- This is going to start the mark of v4
- pairs() implemented on tables for pairs-like iteration
- .error_handler variable is now hidden from view on protected function! Use set_error_handler/get_error_handler
- packaged_coroutine skeleton, but code is blacked out
- tests rearranged a little bit
- auxiliary keys lookup, search, and storage improved
- overhead of stored tables improved
- usertype registration and un-registration improved
- Make sure multiple states don't interfere with each other's keys
- Make sure re-registration of the same type does not cause too many problems
- Make sure the variadics test is present
- Make sure the storage keys are cleared out properly
- Ensure stack field_get/field_set has consistent metaprogramming for needing a pushed table, and at what offset
- Avoid duplicate logic in table_core for the same logic by deferring to detail helper
- Fixes#1095
- Not all were done, though; gave up at enumerations in an enum class being marked as shadowers.
- Which is just the dumbestt hing I've ever had to deal with. At least Clang is smarter!
- Inheritance was causing __index and __new_index to be overriden on the named metatables
- This meant special constructors and other metamethods might not have been set properly
- Use lua_rawset to directly slam dunk the value into the table
- add noexcept to some of the core function calls
- type usage for both trampolines and yields can now be fully tracked, at the expense of more template instantiations when using both
- exceptions should be less prone to explosion in C versions, but will break in C++ code (the trampolines need to be modified for usertype calls to avoid this problem, specifically)
- Add a new feature for SOL_FUNCTION_CALL_VALUE_SEMANTICS. Fixes#1059
- Detect deprecated/removed bit32 library. Fixes#1086
- Format the entire repository. Close#1073
- SOL_LUAJIT was not being properly defined. Close#1082
- Fix alignment issue. #1071
- readonly(...) should no longer issue errors in bogus manners
- binding for member function pointers and normal pointers should behave better now
- coroutines should now properly transfer the state/identity of their error handlers
- Fix empty base class optimization type
- Improve bind traits metaprogramming
- Add new define to catch-all for exception propagation
- Improve natvis
- Move test defines to build system, as we should
- Add a new example for multiple thread/states in Coroutines
- Add proper direct to mangling in Clang for Windows
- Exceptions are still busted on some platforms (Clang being quite funky)
Rationale: since sol2 is a header-only library, some projects might
use it as a private dependency instead of exposing it through their API.
Thus, it is useful to have an option to not install sol2, in order to
reduce the installation size as well as to avoid accidentally exposing
internal dependencies in the top-level project.
The project I work on is a demonstration of how the feature might be
used:
f4a20ec39e/external/CMakeLists.txt (L84)
(mind the different name of the CMake variable, I renamed it to fit the rest
of the sol options convention)
We have an install-check script script which verifies that only expected
headers, libs, targets and docs are exported (the one which are interesting
from users PoV). I wouldn't mind contributing the script as github workflows
if you want to use it as well - please provide feedback if that's desired.
Here is a link to our script for preview:
https://github.com/GENIVI/ramses-logic/blob/master/ci/scripts/installation-check/check-installation.py
While modifying the CMake code, I noticed a 4-space instead of TAB in one
place and fixed it to be consistent with the rest of the CMake code.
- Not at all type safe: there should be some investigation into making it less unsafe to work with these things (albeit it looks like it would cost +1 pointer to serialize a string name for each callable in Lua, at LEAST)
- Must be opted into - see the documentation
- Fixes#1015
- A few drive-by fixes here and there
- New configuration test harness with CMake
- Weed out bugs in the numeric checking implementation
- Solve the problem with load_result/protected_function_result/unsafe_function_result being copyable and thus amenable to explosions. Fixes#995.
- Resolve the warning in sol::readonly. Fixes#1000.
- Looke into #1014. Not sure I can help there, honestly; looks like a mix up of multiply-loaded libraries and mixing the two.
std::is_const<T> will always be false if std::is_lvalue_reference<T> is true, we first have to strip off the reference to look at the const-ness. See https://en.cppreference.com/w/cpp/types/is_const
- Just let MSVC have crap codegen. The compiler sucks anyways. Fixes
- Add incremental GC power. Fixes#997.
- Add non-null offset manipulation. Did I mention I hate VC++ and how it implemented _Aligned? Fixes#1017.
Add a note to the documentation about SOL_NO_LUA_HPP being less preferred over the C++ defines, to Fix#1013.
Add regression tests (not yet fixed) so I can get to the bottom of VC++ being ass.
