| commit | 69bd3e9d9376152b9c1f6c9ea01e2b4a96baa752 | |
| author | Marcelo Bezerra <23555060+mmosca@users.noreply.github.com> | |
| Thu, 25 May 2023 11:12:03 +0000 (25 13:12 +0200) | ||
| committer | GitHub <noreply@github.com> | |
| Thu, 25 May 2023 11:12:03 +0000 (25 13:12 +0200) | ||
| tree | 42f57f14e1f936330c3a18ba1d20b12cd78128c5 | treesnapshot (tar.gz zip) |
| parent | ccd75e7a2e69443bc2a8acca83e380a08840e58a | commitdiff |
[SITL OSX] Fix some of the warnings and add macosx SITL build to workflows (#9063)
clang can be verbose with warnings, but some of it is probably valid when building a 64bit binary.
Highlighted changes:
* Call float versions of math functions to avoid conversion to double by the compiler (absf, sqrtf, roundf, etc)
* Make sure floating point constants are marked as floats, to avoid conversion to double by the compiler. (1.0 is a double, 1.0f is a float and when doing math between float and double, all values get upgraded to double!)
* Changed memcpy_fn in unit test AND SITL builds to be a macro to memcpy (instead of inline function calling memcpy), this fixes linker errors for memcpy as macos compiler mangles the symbol in a different way and would not work with asm("memcpy") call.
* Some simulator code made heavy use of doubles, but since all the data in INAV is float, that is probably overkill and some functions/macros had float in the name, while upconvertting to double.
Open questions:
* Should scale in osdFormatCentiNumber be changed to float? It is currently uint32_t but some of the scale defines are, correctly, not integer numbers.
* I changed CENTIDEGREES_TO_DEGREES to use float on the division path, but the code seems to be ok, or assuming it would be converted to integer anyway. Is this the correct solution?
* This still does not fix the invalid settings issues, I suspect it is related to the lack of linker scripts in clang, to preserve the section data of settings.
* Binary is still not multi platform (arm/x86_64).
clang can be verbose with warnings, but some of it is probably valid when building a 64bit binary.
Highlighted changes:
* Call float versions of math functions to avoid conversion to double by the compiler (absf, sqrtf, roundf, etc)
* Make sure floating point constants are marked as floats, to avoid conversion to double by the compiler. (1.0 is a double, 1.0f is a float and when doing math between float and double, all values get upgraded to double!)
* Changed memcpy_fn in unit test AND SITL builds to be a macro to memcpy (instead of inline function calling memcpy), this fixes linker errors for memcpy as macos compiler mangles the symbol in a different way and would not work with asm("memcpy") call.
* Some simulator code made heavy use of doubles, but since all the data in INAV is float, that is probably overkill and some functions/macros had float in the name, while upconvertting to double.
Open questions:
* Should scale in osdFormatCentiNumber be changed to float? It is currently uint32_t but some of the scale defines are, correctly, not integer numbers.
* I changed CENTIDEGREES_TO_DEGREES to use float on the division path, but the code seems to be ok, or assuming it would be converted to integer anyway. Is this the correct solution?
* This still does not fix the invalid settings issues, I suspect it is related to the lack of linker scripts in clang, to preserve the section data of settings.
* Binary is still not multi platform (arm/x86_64).
42 files changed: