Skip to content

limitations.md

Latest commit

 

History

History
63 lines (51 loc) · 2.76 KB

limitations.md

File metadata and controls

63 lines (51 loc) · 2.76 KB
Raw

Limitations

avr-libstdcpp contains the complete libstdc++ headers, but only a few sources of the compiled library. Everything that is implemented in the headers works, everything that requires the compiled library fails to link.

What works

Everything that is implemented in templates or inline functions, for example:

  • Containers: array, vector, deque, list, forward_list, map, set, unordered_map, unordered_set, queue, stack, bitset, span
  • Strings: string, string_view, to_chars, from_chars, to_string
  • Algorithms and ranges: <algorithm>, <numeric>, <ranges>
  • Utilities: optional, variant, any, expected, tuple, function, unique_ptr, bit, ratio, type_traits, concepts, compare
  • <chrono> durations, time points and calendar types
  • <cmath>, <complex>, <numbers>, <random> (except random_device)
  • <atomic> with the runtime
  • Coroutines

The integration tests show more examples.

What does not work

These features require the compiled library, so they fail to link:

  • I/O streams (<iostream>, <sstream>, <fstream>)
  • Locales (<locale>)
  • Formatting (<format>, <print>)
  • std::random_device
  • The time zone database of <chrono>
  • Threads (<thread>, <mutex>, <condition_variable>): there is no operating system
  • Filesystem (<filesystem>)

These are disabled on purpose:

  • Exceptions and RTTI are not supported. Compile with -fno-exceptions -fno-rtti. Errors that would throw an exception call abort(), see getting started.
  • The mathematical special functions of C++17 (std::beta(), std::legendre(), …) are disabled, since they are implemented in the TR1 headers, which are not included.
  • Wide characters (wchar_t, <cwchar>, <cwctype>) are not supported by avr-libc.
  • The <cfenv> and <ctgmath> headers, since avr-libc has no <fenv.h> and <tgmath.h>.
  • Some C99 functions of <cstdlib> and <cstdio> that avr-libc does not implement, such as std::lldiv() and std::vsscanf().

Things to keep in mind

  • int and size_t are only 16-bit wide, which limits the size of containers and the range of many algorithms. -mint8 is not supported.
  • double is only 32-bit wide by default, the same as float. Since avr-gcc 10 long double is 64-bit and is implemented in software by libgcc, which is much slower and larger.
  • Dynamic memory is very limited: an ATmega328P has only 2kB of RAM. Consider containers with a fixed capacity like std::array, or custom allocators.
  • avr-libc's printf() does not support floating-point numbers by default.
  • The __int24 and __uint24 types of avr-gcc are not recognized by <type_traits>.