basic conversion to C++20 #2645
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #2645 +/- ##
==========================================
+ Coverage 64.62% 64.65% +0.02%
==========================================
Files 104 104
Lines 22239 22235 -4
Branches 10911 10909 -2
==========================================
+ Hits 14371 14375 +4
+ Misses 5626 5622 -4
+ Partials 2242 2238 -4 ☔ View full report in Codecov by Sentry. |
neheb
force-pushed
the
h
branch
3 times, most recently
from
e68ae8a
to
4cddaa6
Compare
|
Hrm I’m unsure of this PR. Compiler minimums have to increase for everything. |
neheb
force-pushed
the
h
branch
2 times, most recently
from
011b5e9
to
84e45a3
Compare
For library code it is usually good to use newer standards only in the implementation, but not in the interface. However I do not know how to map this in Windows or whenever something links to features you need a newer standard C++/template library because then people still need to upgrade. So the safe way would be to make the library stuff compatible from whatever minimum C++ standard version is desired (11/14/17) but pushing all users to C++20 might make exiv2 unsuitable for many of those who use it today. |
|
Right. The headers are still kept at C++11. On that note, if constexpr should be removed from them as compilers are not free to add it implicitly as I originally thought, making the templates much less efficient. As far as Windows is concerned, that's not an issue. In the case of MSVC, the libraries are upgrade-able, I actually don't know if older MSVC can even compile exiv2. I remember MSVC2019 or something just crashing when trying to compile exiv2. No errors printed. On that note, there's currently an issue where Exiv2 is incompatible with msvcrt. Given that it's deprecated by Microsoft, I see no value in supporting it. See #2637 |
neheb
force-pushed
the
h
branch
5 times, most recently
from
9c58408
to
676aac1
Compare
|
Btw, might be an idea to hold off on this for a couple of 0.28.x bugfix releases, just so it remains easy to backport stuff from the main branch? |
|
This is still draft after all. |
neheb
force-pushed
the
h
branch
4 times, most recently
from
21cbd59
to
959b4df
Compare
|
Yeah I'm not sure of this PR anymore. People are still using GCC7 to compile Exiv2. std::endian comes with GCC8 and starts_with GCC9. Let's not mention other projects that are on C++14 to allow building with CentOS7. This is going to be a draft for a while.... |
Signed-off-by: Rosen Penev <rosenp@gmail.com>
18.04 is now EOL. 20.04 comes with 9 standard. Allows building with C++20. Signed-off-by: Rosen Penev <rosenp@gmail.com>
Signed-off-by: Rosen Penev <rosenp@gmail.com>
There's std::endian now. Signed-off-by: Rosen Penev <rosenp@gmail.com>
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| template <size_t N, const char* const (&keys)[N]> | ||
| ExifData::const_iterator findMetadatum(const ExifData& ed) { | ||
| for (const auto& k : keys) { | ||
| auto pos = ed.findKey(ExifKey(k)); |
Check warning
Code scanning / CodeQL
NULL iterator deref Warning
No description provided.