#2733 False "overflow in implicit constant conversion" warning

[Philipp Klaus Krause]

As of [r10267] (and probably for quite a while before that),

unsigned char c = 1  <<  7;

results in a warning "overflow in implicit constant conversion" . However, the constant is 128, which is perfectly within the range of unsigned int. I noticed this issue due to a failure of the valdiag regression testing.

Philipp

[Philipp Klaus Krause]

Looks like the 1 << 7 at some point is considered to be a signed char literal.

Philipp

[Philipp Klaus Krause]

The same problem happens for types other than char:

unsigned int i = 1u << 15;
unsigned long int l = 1ul  << 31;
unsigned long int ll = 1ull  << 31;

All give the same warning.

Philipp

[Philipp Klaus Krause]

However, for unsigned char, the false warning is always given, both on initialization and assignment. For the other unsigned types the false warning is only given on initialization.

Philipp

[Philipp Klaus Krause]

This, or a similar issue also affects operations other than <<:
https://www.mikrocontroller.net/topic/474289

[Erik Petrich]

I trying to resolve the similar [#2877] just before the last release, but couldn't find a correction simple enough that I had confidence that I wasn't inadvertently breaking something else. Because of the assignment to unsigned char, decorateType() is propagating RESULT_TYPE_CHAR which ends up causing the literal to convert from int to signed char.

[Diego Herranz]

As of r12014, I think this still happens.

I attach an example where 2 very similar expressions raise the warning or not, in case that helps.

Simply compiled using: sdcc main.c

unsigned char a = 0x80 | 0x00 | (0x01 << 2); //warning 158
unsigned char b = 0x80 | (0x01 << 2); //no warning

Many thanks!