On 29 December 2016 at 00:33, Adam Borowski <kilobyte@xxxxxxxxxx> wrote: > On Wed, Dec 28, 2016 at 10:02:11PM +0000, Sami Kerola wrote: >> Reset terminal size by assessing maximum row and column. This is useful >> when actual geometry and kernel terminal driver are not in sync. >> >> Addresses: http://bugs.debian.org/835636 >> Based-on-work-by: Adam Borowski <kilobyte@xxxxxxxxxx> >> Signed-off-by: Sami Kerola <kerolasa@xxxxxx> > > Awesome, thanks for taking a look at it! > Alas, at least the delay length makes your version not work for me. Fair enough, updates are available in the git repository at: git://github.com/kerolasa/lelux-utiliteetit.git setterm https://github.com/kerolasa/lelux-utiliteetit/commit/9e0d5c6b0f8294ceabf38212f6ec14f148e5e642 >> --- a/term-utils/setterm.1 >> +++ b/term-utils/setterm.1 >> @@ -228,6 +228,10 @@ Turns keyboard repeat on or off. >> +\fB\-\-resize\fP >> +Reset terminal size by assessing maximum row and column. This is useful >> +when actual geometry and kernel terminal driver are not in sync. > > Perhaps a remark about serial consoles as the principal use? They don't > transfer ioctls, just a byte stream plus breaks. Added. Could you consider sending an update to manual that clarifies why setterm --resize works when stty(1) and resize(1) fail. Ref: http://docstore.mik.ua/orelly/unix/upt/ch42_05.htm >> diff --git a/term-utils/setterm.c b/term-utils/setterm.c >> index abcf8b291..691c3bc2f 100644 >> --- a/term-utils/setterm.c >> +++ b/term-utils/setterm.c >> + tv.tv_sec = 0; >> + tv.tv_usec = 100; >> + return select(1, &set, NULL, NULL, &tv); > > This delay is too short even on my amd64 in xfce4-terminal, much less on a > serial link. Even at 115200 baud, 100usec = 1/10000sec is enough to ferry > less than a byte and a half, one way, without giving any time for the > terminal to respond. As the query will be often preceded by a login screen > -- sometimes just a banner, sometimes a picture, sometimes a screenful of > ALL CAPS "authorized use only" legal blathering, even my original timeout of > 3.0 seconds is quite short. > > In a vast majority of cases the timeout won't be reached -- it's just a > safeguard against the terminal being disconnected or not implementing cursor > position queries. Fair enough. Timeout is set to be 10 seconds. >> + * \e7 Save current state (cursor coordinates, attributes, >> + * character sets pointed at by G0, G1). >> + * \e[r Set scrolling region; parameters are top and bottom row. >> + * \e[99999E Move cursor down 99999 rows. >> + * \e[99999C Move cursor right 99999 columns. >> + */ >> + static const char *setup = "\e7\e[r\e[99999E\e[99999C"; > > I'm afraid some old terminals might use shorts there -- 9999 or 32766 (32767 > might be a magic number) could be better. Finding max value from termcap would be nice, but I don't know how to do that. Meanwhole INT16_MAX - 1 is reasonable compromise. >> + /* \e[6n Report cursor position. */ >> + static const char *getpos = "\e[6n"; >> + /* \e8 Restore state most recently saved by \e7. */ >> + static const char *restore = "\e8"; > >> + if (write(STDIN_FILENO, setup, strlen(setup)) < 0 || >> + write(STDIN_FILENO, getpos, strlen(getpos)) < 0) { >> + warn(_("write failed")); > > What about blasting all three strings to the terminal together? > > The former two have no logical reason to be apart, writing the third one > unconditionally means the cursor won't go to an odd place if the read fails, > thus they can go in one write. Striongs are merged, and I changed write() to write_all() to be more robust. There is also some signal safety now, primarily EINTR is checked at when select(2) and read(2) return. -- Sami Kerola http://www.iki.fi/kerolasa/ -- To unsubscribe from this list: send the line "unsubscribe util-linux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
