Fix for buffer overflow in fp_traits.hpp on aarch64 / mingw #1348

Activity

1. ckormanyos commented on Jan 10, 2026

   

   ckormanyos

   on Jan 10, 2026

   Member

   More actions

   Hi @joostn, thanks for this good find.

   We can handle this for sure. But we should discuss a few details in order to find the best way to handle this issue.

   It looks like you are getting an error via -Werror regarding a suspected buffer overflow. Some similar warnings we encountered in the past have proven to be false negative, bogus warnings.

   So we need to see if the warning can safely be disabled for this compiler and code sequence. if so, then we simply disable the warning within a PP-block in the header itself.

   So just to be sure, without -Werror, is this just a warning?
   Then we can look at the warning and try to determin if it is real/bogus.
   Then potentially disable.
2. ckormanyos commented on Jan 10, 2026

   

   ckormanyos

   on Jan 10, 2026

   Last edited by ckormanyos

   Member

   More actions

   And yes, the hard-coded number 4 does look like it might simply be wrong for this architecture. We need to catch that if this is the case. Then that would be a real warning/error and not a bogus one.
3. jzmaddock commented on Jan 11, 2026

   

   jzmaddock

   on Jan 11, 2026

   Collaborator

   More actions

   And yes, the hard-coded number 4 does look like it might simply be wrong for this architecture. We need to catch that if this is the case. Then that would be a real warning/error and not a bogus one.

   It's based on the assumption that the final #else is always a 128-bit long double, but this DR proves that not to be true, so maybe we should add a bit more error checking... I'll file a PR.

   👍React with 👍1Reacted by Christopher Kormanyos
4. joostn commented on Jan 11, 2026

   

   joostn

   on Jan 11, 2026

   Author

   More actions

   I tested it for the 5 clang toolchains I have at hand:

   #if defined(__APPLE__)
     #if defined(__aarch64__)
       static_assert(sizeof(long double) == 8);
     #elif defined(__x86_64__)
       static_assert(sizeof(long double) == 16);
     #else
       #error "Unsupported architecture"
     #endif
   #elif (defined(_WIN32) && defined(_MSC_VER) && defined __clang__)
     #if defined(__aarch64__)
       static_assert(sizeof(long double) == 8);
     #elif defined(__x86_64__)
       static_assert(sizeof(long double) == 8);
     #else
       #error "Unsupported architecture"
     #endif
   #elif (defined(_WIN32) && defined(__MINGW32__) && defined __clang__)
     #if defined(__aarch64__)
       static_assert(sizeof(long double) == 8);
     #elif defined(__x86_64__)
       static_assert(sizeof(long double) == 16);
     #else
       #error "Unsupported architecture"
     #endif
   #elif (defined(__linux__) && defined __clang__)
     #if defined(__x86_64__)
       static_assert(sizeof(long double) == 16);
     #else
       #error "Unsupported architecture"
     #endif
   #else
     #error "untested"
   #endif

   I think it would be good to add static assertions to fp_traits.hpp, mismatches would be caught at compile time
5. jzmaddock

   added a commit that references this issue on Jan 13, 2026

   Update fptraits.hpp for gcc/windows/aarch64.

   ...

   This issue will close once commit b390d09 is merged into the 'develop' branch.

   b390d09
6. mborland

   closed this as completedin #1350on Jan 24, 2026
7. ckormanyos

   reopened this on Jan 24, 2026
8. ckormanyos commented on Jan 24, 2026

   

   ckormanyos

   on Jan 24, 2026

   Last edited by ckormanyos

   Member

   More actions

   I'd like to support embedded arm-none-eabi also, even if this is not an OFFICIALLY supported compiler, it's nice to have in Math.
9. ckormanyos commented on Jan 24, 2026

   

   ckormanyos

   on Jan 24, 2026

   Member

   More actions

   See also #1352
10. NAThompson commented on Sep 20, 2026

    

    NAThompson

    on Sep 20, 2026

    Collaborator

    More actions

    Resolved on develop
11. NAThompson

    closed this as completedon Sep 20, 2026