[PATCH] D40319: [libcxx] Support getentropy as a source of randomness for std::random_device

[Jonathan Wakely via Phabricator via cfe-commits]

jwakely added subscribers: fweimer, dalias.
jwakely added inline comments.


================
Comment at: libcxx/trunk/src/random.cpp:29
+#if defined(_LIBCPP_USING_GETENTROPY)
+#include <sys/random.h>
+#elif defined(_LIBCPP_USING_DEV_RANDOM)
----------------
mcgrathr wrote:
> jwakely wrote:
> > jwakely wrote:
> > > musl only declares `getentropy` in `<unistd.h>` not `<sys/random.h>`. Glibc declares it in both. Should `<unistd.h>` also be included when `_LIBCPP_USING_GETENTROPY` is defined, so it can work more portably?
> > I suppose it doesn't really matter, since `_LIBCPP_USING_GETENTROPY` is currently only defined for fuchsia and wasi, which both declare `getentropy` in `<sys/random.h>`.
> FWIW, if there is an emerging norm to declare getentropy in <unistd.h> then Fuchsia's libc can add it to our <unistd.h> and there's no real problem relying on that in libc++ sources within about a week of when we land that change in Fuchsia's trunk.
> 
@fweimer added it to glibc's unistd.h for compatibility with existing BSD code:
https://sourceware.org/bugzilla/show_bug.cgi?id=17252#c9

Maybe @dalias can confirm whether musl did it for the same reason?

Does it make sense to "standardize" on declaring `getentropy` in `unistd.h`?


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D40319/new/

https://reviews.llvm.org/D40319