Quoth Mark Walters on Nov 25 at 2:31 pm: > > Hi > > This series looks good to me (I have not reviewed the two bindings > patches). Patch 2 looks like it makes things much easier to follow than > the current code (if I understood the current pointer stuff it > constructs the top-level list by doing pointer stuff to remove all > messages which are replies from the complete message list). Indeed, the > diff is more complicated than the new code! > > On Sun, 25 Nov 2012, Austin Clements <amdragon@MIT.EDU> wrote: > > This series adds a library API for iterating over all messages in a > > thread in sorted order. This is easy for the library to provide and > > difficult to obtain from the current API. Plus, if you don't count > > the code added to the bindings, this series is actually a net > > decrease of 4 lines of code because of simplifications it enables. > > > > Do we want the API to do more? Currently it's very minimal, but I can > > imagine two ways it could be generalized. It could take an argument > > to indicate which message list to return, which could be all messages, > > matched messages, top-level messages, or maybe even unmatched messages > > (possibly all in terms of message flags). It could also take an > > argument indicating the desired sort order. Currently, the caller can > > use existing message flag APIs to distinguish matched and unmatched > > messages and there's a separate function for the top-level messages. > > However, if the API could do all of these things, it would subsume > > various other API functions, such as notmuch_thread_get_*_date. > > I don't know if this is the right API. For the matched message etc I > think using the existing message flag APIs is simple enough. I am not > sure about sort orders though: that looks like it would be much easier > for the caller to have the correct sort by I am not sure what users > would need it. For sort order, I would be inclined to simply construct the reverse list the first time a caller asks for it. Theoretically the caller could do this just as easily as the library, except that we don't expose the list routines. If I do add sort order, I would also want to add some control over which list is returned, since it would be asymmetric to be able to request all messages in either order, but top-level messages only in oldest-first. I think this would be pretty simple, and would give us a reasonably general-purpose and extensible API. (It would also solve the naming conundrum I mentioned below in my original email.) > Best wishes > > Mark > > > > > > > > Also, is this the right name for the new API? In particular, if we do > > later want to add a function that returns, say, the list of matched > > messages, we'll have a convention collision with > > notmuch_thread_get_matched_messages, which returns only a count.