C++17/C++20 support

[COM8]

We should update the complete library to be C++20 or at least C++17 ready.

My plan is to make use of C++17 features like std::opt and the new default deleter for std::unique_ptr and std::shared_ptr. Besides that create a better RAII wrapper for curl sessions to properly support multithreaded access. Perhaps even offer support for lambdas and coroutines.

Since it's rather early for C++20 features in such a library, this will be optional stuff and compilations should be possible with C++17 and C++20.

The old CPR (<C++17) will receive from then one only bug fixes. New features will only be added to the C++17 branche (master).

[zamazan4ik]

I cannot understand the reason? Why? Do you want port whole CPR project to C++17 or even C++20? Remember, that a lot of users don't have even C++17 in their applications. So if you update the library to such new C++ standard, a lot of users will not be able to use CPR.

Do you need any specific stuff from C++17/20?

My personal point of view - that's acceptable to grade someday to C++17, but for now that's totally unacceptable grade the library to C++20 (since it's even officially released and C++ compilers have poor support for C++20). Please - don't upgrade cpr to C++20 now.

Thank you.

[Chlorie]

I cannot understand the reason? Why? Do you want port whole CPR project to C++17 or even C++20? Remember, that a lot of users don't have even C++17 in their applications. So if you update the library to such new C++ standard, a lot of users will not be able to use CPR.

Do you need any specific stuff from C++17/20?

My personal point of view - that's acceptable to grade someday to C++17, but for now that's totally unacceptable grade the library to C++20 (since it's even officially released and C++ compilers have poor support for C++20). Please - don't upgrade cpr to C++20 now.

Thank you.

I think what they meant was to make CPR better integrated with C++17/20 if your build settings support those, like with version detection macros so that if you use C++17/20 CPR will support new fancy things (like coroutines?) but if your compiler don't have the support those parts of the library are simply not accessible (disabled by preprocessor). I don't see how this will be a problem for people using older compilers.

[COM8]

My plan is to make use of C++17 features like std::opt and the new default deleter for std::unique_ptr and std::shared_ptr. Besides that create a better RAII wrapper for curl sessions to properly support multithreaded access. Perhaps even offer support for lambdas and coroutines.

Since it's rather early for C++20 features in such a library, this will be optional stuff and compilations should be possible with C++17 and C++20.

The old CPR (<C++17) will receive from then one only bug fixes. New features will only be added to the C++17 branche (master).

[HellsingDarge]

Also being able to pass string_view into cpr::Url and other applicable locations would also be nice

[COM8]

Also being able to pass string_view into cpr::Url and other applicable locations would also be nice

Yes, good idea!

[xloem]

I cannot understand the reason? Why? Do you want port whole CPR project to C++17 or even C++20? Remember, that a lot of users don't have even C++17 in their applications. So if you update the library to such new C++ standard, a lot of users will not be able to use CPR. Do you need any specific stuff from C++17/20? My personal point of view - that's acceptable to grade someday to C++17, but for now that's totally unacceptable grade the library to C++20 (since it's even officially released and C++ compilers have poor support for C++20). Please - don't upgrade cpr to C++20 now. Thank you.

I think what they meant was to make CPR better integrated with C++17/20 if your build settings support those, like with version detection macros so that if you use C++17/20 CPR will support new fancy things (like coroutines?) but if your compiler don't have the support those parts of the library are simply not accessible (disabled by preprocessor). I don't see how this will be a problem for people using older compilers.

Just to note, it does look like the community disagrees a little here. COM8 seemed to be proposing to stop providing feature updates for c++11/c++14, possibly due to the limits of their capacity being maintainer. I wonder if there is enough community support to sustain maintenance for older compilers, while still offering features for new. Could others possibly step in to backport and forward-port features, or sort out preprocessor defines? If not, we could always propose that one of the variants become a fork.

[acidicoala]

C++ 17 support is wholeheartedly supported, because for now we have to suppress CXX17 deprecation warnings, that prevent compilation on MSVS.

[shyney7]

Please dont make C++20 mandatory. Many systems still dont have a c++20 compiler and it gets even worser if you do cross-compilation for specific embedded systems.

[zydxhs]

I have never used C++20, and I will not consider C++20 for a long time. C++17 is quite new and acceptable.

[simon-berger]

I would like to work on this issue. My plan is to work out a concept first and then update CPR with as few compatibility restrictions as possible.

[simon-berger]

Before I start to work on a PR I would like to discuss some of my ideas.

As far as I understand, the plan is to use C++17 as default (required) and if needed support optional C++20 compilation to enable certain special features. Accordingly, I would increase the CMake CXX_STANDARD to 17 first and consider C++20 only after C++17 has been integrated.

Here is a tentative list of potential improvements / features that I have in mind to implement as part of the update to C++17:

• Implement std::string_view support (e.g. for cpr::Url as proposed by @HellsingDarge)
• Improve lambda functions:
  • Type determination with auto
  • Initialization of function-local constants directly in captures
• Use of std::optional (e.g. for ranges which are only partially defined)
• Use the new filesystem library
• Use std::byte instead of char / unsigned char
• Support std::chrono_literals for timeouts
• Use auto as return type for functions without decltype
• Use of make_unique()
• ~~Make cpr::Session thread safe using std::scoped_lock~~
• Investigate lambda capture of *this (might be interesting for async requests)
• ...

@COM8 can you briefly explain me what you meant by that:

My plan is to make use of C++17 features like [...] the new default deleter for std::unique_ptr and std::shared_ptr.

Feedback is very appreciated :)

[COM8]

Yes. cpr 1.9.0 won't change anything in regards to the minimum required version of C++. BUT for 1.10.0 we will increment it to cpp17 with optional cpp20 features (CMake option).

All your suggestions sound good. Only making cpr::Session thread safe is not needed in my eyes, since using a session in parallel is a broken concept in my eyes.

Regarding the default deleter: This already has been fixed, when we moved the code to cpp11. There default_delete got introduced. Therefore this is no concern any more.