Hi Andrew, Andrew Cagney a écrit : > > Hello, > > Up until now I've been of the opinion that the TUI should be left alone > until after 5.1. Given that I've dragged my feet on 5.1, I'm thinking > that the question should be re-considered. > > The person I'm really interested in hearing an opinion from is Stephane > since he played with the TUI a little and got it to build. Stephane, do > you think that the TUI is salvageable or that a re-implementation would > be better. > The point I reached is not perfect but I can build and use the gdb-tui for i386 and m6811. - I removed all readline/termcap hacks; - I cleaned up various hacks in gdb and started to use several hooks used by insight; - I introduced a specific readline keymap for tui-mode and the ability to switch between the normal keymap (command line) and tui (a kind of short-cut, or, single key command) There are several issues: - There are few TUI hacks inside GDB that I've not yet solved. For some of them, some kind of callback like `create_breakpoint_hook' would be nice to have. So, I guess they are fixable in a clean way. - For a robust curses management, it is necessary to know (within tui) when gdb switches to the terminal inferior or to the gdb terminal. Typically, what would be necessary is to catch some of the target `to_terminal_*' (without overriding them). - Readline is not clean in its management of output. The way HP solved this is really terminal dependent, by defining a locked scroll region on the terminal. This does not work very well and readline output interact badly with curses (which is not aware of that). I solve this by in a Unix way. I created a pipe that is read by tui code and that serves as output for readline. It is clean as far as gdb/readline are concerned. But this is really Unix oriented. I think it's a little bit too early (for me) to declare TUI obsolete. I suggest I spend some more time on it to see if I can do something useful and acceptable for it (let's say before mid or end July). When do you plan to do a 5.1 release? Could you state a date when no new feature should be integrated? > One thing to keep in mind is that we're (I'm?) trying to reduce the > number of interfaces into GDB core with ui-out/ui-file and ``libgdb'' > becomming the only interfaces. For the TUI to survive it would need a > major rewrite anyway. > > Andrew Yes. Clean GDB API is important. Stephane ----------------------------------------------------------------------- Home Office E-mail: stcarrez@worldnet.fr Stephane.Carrez@sun.com WWW: http://home.worldnet.fr/stcarrez http://www.sun.com Mail: 17, rue Foucher Lepelletier 6, avenue Gustave Eiffel 92130 Issy Les Moulineaux 78182 Saint Quentin en Yvelines France