Build issue with clang8 on amd64 #594

In file included from src/secp256k1.c:12:
In file included from ./src/scalar_impl.h:20:
./src/scalar_4x64_impl.h:379:19: error: value '4624529908474429119' out of
range for constraint 'n'
    : "S"(l), "n"(SECP256K1_N_C_0), "n"(SECP256K1_N_C_1)
		  ^~~~~~~~~~~~~~~
./src/scalar_4x64_impl.h:17:25: note: expanded from macro 'SECP256K1_N_C_0'
#define SECP256K1_N_C_0 (~SECP256K1_N_0 + 1)
			^~~~~~~~~~~~~~~~~~~~
./src/scalar_4x64_impl.h:458:74: error: value '4624529908474429119' out of
range for constraint 'n'
    : "g"(m0), "g"(m1), "g"(m2), "g"(m3), "g"(m4), "g"(m5), "g"(m6),
"n"(SECP256K1_N_C_0), "n"(SECP256K1_N_C_1)
									
^~~~~~~~~~~~~~~
./src/scalar_4x64_impl.h:17:25: note: expanded from macro 'SECP256K1_N_C_0'
#define SECP256K1_N_C_0 (~SECP256K1_N_0 + 1)
			^~~~~~~~~~~~~~~~~~~~
./src/scalar_4x64_impl.h:504:64: error: value '4624529908474429119' out of
range for constraint 'n'
    : "g"(p0), "g"(p1), "g"(p2), "g"(p3), "g"(p4), "D"(r),
"n"(SECP256K1_N_C_0), "n"(SECP256K1_N_C_1)
							       ^~~~~~~~~~~~~~~
./src/scalar_4x64_impl.h:17:25: note: expanded from macro 'SECP256K1_N_C_0'
#define SECP256K1_N_C_0 (~SECP256K1_N_0 + 1)
			^~~~~~~~~~~~~~~~~~~~

Is there anything special about how you're building it? You're not trying to build a 32bit binary?

This looks like clang is broken: SECP256K1_N_C_0 is a 63 bit number and should be fine for "n" on x86_64. SECP256K1_N_C_1 has a similar magnitude (it's larger, in fact) and your output is not complaining about that.

Nothing special, here you can see the full build log: http://package18.nyi.freebsd.org/data/headamd64PR236062-default/2019-03-01_06h42m53s/logs/errors/secp256k1-0.1.20190204.log

I guess it doesn't complain about SECP256K1_N_C_1 just because it breaks before on SECP256K1_N_C_0.

Maybe we should just get rid of this mechanism of passing constants into the asm code, and instead inline them directly? I can't even find any documentation anymore about this "n" constraint; I suspect it isn't very standard (to the extent that GCC's extended asm syntax can be considered standard in the first place).

I can reproduce this on debian sid with

$ clang-8 --version
clang version 8.0.0-+rc3-1~exp1 (tags/RELEASE_800/rc3)
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin

$ ./autogen.sh && ./configure CC=clang-8 && make
...
Build Options:
  with endomorphism   = no
  with ecmult precomp = yes
  with jni            = no
  with benchmarks     = yes
  with coverage       = no
  module ecdh         = no
  module recovery     = no

  asm                 = x86_64
  bignum              = no
  field               = 64bit
  scalar              = 64bit

  CC                  = clang-8
  CFLAGS              = -g -O2 -W -std=c89 -pedantic -Wall -Wextra -Wcast-align -Wnested-externs -Wshadow -Wstrict-prototypes -Wno-unused-function -Wno-long-long -Wno-overlength-strings -fvisibility=hidden -O3
  CPPFLAGS            = 
  LDFLAGS             = 

  CC       src/bench_verify.o
gcc -I. -g -O2 -Wall -Wextra -Wno-unused-function -c src/gen_context.c -o gen_context.o
gcc -g -O2 -Wall -Wextra -Wno-unused-function  gen_context.o -o gen_context
./gen_context
  CC       src/libsecp256k1_la-secp256k1.lo
In file included from src/secp256k1.c:12:
In file included from ./src/scalar_impl.h:20:
./src/scalar_4x64_impl.h:379:19: error: value '4624529908474429119' out of range for constraint 'n'
    : "S"(l), "n"(SECP256K1_N_C_0), "n"(SECP256K1_N_C_1)
                  ^~~~~~~~~~~~~~~
./src/scalar_4x64_impl.h:17:25: note: expanded from macro 'SECP256K1_N_C_0'
#define SECP256K1_N_C_0 (~SECP256K1_N_0 + 1)
                        ^~~~~~~~~~~~~~~~~~~~
./src/scalar_4x64_impl.h:458:74: error: value '4624529908474429119' out of range for constraint 'n'
    : "g"(m0), "g"(m1), "g"(m2), "g"(m3), "g"(m4), "g"(m5), "g"(m6), "n"(SECP256K1_N_C_0), "n"(SECP256K1_N_C_1)
                                                                         ^~~~~~~~~~~~~~~
./src/scalar_4x64_impl.h:17:25: note: expanded from macro 'SECP256K1_N_C_0'
#define SECP256K1_N_C_0 (~SECP256K1_N_0 + 1)
                        ^~~~~~~~~~~~~~~~~~~~
./src/scalar_4x64_impl.h:504:64: error: value '4624529908474429119' out of range for constraint 'n'
    : "g"(p0), "g"(p1), "g"(p2), "g"(p3), "g"(p4), "D"(r), "n"(SECP256K1_N_C_0), "n"(SECP256K1_N_C_1)
                                                               ^~~~~~~~~~~~~~~
./src/scalar_4x64_impl.h:17:25: note: expanded from macro 'SECP256K1_N_C_0'
#define SECP256K1_N_C_0 (~SECP256K1_N_0 + 1)
                        ^~~~~~~~~~~~~~~~~~~~
3 errors generated.
make: *** [Makefile:1047: src/libsecp256k1_la-secp256k1.lo] Error 1

I can't even find any documentation anymore about this "n" constraint; I suspect it isn't very standard (to the extent that GCC's extended asm syntax can be considered standard in the first place).

see https://gcc.gnu.org/onlinedocs/gcc/Simple-Constraints.html#Simple-Constraints

@real-or-random Right, that's it. It's not listed in the machine specific ones for x86 though. @alexdupre @MarcoFalke Could you try replacing all instances of "n" with "i" in src/scalar_4x64_impl.h? (both work for me in GCC 8.2).

Even if using i fixes it, we might want to open a bug for clang.

https://bugs.llvm.org/show_bug.cgi?id=40890

I confirm that replacing "n" with "i" fixes the compilation and ./tests run successfully

Looks like this issue will be fixed upstream before the official release of clang 8, so it's probably not worth working around here.

I agree.

Compiles in clang-8-rc4