison operators #3664
• Fix to_json conversion of std::vector<bool>::reference and std::vector<bool>::const_reference for STLs where these are the same as basic_json::boolean_t& and basic_json::boolean_t, respectively. Fix -Wignored-reference-qualifiers #3677 Fix 'const' qualifier on bool& has no effect #3678

Improvements

• Restructure inline namespace and allow version component to be disabled. See documentation. Restructure inline namespace and allow version component to be disabled #3683 MSVC natvis visualizer does not work after introduction of inline ABI namespace #3696 Add Python script for generating Natvis file and update file for 3.11.2 #3697 generate_natvis.py: validate version number; cleanup #3698
• Avoid heap allocations in BJData parser. Minor BJData fixes #3637

Documentation

• Publish documentation on every push to develop branch. Publish documentation on push to develop branch #3660 Refine 'Publish documentation' workflow #3673
• Add missing examples for the public API. Documentation change #3672 Documentation updates for 3.11.2 #3686
• Fix typo in json_pointer documentation. fix typo in json_pointer.md #3692

Community

• Add a badge for the Discord chat to the README file. The goal of that additional communication channel beyond the existing ones is to quickly coordinate between the contributors. Add Discord badge to README #3651
• Complete contributor list. Complete contributor list #3662 Complete contributor list #3670

CI

• Remove the macos-10.15 image from the CI as it is removed by GitHub Actions. macOS 10.15 Actions runner image deprecation #3612 Bump macos-10.5 runner to macos-11 and update Xcode versions #3615 Update CI #3626
• Remove hardcoded paths in Ubuntu workflow. Update CI #3626
• Only trigger AppVeyor builds if relevant files have been changed. Update CI #3626
• Fix CodeQL warning. Update CI #3626
• Harmonize naming of GitHub Actions jobs. Adjust naming of GitHub action jobs #3661
• Add labeler action to automatically add PR labels based on the changed files. Add labeler action #3671 Attempt to fix labeler permissions #3674 Fix whitespace in workflow files #3675
• Lint API documentation in the CI. Documentation change #3672
• Add local LGTM configuration and suppress warnings on third-party scripts. Miscellaneous small fixes #3643

Deprecations

• This release deprecations comparing JSON Pointers with strings via operator== and operator!=. To compare a json_pointer p with a string s, convert s to a json_pointer first and use json_pointer::operator== or json_pointer::operator!=. Deprecate json_pointer/string_t comparisons #3684

[falbrechtskirchinger]

Here's my list of PRs that are ready to go (possibly pending CI run), starting with the easy ones:

• Properly constrain non-string json_pointer overloads #3681
• Fix 'const' qualifier on bool& has no effect #3678
• Amalgamate the forward declaration header #3679
• Miscellaneous small fixes #3643

Easy but may need discussion:

• Deprecate json_pointer/string_t comparisons #3684
• Make json_pointer usable as map key (again) #3685

These are the bigger/more complicated ones:

• Reimplement value() access functions #3663
• Restructure inline namespace and allow version component to be disabled #3683

[falbrechtskirchinger]

Bug fixes, first sentence 3.10.0 -> 3.11.0:

Fix the value function which was broken for string and size types in release 3.11.0

Last sentence:
Fix to_json conversion of `std::vector<bool>::reference` and `std::vector<bool>::const_reference` for STLs where these are the same as `basic_json::boolean_t&` and `basic_json::boolean_t`, respectively.

(With extra backticks for your convenience :-))

[nlohmann]

I appreciate the extra backticks :)

[steffenb7333]

Please do not forget to update nlohmann_jsonConfigVersion.cmake.in (#3657 (comment)).

[falbrechtskirchinger]

And because of those inconsistencies, I feel same-version-compatibility would be unnecessarily limiting. 🤷‍♂️

• As a maintainer for multiple packages depending on each other I cannot even express that in a sane way without recompiling everything and anything using json.

Yup.

[steffenb7333]

A change of nlohmann_jsonConfigVersion.cmake.in towards same-version-compatibility would basically only affect packages citing json as a dependency via find_dependency and remove the inconsistency explained. The "normal" user, using the package via find_package would not be affected by the change.
Basically the current documented state is (API=major-version-compatibility, ABI=major-version-compatibility) whereas the reality is (API=major-version-compatibility, ABI=same-version-compatibility).

[falbrechtskirchinger]

find_package(nlohmann_json 3.9.0 REQUIRED)

I request version 3.9.0 because it introduced a feature I want to use. Newer versions would work for me as well. If v3.10.5 is installed, this line would fail with same-version-compatibility, would it not?

[steffenb7333]

You are correct and I see your point.
Nevertheless the current situation either breaks compiled libraries with nlohmann-symbols (and a later library update) or versioned find_package-calls.

I am out of ideas ATM.

[falbrechtskirchinger]

Indeed. Just note that things were already broken and the inline namespace will prevent known segfault scenarios.

[falbrechtskirchinger]

Fix regression that broke using json_pointer as object keys.

"as map key" or "as key in associative containers"