This file documents what should be changed to this program in the future. It is mainly for "internal use only", so don't worry if it's completely rubbish. 1) Ideas and suggestions for enhancements ----------------------------------------- - More "man" files for _all_ tools would be nice... - when an error occurs in the middle of hardware programming, recovering from it would be nice instead of just bailing out, and leaving all registers in an undefined state (= a bad state) ... How to to this? Is it worth the trouble? Maybe use the "save_regs" function from the XFREE code before doing anything, and perform a "restore_regs" afterwards. This is only realistic when using more of the XFREE code. [ this is unlikely to be ever implemented ] - Of course, even MORE CHIPSETS! - Should really consider moving SVGATextMode closer to the XFREE3.x code, so developing new chipset-support is faster. Also use their code for some things (chipset detection, clock selection, register programming macro's...). check out xfree86/vga256/vga/vgaHW.c: contains register programming stuff This would allow me to use CLKREG_SAVE and RESTORE, so when something goes wrong, I can always restore registers. Much safer... [ this is unlikely to be ever implemented ] - A new program: "Xclockprobe": probes for the ENTIRE clocks line (like X): do one clockprobe for the 25 (or 28 MHZ) clock, warn if 25 and/or 28 are not there (=non-VGA), do other clocks like X (or not?) Could thsi be combined with grabmode? [ this is unlikely to be ever implemented ] - Try to parse text mode label, and see if the X x Y values match the real ones, and give warning if not! (useful for people who don't understand the non-relationship between the label and the actual text mode size.) [ this is unlikely to be ever implemented ] - some people (correctly) argue that the lex/yacc parser does too much in lex that could be done in yacc. Most of them agree this is not a big issue... This is not a high-priority issue. - Chips&Tech chipsets are still lacking. - translate output from "X -probeonly" to a piece of TextConfig file that can be appended to the default one. - make STM programs print out in which kernel they were compiled, next to the the version number - Must include Harald's "user_time" patches to compile grabmode for Windows. This would make for a highly useful tool... They are inclued now, but don't work yet. 2) Really wild or less useful ideas ------------------------------------- [ this is unlikely to be ever implemented ] - Measure the V-freq of the mode that was programmed in the chips, and compare it with the calculated one. If there's too much difference, do something nice (like restoring the old state)... (a "paranoia" switch ?) - check out mkmf, mkmk!! - Consider an X/Windows style "rgb.txt" and being able to say in TextConfig "Palette 0 navy blue" & "Palette 7 IvoryWhite". (K. Albanowski again ;-) - Extend the monitor range checking to allow the user to ask for the "best fitting (refresh, font size, ...) "X*Y" mode for his setup. Maybe call the the "-b" option ("best"). (suggested by Mr Albanowski) - Another obvious extension is full Monitor/Card blocks, like in XFree. 3.1. (suggested by the same one...) - A utility to test the monitor capabilities by sweeping through the freq ranges: first H, keeping V constant at 60 Hz (by decreasing H-size), then, using the H limits, test V! Dangerous... (?) - In theory, H-timings CAN be given with PIXEL precision instead of 8-pixel, using the "pixel panning" registers for both screen and cursor. - Allow the user to select color or monochrome mode from the config file. - grabmode/clockprobe: let them scan for config file, and if there is one, use device-dependent extensions. Of course, this breaks the entire "chipset-independance" idea behind the mode grabber. - try to create a better 32-high font, and maybe even some fonts between the 16 and 32 pixel boundaries. Let someone else do this :-) - sorting the default TextConfig per Hor refresh frequency instead of the hodgepodge it is now. - see Ferraro (I) pg 363: DES could be used to allow VGA more time to access font/char data. e.g. start using this when pixclock > (DEFAULT_MAXCLK - 10). - See if DOSEMU can be compiled with SVGALIB support, so it can restore chipset regs for more chipsets/clockchips. This way they don't have to do it themselves (again). - for V2.0: maybe use svgalib interfaces for hardware instead of custom routines. Could be compile-time switch? - try making this into MODULE that can be loaded VERY early on (for seeing error msg) - GGI support! 3) BUGS and other "features" that should be improved. ----------------------------------------------------- - VT_RESIZE problem: when a previous attempt via 1x1 failed, then the next user attempt at restoring the screen back from 1x1 to something a little more useful will not even TRY, since it believes the screen doesn't need to be resized (intelligent software ...). In fact, I need to re-think use of VT_RESIZE and the out-of-memory problem completely. It sucks (pond slime). Maybe we should check on the amount of free memory BEFORE doing the VT_RESIZE call (/proc/meminfo?) Maybe even wait until memory is there (with proper warning, as fron the NFS-server when it loses connection: "Not enough memory... waiting until we get some") using shrink_buffers() could help : do it with priorities going up on each attempt (start with 4, then 3...0). - LOTS of DOSEMU compatibility problems...: * ET4000W32p/ics5431: DOSEMU re-programs original values in clock regs! So when you switch back to linux console, your text mode clock is wrong. * Maybe even use ClockProg as in X to restore textmode clock when returning to text mode. This is especially useful for clockchip owners where clock is not restored in programmable clock regs) - For some reason, ClockPrg is VERY flaky in X! On some et4000's, the SEQ regs cannot be programmed from the ClockProg... - dependencies: if a file in XFREE/common_hw has changed, "make" doesn't re-make anything! Should we use a Makefile in that directory? - clear all chipset-specific H/V size extended regs (they are not used, and not even programmed in STM, so they could cause mayhem. They also reduce the chance for succes when trying to restore a broken textmode with STM). - grabmode: interlace detection: maybe check vga address each V retrace (how?), if +- equal, non-interlace. - mode 7 in svgalib's vgatest program still detects as 640x400 instead of 320x400 (on S3 805) - S3 + ti3026 still doesn't work for some people. Without proper feedback/help/testcard, this will never get fixed. - grabmode: - first attempt with 50 measurements, second with 100, 3rd with 200! - make it also remember the tv_sec field of the gettimeofday() call, so it can determine (and fold back!) wrapped measurements. - bug with debug mode: center detection is wrong (always last one time band taken). This looks like a compiler bug... (making some vars "volatile" fixes the problem). - detect multiple chipset/clockchip/... entries in config file, and flag them - include Stephen Lee's suggestions for improving the parser (esp. the quote-removal thingie). Here's what he said: > Well, let me atone myself by telling you how to do the quoted string > thing. Actually it is in flex.info. It has a longer one for doing C-like > escapes and etc. But I guess you won't need that. My shorter version is > below. > > Also, I'm quite puzzled by all the <X_????> you use in the lex file. I > don't think you need them. The only reason I give to their existence is > that you can have that > > . { yyerror("Unknown token"); } > > rule (instead of "parse error" from yacc), but that doesn't buy you much. > Parsing should be done in the .y file, and if somebody writes: > > Clockchip font > > Then so what? It would not matter whether you pass CLOCKCHIP FONT or > CLOCKCHIP 'f' to the .y file anyway, both would still be an error. And if > you quote all strings, it would work even if somebody writes > > Clockchip "font" > > Thus, I don't see any reason to have a X_CLOCKCHIP state. > > Back to the string thing. In your lex file do this: > > %{ > > int string_caller; > char string_buf[MAX_STR_CONST]; > char *string_buf_ptr; > > %} > > %x X_STR > > %% > > <whatever context that supports the use of a string> > \" { /* start of a string */ > string_caller = YY_START; > string_buf_ptr = string_buf; > BEGIN(X_STR); > } > > <X_STR>{ > \" { /* done */ > BEGIN(string_caller); > *string_buf_ptr++ = '\0'; > yylval.sval = strdup(string_buf); > return (STRING); > } > \n { /* unterminated string, error */ > yyerror("unterminated string"); > } > [^\n\"]+ { > char *yptr = yytext; > > while ( *yptr ) > *string_buf_ptr++ = *yptr++; > /* you might want to check the length as well */ > } > } /* <X_STR> */ > > %% - CRTC index 5 bits 5 and 6 should be "01" for low-res modes (clkdiv2?) to avoid a right shift. grabmode should also know about this: The left/right shifting of the display is correlated only to bits 5 and 6 of CRTC index 5. The description of these bits in vgadoc4b is quite cryptic. Normally, CRTC index 5 has both bits 5 and 6 set to 0. Setting them to 01 (bit 5=1, bit 6=0) shifts the screen 1 char left. Setting them to 10 shifts 2 chars left, Setting them to 11 shifts 3 chars left. My BIOS just sets them to 01 in text 40x25, but it works the same way in all modes I tested, both text and graphics. Strangely, grabmode seems to ignore these bits in all modes (grabbed timings do not depend on them) except in text 40x25, where it reports that strange 512 for hsync end if the bits are set to 01. ... It _depends_ on the shift value. I tried setting up 40x25x9 with STM and I got the usual right-shifted screen. grabmode reported it correctly: "40x25" 12.06 320 344 384 400 200 206 207 225 ... Then I set the "shift" value to 01. The screen was perfectly centered, but grabmode reported: "40x25" 12.06 320 344 512 400 200 206 207 225 ... just as it does with the 40x25 built by the BIOS. Is that a bug in grabmode, some strange interaction between "shift" value and timings, or what? Massimiliano Ghilardi - add minimal chipset detection. Esp. on PCI chips this is easy. NOT full auto-detect, just check if specified chipset is present. If not: refuse (unless "-v" is given) - IBM RGB ramdac: all clocks >80Mhz are set to 80Mhz.