On Wed, May 28, 2014 at 4:15 PM, Daniel Shahaf <d...@daniel.shahaf.name>wrote:
> Stefan Fuhrmann wrote on Wed, May 28, 2014 at 12:30:36 +0200: > > On Wed, May 28, 2014 at 3:41 AM, Daniel Shahaf <d...@daniel.shahaf.name > >wrote: > > > > > Building latest trunk with gcc 4.7: > > > > > > subversion/libsvn_fs_fs/index.c:351:1: warning: always_inline function > > > might not be inlinable [-Wattributes] > > > subversion/libsvn_fs_x/index.c:356:1: warning: always_inline function > > > might not be inlinable [-Wattributes] > > > subversion/libsvn_fs_x/index.c:339:1: warning: always_inline function > > > might not be inlinable [-Wattributes] > > > subversion/libsvn_ra_svn/marshal.c:398:1: warning: always_inline > function > > > might not be inlinable [-Wattributes] > > > > > > If the inlining is really required, we should fix the code such that it > > > either inlines the function or fails to compile, rather than this > > > halfway mode. > > > > > > > None of these inlines is required for functional correctness. > > It is just that the compiler made a poor choice of those hot > > code paths. > > > > Could you please clarify this in the docstrings, as well? > > Right now they say: > > * The forced inline is required as e.g. GCC would inline read() into here > * instead of lining the simple buffer access into callers of get(). > > which could be mistaken as implying "required for functional > correctness". > > I'd do this myself but you'd probably be able to describe what they > _are_ required for better. > Done in r1598273. -- Stefan^2.
