Julian Foad wrote: > Peter Samuelson wrote: > > [Julian Foad] > > > Proposal for switching to a dedicated "fs path" API: > > > > > > Step 1: Introduce a new API for FS paths, as a thin layer that maps to > > > existing path APIs. It must assert that the fspath arguments and fspath > > > return values begin with '/'. Initially it should map directly to > > > svn_uri_*, so that it can be seen to be a direct replacement.
I've just scanned through libsvn_fs*, libsvn_repos and libsvn_ra*, and the only svn_uri_* functions used here on non-URI paths appear to be: _join() _is_child() _basename() So that makes for a nice small start-up requirement for this new API. There are probably others used elsewhere in our codebase, of course. - Julian > > So do these or do these not involve doing URI escaping? %20 for space, > > etc. > > No escaping involved here - these are not URLs and these are not > relative-URLs and so do not have any escaping inside them. When we make > a URL from a base URL and a repo-relpath, *then* we will escape it. > > I believe, but haven't checked recently, that the present set of > svn_uri_* functions doesn't touch or care about percent-escaping when > handling these paths. Because we have forced them to handle this case, > they don't assert that their arguments are canonical. After we make > this transition, then the svn_uri_* functions should be able (from a > correctness point of view) to assert that their 'URI' inputs and outputs > are canonical. > > > > By calling them 'fs paths', the assumption is that these are in > > UTF-8, and there is no marshalling / escaping to even think about, > > except when converting it to or from a URI or dirent. Is that the > > intent? > > Yes. > > > > > Bikeshed: svn_fspath__* or svn_fs__path_* or something else? > > > > svn_fspath__ seems better to me. svn_fs__path_ sounds like this is all > > internal to libsvn_fs, whereas I understand you want this in > > libsvn_subr for layers to use that are _not_ specifically talking about > > svn_fs internals. > > Correct, so yes, I agree. > > - Julian > >
