Fix autoconf check for __builtin_expect #571

pull real-or-random wants to merge 1 commits into bitcoin-core:master from real-or-random:ac-link changing 1 files +1 −1

real-or-random commented at 2:05 AM on October 31, 2018: contributor

Calling the compiler is not enough to check whether a function exists because the compiler will assume an implicit declaration if the function does not exist. We need the linker to check if the function really exists.

Fix autoconf check for __builtin_expect

Calling the compiler is not enough to check whether a function exists
because the compiler will assume an implicit declaration if the
function does not exist. We need the linker to check if the function
really exists.

e5a3d2135f

sipa commented at 2:10 AM on November 6, 2018: contributor

@real-or-random Eh, __builtin_expect is expected to be a builtin for which no runtime symbol exists (you can't write &__builtin_expect for example), so if it compiles, we should be fine.

EDIT: Ah, this is about the case where it's not fine; good point.

ACK
gmaxwell commented at 7:10 PM on December 12, 2018: contributor

Or we could stop using autoconf for this and switch to compiler version sniffing, gnupreq 3 is sufficient for expect.
real-or-random commented at 8:18 PM on December 12, 2018: contributor

I don't know what gnupreq is but clang has __builtin_expect, too.
sipa commented at 1:43 AM on December 16, 2018: contributor

@real-or-random The code here: https://github.com/sipa/minisketch/blob/master/src/util.h#L10L23 works on clang too, I think.
real-or-random commented at 10:50 AM on December 17, 2018: contributor

Closed, #579 is better
real-or-random closed this on Dec 17, 2018

Contributors

real-or-random

sipa

gmaxwell

Linked (view graph)

#579 Use __GNUC_PREREQ for detecting __builtin_expect