CQF compilation fails with gcc 6.2

[ctb]

% make
./setup.py build_ext --inplace
running build_ext
bash -c cd third-party/cqf && make
make[1]: Entering directory `/mnt/research/ged/ctb/khmer/third-party/cqf'
g++ -std=c++11 -fPIC -Wall   -Ofast  -m64 -I. -Wno-unused-result -Wno-strict-ali
asing -Wno-unused-function  gqf.c -c -o gqf.o
cc1plus: error: invalid option argument ‘-Ofast’
cc1plus: error: unrecognized command line option "-std=c++11"
cc1plus: warning: unrecognized command line option "-Wno-unused-result"
make[1]: *** [gqf.o] Error 1
make[1]: Leaving directory `/mnt/research/ged/ctb/khmer/third-party/cqf'
error: command 'bash' failed with exit status 2
make: *** [khmer/_khmer.cpython-35m-x86_64-linux-gnu.so] Error 1

% g++ -v
Using built-in specs.
COLLECT_GCC=g++
COLLECT_LTO_WRAPPER=/opt/software/GCC/6.2/libexec/gcc/x86_64-pc-linux-gnu/6.2.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: ../gcc-6.2.0/configure --disable-multilib --prefix=/opt/software/GCC/6.2
Thread model: posix
gcc version 6.2.0 (GCC)

I removed the offending lines and things seemed to work (compilation succeeded, at least).

% git diff
diff --git a/third-party/cqf/Makefile b/third-party/cqf/Makefile
index 2fc9ba2..3ca26a3 100644
--- a/third-party/cqf/Makefile
+++ b/third-party/cqf/Makefile
@@ -5,7 +5,7 @@ ifdef D
        OPT=
 else
        DEBUG=
-       OPT=-Ofast
+       OPT=-O3
 endif
 
 # using the SSE4.2 extensions doesn't seem to speed things up by a large
@@ -23,9 +23,9 @@ ifdef P
        PROFILE=-pg -no-pie # for bug in gprof.
 endif
 
-CXX = g++ -std=c++11 -fPIC
-CC = g++ -std=c++11 -fPIC
-LD= g++ -std=c++11
+CXX = g++ -fPIC
+CC = g++ -fPIC
+LD= g++
 
 CXXFLAGS = -Wall $(DEBUG) $(PROFILE) $(OPT) $(ARCH) -m64 -I. -Wno-unused-result

[ctb]

all tests passed. Note, this is on the MSU HPC:

module load GNU/6.2
module load Python/3.5.0

[betatim]

This seems super weird. Both command-line options should be valid. What is the version of g++ you get without setting up gcc 6.2? Wondering if because of bash -c cd third-party/cqf && make the shell in which make runs doesn't inherit the setup?

[betatim]

Probably adding something like @echo $(shell which $(CXX)) to the all target in the cqf Makefile so we can see the path to g++ will help.

[ctb]

...this appears to have been fixed invisibly on the HPC; perhaps there was a wonky install or something. At any rate my patchfix stopped working and when I did a clean checkout of the spacegraphcats branch, compilation succeeded and tests passed. so... Whatevs.

[ctb]

OK, ran into the problem again - it was "solved" before because I reverted the branch I was working on to something fairly old, and so the problem cropped up when I updated to the latest master!

Here's a fix for GCC 6.2 -

diff --git a/third-party/cqf/Makefile b/third-party/cqf/Makefile
index 2fc9ba2..fd3cfa3 100644
--- a/third-party/cqf/Makefile
+++ b/third-party/cqf/Makefile
@@ -5,7 +5,7 @@ ifdef D
        OPT=
 else
        DEBUG=
-       OPT=-Ofast
+       OPT=-O3
 endif
 
 # using the SSE4.2 extensions doesn't seem to speed things up by a large
@@ -23,9 +23,9 @@ ifdef P
        PROFILE=-pg -no-pie # for bug in gprof.
 endif
 
-CXX = g++ -std=c++11 -fPIC
-CC = g++ -std=c++11 -fPIC
-LD= g++ -std=c++11
+CXX = g++ -std=c++0x -fPIC
+CC = g++ -std=c++0x -fPIC
+LD= g++ -std=c++0x
 
 CXXFLAGS = -Wall $(DEBUG) $(PROFILE) $(OPT) $(ARCH) -m64 -I. -Wno-unused-result
 
diff --git a/third-party/cqf/gqf.c b/third-party/cqf/gqf.c
index 6ac1d5e..e7e4493 100644
--- a/third-party/cqf/gqf.c
+++ b/third-party/cqf/gqf.c
@@ -28,6 +28,12 @@
 #include <sys/stat.h>
 #include <unistd.h>
 #include <sys/mman.h>
+#include <cstdint>
+#include <limits>
+
+#ifndef UINT64_MAX
+#define UINT64_MAX std::numeric_limits<std::uint64_t>::max();
+#endif
 
 #include "gqf.h"