Eigen Compiler Suite

Thanks, I will look into that. Technically speaking division by zero is undefined in C and in intermediate code, so this behaviour is actually valid. The code should rather use the INFINITY macro from the math.h header to properly represent infinity.

Meanwhile I also tried to process with cdamd32 while I replaced inf by nan which gives me the following error:

./ecsd _PDCLIB_strtod_main.cod _PDCLIB_strtod_main.cod:154: error: invalid operand in 'mov f8 [$fp-48], f8 nan'

so this behaviour is actually valid

As far as I understand the ECS manual, both INF and NAN are valid values, so I would expect that a mov operation of these values would work.

division by zero is undefined in C

In ISO 9899 it indeed is undefined; but the compilers I usually use seem to behave all the same: x/0.0 is inf and 0.0/0.0 is nan. So I would assume that chibicc/ecc should also behave like this.

The code should rather use the INFINITY macro from the math.h

What I currently try to do is actually implementing a standard library for chibicc/ecc, and thus defining what these macros are is actually part of the task.

The attached patch should fix both issues, thanks for reporting.

Being undefined in C, C++, and intermediate code, your code should at least test against INFINITY and NAN and provide the expected outcome yourself rather than adding an implicit dependency on the compiler used to compile your code.

Yes, that's what I did in codegen.cpp (unfortunately not feasible in codegen.c), but by the end of the day INFINITY and NAN are still implementation dependent; but now everything compiles as expected (up to a few symbols I need from runtime.cpp), so thank you again very much for your support!