On Tue, Mar 6, 2012 at 11:01 AM, Marcus Denker <[email protected]>wrote:
> > On Mar 6, 2012, at 7:30 PM, Max Leske wrote: > > > This may be a stupid question… (As far as I understand, condensing the > changes is not the same) > > Why don't you increment the version number of the sources file to 1.1 > and use a fresh changes file? > > > The sources file does not contain the sources and the changes just the > changes. (even though one could think so taking > the names into account.) > > The sources file containes the sources of ages ago (1.0), the .changes is > "on top of that". This means that the .sources > file contains a lot of code from classes and methods that have been > removed (like the changes), > as well as old versions. > > The .sources/.changes mechanism merges three responsabilties: > > 1) be a log of all edited code in case of a crash > 2) store the current source of the methods (when you ask a method for > code, it reads either from the .sources > (method not changed since 1.0) or from the .changes (method was changed). > 3) provide a history for all code > > And all that in 2 big files that one can only append to because? No idea, > actually... > > I personally think that it is a mistake today to merge all these into one > such basic mechanism. Especialy if > you look at how complex the code is in the image... it's actually amazing. > > When you do a #condenseSources (which is sadly broken right now), then it > generates a .sources > file with just the code in the current image. It would be interesting to > see how large that is. > IMO it's also interesting to see how big a condense of only the valid history is, i.e. removing history for missing classes and methods, and only retaining history that is referenced from current methods (which eliminates branches that don't lead to the current definition). After all this is the history one wants, and I already provided code to compute the per-method history that leads to the current definition. Why not have your cake and compress it too? > > Marcus > > -- > Marcus Denker -- http://marcusdenker.de > > > -- best, Eliot
