On 29.12.2010 09:43, Branko Čibej wrote: > On 28.12.2010 23:13, Peter Samuelson wrote: >> [C. Michael Pilato] >>>> svn_fspath__is_canonical >>>> svn_fspath__dirname >>>> svn_fspath__basename >>>> svn_fspath__split >>>> svn_fspath__join >>>> svn_fspath__is_child >>>> svn_fspath__skip_ancestor >>>> svn_fspath__is_ancestor >>>> svn_fspath__get_longest_ancestor >>>> svn_error__locate >>>> >>>> svn_error__malfunction >>> While we've mandated that "__" must be used for semi-public >>> functions, we've never said that it can't be used for private ones. >> Kamesh is talking about public, not private, functions. I.e., ones >> where we actually do have an ABI promise to keep. >> >> I looked at svn_error__locate last week. It's really only useful in >> --enable-maintainer-mode, but the way it's implemented, it ends up >> being used by macros in the public (non-maintainer) ABI, so even if we >> eliminate those callers, we have to supply the function itself forever. >> I've addressed this as best we can, in r1053469. > Now I'm really mystified. When I added those location-tracing functions > and macros, they were only ever enabled with SVN_DEBUG turned on (i.e., > in maintainer-mode). Now I see that those #ifdef wrappers are gone, and > can't recall an explanation as to why that's a good thing. > > Can anyone remember what that change was good for?
Found it, r843793 and I agree with the change. So, effectively, svn_error__locate has been public since then, and r1053469 should be reverted anyway because it breaks the ABI. The name may or may not be misleading, depending on interpretation ... no-one should call svn_error__locate directly, the public API are effectively the macro wrappers. -- Brane
