[clang] [clang][HIP] Disallow non-constant format string with pointer argument (PR #238287)
Steffen Larsen via cfe-commits
cfe-commits at lists.llvm.org
Thu Nov 12 00:50:20 PST 2026
steffenlarsen wrote:
> This seems unnecessarily pessimistic. `llvm::emitAMDGPUPrintfCall()` was designed with the intention of simply printing a pointer when it's not clear that a given argument actually points to a string corresponding to a `%s` specifier. `printf` is a generally useful function for debugging, so it was deemed the users responsibility if they don't see the string argument they expected. The general message is, "don't do fancy stuff, there are limitations to what you will see, but we will be lenient in what we accept".
Do we make it clear that it is the case that "%s" only currently works if the format string can be determined at compile-time? They way it is currently implemented, it's very easy for a user to go from printing a string through "%s" in a way that works to a way that doesn't work. I agree that this solution is pessimistic, as we cannot determine if a "%s" is inside the format string, but AFAICT the only valid case this should break is when a user wants to print the address of a pointer ("%p" or any integer format), but for users running into this restriction it should be easy to work around it by casting the pointer to an appropriate integer type. As such, I would argue that we pay a small cost to prevent the user shooting themselves in the foot.
https://github.com/llvm/llvm-project/pull/238287
More information about the cfe-commits
mailing list