Hello all, Thanks to everyone following svnbrowse.
I believe this project has come to a state of a decently stable milestone. It is capable for general browsing around in a tree and overall interactivity feels good. However, there are a few things which need some further attention. 1. Opening a file would crash the entire application; I don't know what the best way to display them would be. 2. Handling of --revision and @pegrevision doesn't handle certain cases correctly. The condition that checks and fallbacks to HEAD via SVN_INVALID_REVNUM (passed to svn_ra_get_dir) misbehaves in certain cases (for example, 'svnbrowse ./path/in/wc -r head' does not work). 3. When launched from a colourless terminal (or with a limited colour support), the UI elements are displayed wrongly. This will occur primarily in legacy terminals. 4. No cancellation is yet implemented. Would be more like a fancy improvement and it doesn't really affect any user experience. 5. There is a funny segfault (to null dereference) when a node with no author is opened. 6. Implement non-pkgconfig search of a curses library (via find_package). 7. Include in autoconf/configure build. 8. Print path to stdout when exiting [2]. I would really appreciate help with 1 (I'm out of ideas about the user experience to deal with files), 2 (I have tried to find the correct way to go with. Somebody more experienced than me should step ahead here), and 7 (zero clue how autoconf works). The rest would be pretty obvious to tackle. Of course if you can run a dev build, I'm happy to hear some feedback. Also I start thinking about what the target use could be (besides wandering around in directories which is not that useful on its own). I believe it might sometimes come in handy for scripts to ask a user to pick a path. For example, we might target on doing something like this: [[[ $ svn checkout $(svnbrowse https://svn.apache.org/repos/asf) # the user picks a path and quits the app (enter does not work because it the # opens selected directory) # it is printed to stdout # $() puts it into an argument [1] and executes svn as usual ]]] As another application we currently have to type full branch URL when switching/merging (or use ^/ syntax which is still pretty much the same thing because one must remember its location in a repository). It could be either 'svn merge' and 'svn switch' themselves launch the browser (most likely they would do that in the same process) or it could be a simple shell alias. Not really another application, but a different way to describe that chunk of script from above. But... This on the other hand might be a bad idea; A tool starts doing multiple things at the same time. Instead of first deciding on the operation's scope and then executing the actions, it would do both at the same time. If let's say during a checkout, a user picks location and expects it to work, but then they realise that suddenly the internet connection just died so checkout failed (or name any other reason), all the effort required to pick that location will be lost. When it's a manually typed shell command, one might simply go back and retry/edit wrong part. [1] I did not test it but I'm pretty sure this is how shell works; it should be the thing as let's say /bin/basename. [2] it's a bit tricky to use stdout as right now this is the channel used for the interactive UI. There should be a some way to force curses to use stderr instead which solves the issues. -- Timofei Zhakov
