Daniel Shahaf wrote: > Julian Foad wrote on Wed, Sep 14, 2011 at 16:07:19 +0100: > > Enable libsvn_client APIs to re-use a previous RA session instead of always > > opening a new connection. > > > > Basically: > > > > - Declare an opaque "RA session cache" object in the public API. It holds > > one open RA session per repository, for any number of repositories. > > Why per repository?
Because the caller of a libsvn_client API generally provides a "path_or_url" without knowing which repository it refers to. If the client is calling e.g. "update a/", "update b/", "update c/", it doesn't know in advance whether those paths refer to different repositories. Suppose only "a" and "c" are in the same repo; the client can't pass just a single cached session if we want to re-use the "a" session for "c". > What about a client that runs 'svn update' on several > wc's of the same URL concurrently? I haven't attempted to support concurrent calls. The cache is for sequential use only (one user of the cache at a time). If a client is doing parallel updates and want to use this kind of cache it must have a separate cache per thread. I think. > > - Implement a private API for libsvn_client functions to use instead of > > simply opening a new session. It re-uses a session from the cache, if > > present, or opens a new connection if necessary. > > As far as I can see, the "get a session from the cache" code assumes > that the cache is used by a single thread. (Consider what happens if > one thread retrieves or uses a cached session while another retrieves > the same session from the cache.) Will the cache be thread-safe? See above. - Julian
