Is this a C51 bug?

What is the sign-ness of char for your build?

Assuming a char is signed then:

You have an element with a width of 1 bit as a signed char.
Does a 1 bit twos complement value make any sense?

When you return the value it is being converted to an unsigned char.
What conversion will occur?

Likely that you intended your 1 bit variable to be unsigned.
Try it.

struct tag
{
    unsigned char a:7;
    unsigned char b:1;
};

unsigned char foo(void)
{
    struct tag test;

    test.b = 1;

    return test.b;
}

void main(void)
{
    unsigned char dat;

    if(foo() == 1)// ***foo return 0x80 not 0x01
    {
        dat = 2;
    }
    while(1);
}

17: void main(void) 18: { 19: unsigned char dat; 20: 21: if(foo() == 1)// ***foo return 0x80 not 0x01
C:0x000F 12001A LCALL foo(C:001A)
C:0x0012 BF0103 CJNE R7,#0x01,C:0018 22: { 23: dat = 2;
C:0x0015 750802 MOV 0x08,#0x02 24: }
C:0x0018 80FE SJMP C:0018 8: unsigned char foo(void) 9: { 10: struct tag test; 11: 12: test.b = 1; 13:
C:0x001A 430980 ORL 0x09,#P0(0x80) 14: return test.b;
C:0x001D E509 MOV A,0x09
C:0x001F 5480 ANL A,#P0(0x80)
C:0x0021 FF MOV R7,A 15: }

THE RESULT IS THE SAME!

Keil support should be interested in this. Contact them.

As written, the code worked for me: dat is assigned a value of 2 and its branch is executed.

however, if you change

char dat;

by giving dat a value, like

char dat=1;

the branch isn't executed.

I didn't take a look at the disassembly but I tend to think that something isn't right here.

... I tend to think that something isn't right here.

I thought the same as you. When I first tried it with the signed char, things were obviously wrong and I thought I understood. When I converted to unsigned char, it looked right. I then tried it later and it didn't work. Must have done something different, but don't know what. This was all determined by looking at the disassembly.