> Compilers delete null pointer checks by default (on typical platforms) because it's what most users want.

It's what users think they want (it leads to e.g. higher numbers on meaningless microbenchmarks).

But the null pointer check was left in. (And, sure enough, adding -fno-delete-null-pointer-checks makes no difference.)

That's because y is never used, so why should the pointer be dereferenced? If you call bar(y) instead of just bar(), "x86-64 clang 3.9.1" with -O3 does the load as well (but after the check)

    foo(int*):                               # @foo(int*)
        test    rdi, rdi
        je      .LBB0_1
        mov     edi, dword ptr [rdi]
        jmp     bar(int)                 # TAILCALL
    .LBB0_1:
        ret

Only GCC does the kind of aggressive optimization the article mentions (and might need to be tamed by -fno-delete-null-pointer-checks).

Ha... yes, a good point. That's a reasonable reason not to generate the load, and stupid me for not noticing. What's also dumb is that my eyes just glossed over the ud2 instruction that both put in main too. The program (not unreasonably) won't even run properly anyway.

gcc does seem to be keener than clang to chop bits out - I think I prefer clang here. But let's see how I feel if I encounter this in anger in a non-toy example ;) I must say I'm still a bit suspicious, but I can't really argue that this behaviour is especially surprising here, or difficult to explain.