mullen@sdsu.UUCP (deborah mullen) (10/07/88)
Unless I lost a character when downloading (never happened before), line 204 of console.c (1.3c update) is missing a closing comment. I assume this would cause the next line of code to be skipped. The PUBLIC and PRIVATE directives do not mean much. There are many routines declared PRIVATE that are called from other source files. Examples include flush, set_leds, tty_init, ..... I know the MINIX compiler does not care, but for the sake of documentation. How much was this version tested after all the new stuff was added? I'm not that keen on all the ega stuff. It makes the code large and messy. I believe the author was expressing some doubts in one of his postings also. -- Deborah Mullen
ast@cs.vu.nl (Andy Tanenbaum) (10/10/88)
In article <3193@sdsu.UUCP> mullen@sdsu.UUCP (deborah mullen) writes: >line 204 of console.c (1.3c update) is missing a closing comment. Yes, you are definitely right. >The PUBLIC and PRIVATE directives do not mean much. Right again. >How much was this version tested after all the new stuff was added? >I'm not that keen on all the ega stuff. It makes the code large and >messy. I believe the author was expressing some doubts in one of his >postings also. It has been moderately well tested. How the missing comment symbol slipped by is a mystery. I guess it only affects large transfers. I completely agree about the EGA stuff. I don't have an EGA board, so I didn't do that stuff myself. The author (Charles Hedrick) and I have been negotiating about that stuff. What I would like is to have two scrolling methods available, toggled by the F3 key: 1) plain vanilla 6845 hardware scrolling, like in 1.1 2) video RAM crawl software scrolling. The idea here is when the system comes up, the lowest 25 x 80 x 2 bytes are used to hold the screen image. When a new line is added, it goes at video RAM address 25 x 80 x 2 and the base register is shifted up by 160 (80 characters + 80 attribute bytes) Each new line uses a higher piece of video RAM until some limit is hit. I think all known video cards have >= 16K except the standard mono card, for which 6845 hardware scrolling works. When the video image has crawled up to the limit, the whole thing is copied back to the beginning of the video RAM and we start all over. With 16K RAM and a 4K image, the copying operation will be needed about every 77 lines. Assuming it can be done with interrupts enabled (for the RS232 line), the performance should be pretty good and the code pretty simple. In return for his doing this, I am willing to tolerate linewrapping (as an #ifdef), very much against my taste and better judgment. If anyone has other suggestions for simplifying the video business, please post them so we can have a discussion. These changes will apply to 1.4, not 1.3. Below is a shar file containing the fixes I made in response to Deborah Mullen's comments, as well as a crc * listing of the current kernel dir. These fixes apply to 1.3. I had to unseal the box anyway to fix the bootblok.s problem discussed in an earlier posting, so I did this too. Andy Tanenbaum (ast@cs.vu.nl) : This is a shar archive. Extract with sh, not csh. : This archive ends with exit, so do not worry about trailing junk. : --------------------------- cut here -------------------------- PATH=/bin:/usr/bin:/usr/ucb echo Extracting 'console.cdif' sed 's/^X//' > 'console.cdif' << '+ END-OF-FILE ''console.cdif' X*** console.old Mon Oct 10 11:59:02 1988 X--- console.c Mon Oct 10 11:56:50 1988 X*************** X*** 1,4 **** X- X /* Code and data for the IBM console driver. */ X X #include "../h/const.h" X--- 1,3 ---- X*************** X*** 201,207 **** X flush(tp); /* clear out the pending characters */ X X /* Update terminal data structure. */ X! count = offset - offset1; /* # characters printed * X tp->tty_phys += count; /* advance physical data pointer */ X tp->tty_cum += count; /* number of characters printed */ X X--- 200,206 ---- X flush(tp); /* clear out the pending characters */ X X /* Update terminal data structure. */ X! count = offset - offset1; /* # characters printed */ X tp->tty_phys += count; /* advance physical data pointer */ X tp->tty_cum += count; /* number of characters printed */ X X*************** X*** 213,219 **** X /*===========================================================================* X * out_char * X *===========================================================================*/ X! PRIVATE out_char(tp, c) X register struct tty_struct *tp; /* pointer to tty struct */ X char c; /* character to be output */ X { X--- 212,218 ---- X /*===========================================================================* X * out_char * X *===========================================================================*/ X! PUBLIC out_char(tp, c) X register struct tty_struct *tp; /* pointer to tty struct */ X char c; /* character to be output */ X { X*************** X*** 449,455 **** X /*===========================================================================* X * flush * X *===========================================================================*/ X! PRIVATE flush(tp) X register struct tty_struct *tp; /* pointer to tty struct */ X { X /* Have the characters in 'ramqueue' transferred to the screen. */ X--- 448,454 ---- X /*===========================================================================* X * flush * X *===========================================================================*/ X! PUBLIC flush(tp) X register struct tty_struct *tp; /* pointer to tty struct */ X { X /* Have the characters in 'ramqueue' transferred to the screen. */ X*************** X*** 1024,1030 **** X /*===========================================================================* X * set_leds * X *===========================================================================*/ X! PRIVATE set_leds() X { X /* Set the LEDs on the caps lock and num lock keys */ X X--- 1023,1029 ---- X /*===========================================================================* X * set_leds * X *===========================================================================*/ X! PUBLIC set_leds() X { X /* Set the LEDs on the caps lock and num lock keys */ X X*************** X*** 1051,1057 **** X /*===========================================================================* X * tty_init * X *===========================================================================*/ X! PRIVATE tty_init() X { X /* Initialize the tty tables. */ X X--- 1050,1056 ---- X /*===========================================================================* X * tty_init * X *===========================================================================*/ X! PUBLIC tty_init() X { X /* Initialize the tty tables. */ X X*************** X*** 1194,1200 **** X /*===========================================================================* X * func_key * X *===========================================================================*/ X! PRIVATE func_key(ch) X char ch; /* scan code for a function key */ X { X /* This procedure traps function keys for debugging purposes. When MINIX is X--- 1193,1199 ---- X /*===========================================================================* X * func_key * X *===========================================================================*/ X! PUBLIC func_key(ch) X char ch; /* scan code for a function key */ X { X /* This procedure traps function keys for debugging purposes. When MINIX is + END-OF-FILE console.cdif chmod 'u=rw,g=r,o=r' 'console.cdif' set `wc -c 'console.cdif'` count=$1 case $count in 4158) :;; *) echo 'Bad character count in ''console.cdif' >&2 echo 'Count should be 4158' >&2 esac echo Extracting 'rs232.cdif' sed 's/^X//' > 'rs232.cdif' << '+ END-OF-FILE ''rs232.cdif' X*** rs232.old Mon Oct 10 11:59:17 1988 X--- rs232.c Mon Oct 10 12:00:20 1988 X*************** X*** 405,411 **** X /*===========================================================================* X * tty_o_done * X *===========================================================================*/ X! PRIVATE int tty_o_done() X { X /* A write request on an RS232 line has completed. Send FS a message. */ X X--- 405,411 ---- X /*===========================================================================* X * tty_o_done * X *===========================================================================*/ X! PUBLIC int tty_o_done() X { X /* A write request on an RS232 line has completed. Send FS a message. */ X X*************** X*** 432,438 **** X /*===========================================================================* X * rs_sig * X *===========================================================================*/ X! PRIVATE rs_sig(tp) X struct tty_struct *tp; X { X /* Called when a DEL character is typed. It resets the output. */ X--- 432,438 ---- X /*===========================================================================* X * rs_sig * X *===========================================================================*/ X! PUBLIC rs_sig(tp) X struct tty_struct *tp; X { X /* Called when a DEL character is typed. It resets the output. */ + END-OF-FILE rs232.cdif chmod 'u=rw,g=r,o=r' 'rs232.cdif' set `wc -c 'rs232.cdif'` count=$1 case $count in 1396) :;; *) echo 'Bad character count in ''rs232.cdif' >&2 echo 'Count should be 1396' >&2 esac echo Extracting 'log' sed 's/^X//' > 'log' << '+ END-OF-FILE ''log' X33568 3218 at_makefile X64098 16075 at_wini.c X61767 9461 clock.c X29239 40519 console.c X46559 2880 const.h X09961 3543 dmp.c X37030 26551 floppy.c X57617 941 glo.h X24617 27165 klib88.s X01335 11156 main.c X36060 5296 memory.c X09331 13834 mpx88.s X65108 3375 pc_makefile X56408 9952 printer.c X10507 13753 proc.c X21644 2324 proc.h X12176 18813 ps_wini.c X20815 18995 rs232.c X61988 19960 system.c X38305 3634 table.c X61842 28048 tty.c X61705 7512 tty.h X44806 4708 ttymaps.h X27554 742 type.h X14198 26433 xt_wini.c + END-OF-FILE log chmod 'u=rw,g=r,o=r' 'log' set `wc -c 'log'` count=$1 case $count in 537) :;; *) echo 'Bad character count in ''log' >&2 echo 'Count should be 537' >&2 esac exit 0
dtynan@sultra.UUCP (Der Tynan) (10/11/88)
In article <1499@ast.cs.vu.nl>, ast@cs.vu.nl (Andy Tanenbaum) writes: > > In return for his doing this, I am willing to tolerate linewrapping (as an > #ifdef), very much against my taste and better judgment. > > If anyone has other suggestions for simplifying the video business, please > post them so we can have a discussion. These changes will apply to 1.4, not > 1.3. > > Andy Tanenbaum (ast@cs.vu.nl) Sorry to disagree with you, but personally, I favor line-wrap. Unfortunately, this is *not* a perfect world, and there are many programs which disappear off the side of the screen. I think it is the responsibility of the tty driver to cure this. For example, if you use term, and call a remote system, which answers at a different speed, it takes a few lines of garbage before you get any real data. Unfortunately, at this stage, everything has disappeared out of view. I can cite numerous examples of programs that behave this way, and unfortunately, it would take a lot to fix them all. On the other point, I think the console driver should be split. There should be one for EGA, and one for the rest of us. I know this is difficult to do, because the distribution disks must work with both, so maybe a hybrid is in order. In other words, a quick-and-dirty tty driver, which works with both types. Once the system is running, the end-user can select the appropriate code, and re-compile the kernel. Somewhat similar to the xx_wini.c deal. - Der -- Reply: dtynan@sultra.UUCP (Der Tynan @ Tynan Computers) {mips,pyramid}!sultra!dtynan Cast a cold eye on life, on death. Horseman, pass by... [WBY]
hedrick@athos.rutgers.edu (Charles Hedrick) (10/12/88)
In article <2571@sultra.UUCP>, dtynan@sultra.UUCP (Der Tynan) writes: > On the other point, I think the console driver should be split. There should > be one for EGA, and one for the rest of us. I know this is difficult to do, It's no longer worth doing in my opinion. I'm now using a kernel with the odd EGA register stuff removed. There are a few conditionals here and there, e.g. not waiting for retrace to change the screen, but that's also appropriate for the old monochrome adapter. The only remnant of the old EGA code is one small piece of code in the scroll routine. But it's much smaller than my first attempt. It is now modified so that it works just like the original Minix hardware scroll, except that at the end of the screen memory, the whole screen image is copied back to the start of the screen memory. Since it only has to be done once every N lines, the overhead is not noticable. The EGA-specific code for processing insert-delete line and various other places is gone. At this point there's not enough code to be worth worrying about. (I've sent diffs to ast, but he wants me to put in some more recent fixes and do the diffs again. I don't see that it will matter much, since patch shouldn't care about small differences in line numbers for context diffs, but I'll do that over the weekend and post the results.
srw%pws1@Sun.COM (Scott Wiesner) (10/13/88)
From article <Oct.12.01.06.28.1988.3257@athos.rutgers.edu>, by hedrick@athos.rutgers.edu (Charles Hedrick): > modified so that it works just like the original Minix hardware > scroll, except that at the end of the screen memory, the whole screen > image is copied back to the start of the screen memory. Since it only > has to be done once every N lines, the overhead is not noticable. I know very little about minix, but I am familiar with PC's, and I'd like to warn everyone that some of the 'clone' monochrome display adapters don't have multiple pages of screen memory so the trick of moving the beginning of the visible portion of the screen doesn't work. One that comes to mind is the adapter that came with one of the Wyse 286 AT's. Scott
hannam@elec.uq.oz (Andrew Hannam) (10/24/88)
> On the other point, I think the console driver should be split. There should > be one for EGA, and one for the rest of us. I know this is difficult to do, > because the distribution disks must work with both, so maybe a hybrid is in > order. In other words, a quick-and-dirty tty driver, which works with both > types. Once the system is running, the end-user can select the appropriate > code, and re-compile the kernel. Somewhat similar to the xx_wini.c deal. > - Der I don't see why it should be split. The EGA driver should not be difficult to do. The problem with the existing code is that it hasn't been thought out properly (i.e. remove software scrolling and fix the problem properly). Conceptually the EGA should be upwardly compatable with the CGA (although it isn't when it comes to memory wrap around). I have already produced a new tty driver (which has been posted to the net) which handles this problem properly without having umpteen million lines of specialized code for handling the EGA. As for line-wrap, I'll state again - Please put it in but make it a run-time option (i.e. ioctl or ANSI sequence support) rather than compile time. Whether you have line-wrap or not is dependant on the users preferences not on system policy. Again, this is already in my posted driver. (notice I'm really pushing for people to try it ?) I wish AST would take a good look at my new tty driver as it contains significant improvments to algorithms, methods etc. Some of this code may (!!) be useful to him. I HATE SOFTWARE SCROLLING !!
hedrick@geneva.rutgers.edu (Charles Hedrick) (10/26/88)
> On the other point, I think the console driver should be split. There should > be one for EGA, and one for the rest of us. I know this is difficult to do, > because the distribution disks must work with both, so maybe a hybrid is in This might have been true of 1.3c, but not of the final version. The amount of EGA-specific code is now very small.
julian@acp.OZ (Julian Elischer) (10/27/88)
First, an appology about how late this is..
we had a big blockage here in Western AUS.
ast mentions in his articel of the 10th of Oct
the method of scrolling that has the screen creep
one line higher in memory for each scroll, and then does a bulk
copy when the 'creep limit' is reached, to move to the base of video
ram again.. An alternative that might be worth considering is to
just keep 2 screen-fulls of ram for the screen. Characters are written
to 2 locations for each write, one on screen, and one off. When the
'creep limit' has been reached, the screen is reset to the bottom of
the video ram, which should ALREADY contain identical information
to the top of the video ram, hence no bulk copy is needed.
If it is coded correctly, the extra copy might (no promises)
not prove too much of an extra load ( though the 8086 architecture
may defeat you). Theoretically, this incremental copy MUST
have more overhead, however the elimination of the BULK copies
might be worth the trouble in some situations. ( and anyhow you are
doing the copy anyhow, just at a different time)
I unfortunatly do not have minix OR a machine to try it on
however I would be curious to hear if anybody can think
of any advantages or disadvantages to this scheme.
Julian Elischer
julian@acp.oz
UUCP: {check your map,uunet}!munnari!acp.oz!julian