On Sun, Jun 26, 2011 at 5:34 PM, Reiner Pope <reiner.p...@gmail.com> wrote: > On a similar note, would it be acceptable to join the 'hlRun' and 'hlFocus' > fields. Currently they are: > > hlRun :: Scanner Point Char -> Point -> cache -> cache, > hlFocus :: Map WindowRef Region -> cache -> cache
I am not sure... However, maybe you would enjoy the following explanation on what "focusing" is all about -- I don't think I have documented it anywhere. Focusing on a particular point allows better asymptotic performance. The idea is that focusing is called when a window is "scrolled", so that the AST does not have to be re-traversed from the root every time a display is needed. Otherwise, the AST has to be traversed in pre-order to find the portion that corresponds to the windows, at every refresh. You may retort: "Why don't you cache the beginning/end of the regions corresponding to each node in the AST?" Otherwise, it would break the "online" properties of the parser. (See www.cse.chalmers.se/~bernardy/FunctionalIncrementalParsing.pdf for the explanation of the online parsing) > It seems that we could allow simpler parsers (more on this later) I dare say that a sensible overall plan about parsers should be well thought out before putting the current code under the knife :). I am curious to hear your plans! I could also let you know about mine if you are interested. > if we combined them into one: > > hlRun :: Scanner Point Char -> Point -> Map WindowRef Region -> cache -> > cache > > After my brief search through the code, it appears that these are run in the > main loop as follows > > (in Yi.Core.refreshEditor): > pureM clearAllSyntaxAndHideSelection >>= > ... (do layout stuff) > pureM focusAllSyntax >>= > ... (display everything) > > I don't see any reason why this couldn't be changed to > > ... (do layout stuff) > pureM runAndFocusAllSyntax >>= > ... (display everything I cannot recall why these two calls are separated. Perhaps I just wanted the focusing to be easily removable. That's the last non-trivial thing I implemented, and I wasn't sure the asymptotic performance gain was worth the trouble. > which would allow merging hlRun and hlFocus as described above. In fact, I > tested this by moving the calls to 'pureM clearAllSyntaxAndHideSelection' > and 'pureM focusAllSyntax' next to each other, and nothing appears to have > been broken. Did you also test asymptotic performance as hinted above? (Quite a pain to do, I'll grant you that much). > The reason for merging hlRun and hlFocus is as follows. For example, suppose > I have a simple parser which just recognises keywords based on > regexp-matching (I know it's bad, but it's just an example). Do you have any good example? > If hlRun and > hlFocus were combined, my cache could be > > type Cache = Map WindowRef [(Keyword, Point)] At the moment, the parsers are optimized for a single window showing a given buffer (that's another way to say that I ignored the case of a buffer shown in multiple windows). I'd much prefer a solution that handles multiple windows in a generic way rather than to hack it withing the cache of a specific parser. Maybe that's what you want to do though? Cheers, JP. -- Yi development mailing list yi-devel@googlegroups.com http://groups.google.com/group/yi-devel
