Test: fixing itoa implementation and clean-up of tests and test Makefile by mcspr · Pull Request #8531 · esp8266/Arduino (original) (raw)

Update itoa to be the same as newlib, also fixing edgecase of abs(INT_MIN)
Update WString.cpp:toString() integer conversions to use noniso funcs
Update WString to use C++ limits lib
Remove legacy gcc versions from Makefile and allow overrides
Don't fallback to c11 and c++11, source cannot support that

Additionally, allow test binary to accept arguments

~/d/e/t/host> make TEST_ARGS=[core][String] test
Makefile:31: using g++
Makefile:46: Cannot compile in 32 bit mode (g++-multilib is missing?), switching to native mode
Makefile:57: compiling in native mode
/home/runner/dev/esp8266-Arduino/tests/host/bin/host_tests [core][String]
===============================================================================
All tests passed (288 assertions in 20 test cases)

As mentioned by @d-a-v in the Matrix channel
Not really sure if it's the best idea feature-wise, but it keeps this weird Arduino legacy output throughout the class. Like, the way -100 with base 16 gets converted into ffffff9c instead of -64 (which what I'd expect here, with any other sane API :). Main reason for that is Arduino API does printf equivalent of %X. But, it is consistent with other implementations, and there's no branching required to the printf
ref. https://github.com/arduino/ArduinoCore-API/blob/master/test/src/itoa.cpp, https://github.com/espressif/arduino-esp32/blob/6cfe4613e4b4846e1ab08c7f78b7ea241f52c7da/cores/esp32/WString.cpp#L80-L89

Just using the noniso funcs saves a little bit of code. For example checking on the nm output for int ctor when building the test binary

• 0000000000000150 T String::String(int, unsigned char)

• 0000000000000121 T String::String(int, unsigned char)

I'd like to run these on the board before pushing though, since I have only tried the host test