On 3/17/23 02:45, Eric Blake wrote:
Where it can go wrong is when a public header has a macro that redefines the normal symbol name into the actual library linkage name (think about the public stat() vs. internal __stat64() mess when big files were first introduced, or more recently, figuring out how to support 64-bit time on 32-bit systems).
Ah, I got it now. The autoconf-generated test could theoretically fail (leading to a false negative) because "prctl" might not exist as a linkable symbol *at all* in the C library (so the fake function declaration in the generated test file would not link against anything). The actual function could be provided by the shared object under a symbol like "__prctl_foobar" *only*, and the only way to get from "prctl" to "__prctl_foobar" could be a #define available via inclusion of the <sys/prctl.h> header. Yeah, let's deal with *that* horror if it ever happens. :) Thanks! Laszlo