a simple compiling error with clang in macOS

[Teaonly]

Building error in macOS (version 10.12.3)
Code commit version:

commit 0a893abc7be4dbaf609c1db180dcc148853ad208
Author: Adam Lerer <alerer@fb.com>
Date:   Sat Feb 11 18:49:32 2017 -0800

    fix serialization bug for large files

Error message:

In file included from torch/csrc/serialization.cpp:7:
In file included from /Users/teaonly/Workspace/github/pytorch/torch/lib/tmp_install/include/TH/THGenerateAllTypes.h:8:
In file included from generic/serialization.cpp:1:
/Users/teaonly/Workspace/github/pytorch/torch/csrc/generic/serialization.cpp:106:27: error: no matching function for call to 'min'
      size_t to_convert = std::min(size - i, buffer_size);
                          ^~~~~~~~
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/../include/c++/v1/algorithm:2589:1: note: candidate template ignored: deduced conflicting types
      for parameter '_Tp' ('long long' vs. 'long')
min(const _Tp& __a, const _Tp& __b)
^
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/../include/c++/v1/algorithm:2599:1: note: candidate template ignored: could not match
      'initializer_list<type-parameter-0-0>' against 'long long'
min(initializer_list<_Tp> __t, _Compare __comp)
^
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/../include/c++/v1/algorithm:2581:1: note: candidate function template not viable: requires 3
      arguments, but 2 were provided
min(const _Tp& __a, const _Tp& __b, _Compare __comp)
^
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/../include/c++/v1/algorithm:2607:1: note: candidate function template not viable: requires single
      argument '__t', but 2 arguments were provided
min(initializer_list<_Tp> __t)

The clang will report error in the following code:

 54     for (int64_t i = 0; i < self->size; i += buffer_size) {
 55       size_t to_convert = std::min(self->size - i, buffer_size);

self->size is long
i is int64_t
buffer_size is long
so 'self->size-i' is a long long,

clang will report error when different type in std::min.

[szagoruyko]

you probably forgot to add export MACOSX_DEPLOYMENT_TARGET=10.9 ?

[Teaonly]

/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/../include/c++/v1/algorithm:2599:1: note:
      candidate template ignored: could not match 'initializer_list<type-parameter-0-0>' against 'long long'
min(initializer_list<_Tp> __t, _Compare __comp)
^
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/../include/c++/v1/algorithm:2581:1: note:
      candidate function template not viable: requires 3 arguments, but 2 were provided
min(const _Tp& __a, const _Tp& __b, _Compare __comp)
^
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/../include/c++/v1/algorithm:2607:1: note:
      candidate function template not viable: requires single argument '__t', but 2 arguments were provided
min(initializer_list<_Tp> __t)
^
14 errors generated.
/usr/bin/clang -Wno-unused-result -Wsign-compare -Wunreachable-code -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -I/Users/teaonly/opt/anaconda3/include -arch x86_64 -I/Users/teaonly/Workspace/github/pytorch -I/Users/teaonly/Workspace/github/pytorch/torch/csrc -I/Users/teaonly/Workspace/github/pytorch/torch/lib/tmp_install/include -I/Users/teaonly/Workspace/github/pytorch/torch/lib/tmp_install/include/TH -I/Users/teaonly/Workspace/github/pytorch/torch/lib/tmp_install/include/THPP -I/Users/teaonly/Workspace/github/pytorch/torch/lib/tmp_install/include/THNN -I/Users/teaonly/opt/anaconda3/lib/python3.5/site-packages/numpy/core/include -I/Users/teaonly/opt/anaconda3/include/python3.5m -c torch/csrc/autograd/init.cpp -o build/temp.macosx-10.6-x86_64-3.5/torch/csrc/autograd/init.o -D_THP_CORE -std=c++11 -Wno-write-strings -DWITH_NUMPY
error: command '/usr/bin/clang' failed with exit status 1
aliwork:pytorch teaonly$ echo $MACOSX_DEPLOYMENT_TARGET
10.9

I have setup env. If modify code like following, the pytorch will be installed.

--- a/torch/csrc/generic/serialization.cpp
+++ b/torch/csrc/generic/serialization.cpp
@@ -52,7 +52,7 @@ void THPStorage_(writeFileRaw)(THStorage *self, int fd)
     long buffer_size = std::min(self->size, (long)5000);
     std::unique_ptr<uint8_t[]> le_buffer(new uint8_t[buffer_size * sizeof(real)]);
     for (int64_t i = 0; i < self->size; i += buffer_size) {
-      size_t to_convert = std::min(self->size - i, buffer_size);
+      size_t to_convert = std::min(self->size - (int)i, buffer_size);
       if (sizeof(real) == 2) {
         THP_encodeInt16Buffer((uint8_t*)le_buffer.get(),
             (const int16_t*)data + i,
@@ -103,7 +103,7 @@ THStorage * THPStorage_(readFileRaw)(int fd)
     long buffer_size = std::min(size, (long)5000);
     std::unique_ptr<uint8_t[]> le_buffer(new uint8_t[buffer_size * sizeof(real)]);
     for (int64_t i = 0; i < size; i += buffer_size) {
-      size_t to_convert = std::min(size - i, buffer_size);
+      size_t to_convert = std::min(size - (int)i, buffer_size);
       SYSCHECK(read(fd, le_buffer.get(), sizeof(real) * to_convert));
       if (sizeof(real) == 2) {
         THP_decodeInt16Buffer((int16_t*)data + i,

[kashif]

hopefully #732 fixes this

[kashif]

@szagoruyko do you know where the MACOSX_DEPLOYMENT_TARGET variable is used? As far as I can see, it doesn't do anything...

[apaszke]

@kashif it's not used by PyTorch, but by the compiler

[apaszke]

This should be fixed now by @kashif's PR.