How does the `--defsym` linker flag work to pass values to source code?

Question

In an LJ post, the --defsym flag was used for passing the build date into the source code:

#include <stdio.h>

extern char __BUILD_DATE;

void main(void) {
    printf("Build date: %u\n", (unsigned long) &__BUILD_DATE);
}

by linking with the following flags:

gcc example.c -Xlinker --defsym -Xlinker __BUILD_DATE=$(date +%Y%m%d)

According to the ld manual,

--defsym symbol=expression Create a global symbol in the output file, containing the absolute address given by expression.

I am trying to understand the following:

1. How is the 9-character string for the build date (YYYYmmdd+\0) stored in memory?
2. If --defsym creates a symbol containing an address, why __BUILD_DATE is defined as a char and not as a pointer or as an integral type?
3. Why __BUILD_DATE is defined as char and not unsigned long if it is eventually casted to unsigned long?

Answer 1

Linkers see globals as addresses (pointers to the "actual" global rather than the actual global -- even if the actual global doesn't exist at that address). -Xlinker --defsym -Xlinker __BUILD_DATE=$(date +%Y%m%d) sets the address of __BUILD_DATE not the value. When the __BUILD_DATE linker entity has an address but not a value, you can get the address by declaring the entity as anything and then taking the address of that.

In:

#include <stdio.h>

//extern long __BUILD_DATE[128];
//extern int __BUILD_DATE;
extern char __BUILD_DATE;

int main(void) 
{
    printf("Build date: %lu\n", (unsigned long)&__BUILD_DATE);
}

Any of the three declarations should work. Just don't try to use the value of that (pseudo) global. That would be like dereferencing an invalid pointer.

That should answer 2 and 3. To answer 1, -Xlinker --defsym -Xlinker __BUILD_DATE=$(date +%Y%m%d) stores the number returned (stdout) by $(date %Y%m%d) as the address of __BUILD_DATE. It doesn't store a string.