[PATCH] libc++: Add option to disable access to the global filesystem namespace

[Ed Schouten]

In http://reviews.llvm.org/D8194#137640, @jroelofs wrote:

> Seems reasonable to me. I think you ought to fix a few of the tests that this "breaks", and add a few *.fail.cpp ones to verify that the functions you expect to be removed have actually been removed. For example ./test/std/input.output/file.streams/c.files/cstdio.pass.cpp should have:
>
>   static_assert((std::is_same<decltype(std::fflush(fp)), int>::value), "");
>   #if defined(_LIBCPP_HAS_GLOBAL_FILESYSTEM_NAMESPACE)
>   static_assert((std::is_same<decltype(std::fopen("", "")), std::FILE*>::value), "");
>   static_assert((std::is_same<decltype(std::freopen("", "", fp)), std::FILE*>::value), "");
>   #endif
>   static_assert((std::is_same<decltype(std::setbuf(fp,cp)), void>::value), "");
>   
>
> And then I'd add a ./test/std/input.output/file.streams/c.files/no.global.filesystem.namespace/fopen.fail.cpp with:
>
>   // REQUIRES: CloudABI
>   
>   #include <cstdio>
>   
>   int main() {
>       // fopen is not allowed on systems without the global filesystem namespace
>       std::fopen("", "");
>   }


Yes, exactly! I was planning on doing something similar to what you've described. Good to see that I'm starting to get the hang of it.

So my current roadmap is to first get libc++ to build on CloudABI before I attempt to tackle the test suite. In my opinion it doesn't make sense to patch up the test suite if the tests cannot be run decently. Being able to build libc++ would be a prerequisite. Getting all the tests to pass followed by setting up a buildbot will be my next milestones.

Does that approach sound reasonable to you?

Thanks,
Ed


REPOSITORY
  rL LLVM

http://reviews.llvm.org/D8194

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/