On Tue, Jan 29, 2013 at 10:40 AM, Bert Huijben <b...@qqmail.nl> wrote: > I think there might be third party implementations (CollabNet, WANdisco, > VisualSvn) that might rely on these values? > > So maybe somebody wants to respond? > > > Personally I would prefer to see an error somewhere in SVN_ERR_AUTHZ to at > least allow determining which subsystem produced the error code > programmatically. (I don't maintain code connected to this)
The existing implementation returned non SVN_ERR_AUTHZ_* values even though the documentation suggested otherwise. For example calling svn_repos_authz_read() with must_exist on a non-existant file will return ENOENT (or for that matter whatever error apr_file_open() produces). If we want SVN_ERR_AUTHZ specific error codes we can add them but changing what I changed today will only change the error codes for the in repo files. We'd have to catch a bunch of errors from svn_config_read2() to have SVN_ERR_AUTHZ_* specific error codes for the file paths. Looking now I don't think I should have used SVN_ERR_RA_LOCAL_REPOS_NOT_FOUND. Interestingly enough svn_ra_local__split_URL() doesn't even use it itself. It uses SVN_ERR_RA_LOCAL_REPOS_OPEN_FAILED for both the search for the root_path and for the actual open. I should have just created SVN_ERR_AUTHZ_REPOS_OPEN_FAILED. I'd be happy to add SVN_ERR_AUTHZ_REPOS_ENOENT and SVN_ERR_AUTHZ_REPOS_ENOTDIR (or something to that effect). But I'm not sure it's really worth it, since they would only apply to the in repository files and would create the impression that you're always going to get an SVN_ERR_AUTHZ_* error.
