On Fri, Feb 6, 2015 at 3:43 PM, Kyle J. McKay <mackyle@xxxxxxxxx> wrote: > On Feb 6, 2015, at 12:05, Junio C Hamano wrote: >> "Kyle J. McKay" <mackyle@xxxxxxxxx> writes: >>> So I think it needs to stay #define'd to nothing to be safe in case >>> anything later on ends up including stuff that uses it. >> >> Doesn't the fact that your test failed indicates that it is not jsut >> "to be safe in case" but is required for correctness? >> >> [...] I do not know what >> changes they make to openssl/*.h (which is included just after the >> above header is included, but I would imagine that is where the >> AVAILABLE_MAC_OS_X_VERSION_XXX_AND_LATER_BUT_DEPRECATED_IN_MAC_OS_VERSION_YYY >> macros are checked and annoying warnings that are being squelched by >> the previous change are given? > > Yes. > > Although Eric didn't specify exactly where when he suggested adding this: > > On Feb 6, 2015, at 02:00, Eric Sunshine wrote: >> >> #ifdef __APPLE__ >> #undef DEPRECATED_ATTRIBUTE >> #endif > > I took the suggestion to be after the openssl/*.h headers are included which > would avoid the error of having DEPRECATED_ATTRIBUTE be #undef'd for them. > But, even math.h can end up including AvailabilityMacros.h, so I think > #undef'ing DEPRECATED_ATTRIBUTE after the openssl/*.h headers are included > would be unsafe in general. While we might happen to get away with that > today, if say compat/apple-common-crypto.h changes in the future (or for > that matter any sequence of includes in other files or any headers in the > Apple SDK) we could start seeing the error. > > TLDR; yeah, DEPRECATED_ATTRIBUTE needs to remain defined to nothing. I agree with this analysis. An alternative would be to revert b195aa00 (and not apply this patch), but then we risk having legitimate Git-related compilation warnings lost in the noise of the useless Apple deprecation warnings. Given the glacial pace at which Apple headers changes, and considering the rapid pace of change in Git, it still seems the lesser evil to suppress the useless warnings Apple thrust upon us when they deprecated OpenSSL in its entirety without providing replacements. It's unfortunate that the DEPRECATED_ATTRIBUTE #define will bleed beyond the OpenSSL #includes and potentially allow us to miss some future Apple deprecations, however, given the shortcomings of b195aa00, the proposed patch seems to be the best we can do. -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html