diff options
Diffstat (limited to 'contrib/ncurses/ncurses/tty/lib_mvcur.c')
-rw-r--r-- | contrib/ncurses/ncurses/tty/lib_mvcur.c | 1242 |
1 files changed, 1242 insertions, 0 deletions
diff --git a/contrib/ncurses/ncurses/tty/lib_mvcur.c b/contrib/ncurses/ncurses/tty/lib_mvcur.c new file mode 100644 index 000000000000..9eca0269390e --- /dev/null +++ b/contrib/ncurses/ncurses/tty/lib_mvcur.c @@ -0,0 +1,1242 @@ +/**************************************************************************** + * Copyright (c) 1998 Free Software Foundation, Inc. * + * * + * Permission is hereby granted, free of charge, to any person obtaining a * + * copy of this software and associated documentation files (the * + * "Software"), to deal in the Software without restriction, including * + * without limitation the rights to use, copy, modify, merge, publish, * + * distribute, distribute with modifications, sublicense, and/or sell * + * copies of the Software, and to permit persons to whom the Software is * + * furnished to do so, subject to the following conditions: * + * * + * The above copyright notice and this permission notice shall be included * + * in all copies or substantial portions of the Software. * + * * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS * + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. * + * IN NO EVENT SHALL THE ABOVE COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, * + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR * + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR * + * THE USE OR OTHER DEALINGS IN THE SOFTWARE. * + * * + * Except as contained in this notice, the name(s) of the above copyright * + * holders shall not be used in advertising or otherwise to promote the * + * sale, use or other dealings in this Software without prior written * + * authorization. * + ****************************************************************************/ + +/**************************************************************************** + * Author: Zeyd M. Ben-Halim <zmbenhal@netcom.com> 1992,1995 * + * and: Eric S. Raymond <esr@snark.thyrsus.com> * + ****************************************************************************/ + + +/* +** lib_mvcur.c +** +** The routines for moving the physical cursor and scrolling: +** +** void _nc_mvcur_init(void) +** +** void _nc_mvcur_resume(void) +** +** int mvcur(int old_y, int old_x, int new_y, int new_x) +** +** void _nc_mvcur_wrap(void) +** +** Comparisons with older movement optimizers: +** SVr3 curses mvcur() can't use cursor_to_ll or auto_left_margin. +** 4.4BSD curses can't use cuu/cud/cuf/cub/hpa/vpa/tab/cbt for local +** motions. It doesn't use tactics based on auto_left_margin. Weirdly +** enough, it doesn't use its own hardware-scrolling routine to scroll up +** destination lines for out-of-bounds addresses! +** old ncurses optimizer: less accurate cost computations (in fact, +** it was broken and had to be commented out!). +** +** Compile with -DMAIN to build an interactive tester/timer for the movement +** optimizer. You can use it to investigate the optimizer's behavior. +** You can also use it for tuning the formulas used to determine whether +** or not full optimization is attempted. +** +** This code has a nasty tendency to find bugs in terminfo entries, because it +** exercises the non-cup movement capabilities heavily. If you think you've +** found a bug, try deleting subsets of the following capabilities (arranged +** in decreasing order of suspiciousness): it, tab, cbt, hpa, vpa, cuu, cud, +** cuf, cub, cuu1, cud1, cuf1, cub1. It may be that one or more are wrong. +** +** Note: you should expect this code to look like a resource hog in a profile. +** That's because it does a lot of I/O, through the tputs() calls. The I/O +** cost swamps the computation overhead (and as machines get faster, this +** will become even more true). Comments in the test exerciser at the end +** go into detail about tuning and how you can gauge the optimizer's +** effectiveness. +**/ + +/**************************************************************************** + * + * Constants and macros for optimizer tuning. + * + ****************************************************************************/ + +/* + * The average overhead of a full optimization computation in character + * transmission times. If it's too high, the algorithm will be a bit + * over-biased toward using cup rather than local motions; if it's too + * low, the algorithm may spend more time than is strictly optimal + * looking for non-cup motions. Profile the optimizer using the `t' + * command of the exerciser (see below), and round to the nearest integer. + * + * Yes, I (esr) thought about computing expected overhead dynamically, say + * by derivation from a running average of optimizer times. But the + * whole point of this optimization is to *decrease* the frequency of + * system calls. :-) + */ +#define COMPUTE_OVERHEAD 1 /* I use a 90MHz Pentium @ 9.6Kbps */ + +/* + * LONG_DIST is the distance we consider to be just as costly to move over as a + * cup sequence is to emit. In other words, it's the length of a cup sequence + * adjusted for average computation overhead. The magic number is the length + * of "\033[yy;xxH", the typical cup sequence these days. + */ +#define LONG_DIST (8 - COMPUTE_OVERHEAD) + +/* + * Tell whether a motion is optimizable by local motions. Needs to be cheap to + * compute. In general, all the fast moves go to either the right or left edge + * of the screen. So any motion to a location that is (a) further away than + * LONG_DIST and (b) further inward from the right or left edge than LONG_DIST, + * we'll consider nonlocal. + */ +#define NOT_LOCAL(fy, fx, ty, tx) ((tx > LONG_DIST) && (tx < screen_lines - 1 - LONG_DIST) && (abs(ty-fy) + abs(tx-fx) > LONG_DIST)) + +/**************************************************************************** + * + * External interfaces + * + ****************************************************************************/ + +/* + * For this code to work OK, the following components must live in the + * screen structure: + * + * int _char_padding; // cost of character put + * int _cr_cost; // cost of (carriage_return) + * int _cup_cost; // cost of (cursor_address) + * int _home_cost; // cost of (cursor_home) + * int _ll_cost; // cost of (cursor_to_ll) + *#if USE_HARD_TABS + * int _ht_cost; // cost of (tab) + * int _cbt_cost; // cost of (back_tab) + *#endif USE_HARD_TABS + * int _cub1_cost; // cost of (cursor_left) + * int _cuf1_cost; // cost of (cursor_right) + * int _cud1_cost; // cost of (cursor_down) + * int _cuu1_cost; // cost of (cursor_up) + * int _cub_cost; // cost of (parm_cursor_left) + * int _cuf_cost; // cost of (parm_cursor_right) + * int _cud_cost; // cost of (parm_cursor_down) + * int _cuu_cost; // cost of (parm_cursor_up) + * int _hpa_cost; // cost of (column_address) + * int _vpa_cost; // cost of (row_address) + * int _ech_cost; // cost of (erase_chars) + * int _rep_cost; // cost of (repeat_char) + * + * The USE_HARD_TABS switch controls whether it is reliable to use tab/backtabs + * for local motions. On many systems, it's not, due to uncertainties about + * tab delays and whether or not tabs will be expanded in raw mode. If you + * have parm_right_cursor, tab motions don't win you a lot anyhow. + */ + +#include <curses.priv.h> +#include <term.h> +#include <ctype.h> + +MODULE_ID("$Id: lib_mvcur.c,v 1.57 1999/06/26 22:16:04 tom Exp $") + +#define STRLEN(s) (s != 0) ? strlen(s) : 0 + +#define CURRENT_ATTR SP->_current_attr /* current phys attribute */ +#define CURRENT_ROW SP->_cursrow /* phys cursor row */ +#define CURRENT_COLUMN SP->_curscol /* phys cursor column */ +#define REAL_ATTR SP->_current_attr /* phys current attribute */ +#define WANT_CHAR(y, x) SP->_newscr->_line[y].text[x] /* desired state */ +#define BAUDRATE cur_term->_baudrate /* bits per second */ + +#if defined(MAIN) || defined(NCURSES_TEST) +#include <sys/time.h> + +static bool profiling = FALSE; +static float diff; +#endif /* MAIN */ + +#define OPT_SIZE 512 + +static int normalized_cost(const char *const cap, int affcnt); + +#if !HAVE_STRSTR +char * _nc_strstr(const char *haystack, const char *needle) +{ + size_t len1 = strlen(haystack); + size_t len2 = strlen(needle); + char *result = 0; + + while ((len1 != 0) && (len1-- >= len2)) { + if (!strncmp(haystack, needle, len2)) { + result = haystack; + break; + } + haystack++; + } + return result; +} +#endif + +/**************************************************************************** + * + * Initialization/wrapup (including cost pre-computation) + * + ****************************************************************************/ + +#ifdef TRACE +static int +trace_cost_of(const char *capname, const char *cap, int affcnt) +{ + int result = _nc_msec_cost(cap,affcnt); + TR(TRACE_CHARPUT|TRACE_MOVE, ("CostOf %s %d", capname, result)); + return result; +} +#define CostOf(cap,affcnt) trace_cost_of(#cap,cap,affcnt); + +static int +trace_normalized_cost(const char *capname, const char *cap, int affcnt) +{ + int result = normalized_cost(cap,affcnt); + TR(TRACE_CHARPUT|TRACE_MOVE, ("NormalizedCost %s %d", capname, result)); + return result; +} +#define NormalizedCost(cap,affcnt) trace_normalized_cost(#cap,cap,affcnt); + +#else + +#define CostOf(cap,affcnt) _nc_msec_cost(cap,affcnt); +#define NormalizedCost(cap,affcnt) normalized_cost(cap,affcnt); + +#endif + +int _nc_msec_cost(const char *const cap, int affcnt) +/* compute the cost of a given operation */ +{ + if (cap == 0) + return(INFINITY); + else + { + const char *cp; + float cum_cost = 0; + + for (cp = cap; *cp; cp++) + { + /* extract padding, either mandatory or required */ + if (cp[0] == '$' && cp[1] == '<' && strchr(cp, '>')) + { + float number = 0; + + for (cp += 2; *cp != '>'; cp++) + { + if (isdigit(*cp)) + number = number * 10 + (*cp - '0'); + else if (*cp == '.') + number += (*++cp - 10) / 10.0; + else if (*cp == '*') + number *= affcnt; + } + + cum_cost += number * 10; + } + else + cum_cost += SP->_char_padding; + } + + return((int)cum_cost); + } +} + +static int normalized_cost(const char *const cap, int affcnt) +/* compute the effective character-count for an operation (round up) */ +{ + int cost = _nc_msec_cost(cap, affcnt); + if (cost != INFINITY) + cost = (cost + SP->_char_padding - 1) / SP->_char_padding; + return cost; +} + +static void reset_scroll_region(void) +/* Set the scroll-region to a known state (the default) */ +{ + if (change_scroll_region) + { + TPUTS_TRACE("change_scroll_region"); + putp(tparm(change_scroll_region, 0, screen_lines - 1)); + } +} + +void _nc_mvcur_resume(void) +/* what to do at initialization time and after each shellout */ +{ + /* initialize screen for cursor access */ + if (enter_ca_mode) + { + TPUTS_TRACE("enter_ca_mode"); + putp(enter_ca_mode); + } + + /* + * Doing this here rather than in _nc_mvcur_wrap() ensures that + * ncurses programs will see a reset scroll region even if a + * program that messed with it died ungracefully. + * + * This also undoes the effects of terminal init strings that assume + * they know the screen size. This is useful when you're running + * a vt100 emulation through xterm. + */ + reset_scroll_region(); + SP->_cursrow = SP->_curscol = -1; + + /* restore cursor shape */ + if (SP->_cursor != -1) + { + int cursor = SP->_cursor; + SP->_cursor = -1; + curs_set (cursor); + } +} + +void _nc_mvcur_init(void) +/* initialize the cost structure */ +{ + /* + * 9 = 7 bits + 1 parity + 1 stop. + */ + SP->_char_padding = (9 * 1000 * 10) / (BAUDRATE > 0 ? BAUDRATE : 9600); + if (SP->_char_padding <= 0) + SP->_char_padding = 1; /* must be nonzero */ + TR(TRACE_CHARPUT|TRACE_MOVE, ("char_padding %d msecs", SP->_char_padding)); + + /* non-parameterized local-motion strings */ + SP->_cr_cost = CostOf(carriage_return, 0); + SP->_home_cost = CostOf(cursor_home, 0); + SP->_ll_cost = CostOf(cursor_to_ll, 0); +#if USE_HARD_TABS + SP->_ht_cost = CostOf(tab, 0); + SP->_cbt_cost = CostOf(back_tab, 0); +#endif /* USE_HARD_TABS */ + SP->_cub1_cost = CostOf(cursor_left, 0); + SP->_cuf1_cost = CostOf(cursor_right, 0); + SP->_cud1_cost = CostOf(cursor_down, 0); + SP->_cuu1_cost = CostOf(cursor_up, 0); + + SP->_smir_cost = CostOf(enter_insert_mode, 0); + SP->_rmir_cost = CostOf(exit_insert_mode, 0); + SP->_ip_cost = 0; + if (insert_padding) { + SP->_ip_cost = CostOf(insert_padding, 0); + } + + /* + * Assumption: if the terminal has memory_relative addressing, the + * initialization strings or smcup will set single-page mode so we + * can treat it like absolute screen addressing. This seems to be true + * for all cursor_mem_address terminal types in the terminfo database. + */ + SP->_address_cursor = cursor_address ? cursor_address : cursor_mem_address; + + /* + * Parametrized local-motion strings. This static cost computation + * depends on the following assumptions: + * + * (1) They never have * padding. In the entire master terminfo database + * as of March 1995, only the obsolete Zenith Z-100 pc violates this. + * (Proportional padding is found mainly in insert, delete and scroll + * capabilities). + * + * (2) The average case of cup has two two-digit parameters. Strictly, + * the average case for a 24 * 80 screen has ((10*10*(1 + 1)) + + * (14*10*(1 + 2)) + (10*70*(2 + 1)) + (14*70*4)) / (24*80) = 3.458 + * digits of parameters. On a 25x80 screen the average is 3.6197. + * On larger screens the value gets much closer to 4. + * + * (3) The average case of cub/cuf/hpa/ech/rep has 2 digits of parameters + * (strictly, (((10 * 1) + (70 * 2)) / 80) = 1.8750). + * + * (4) The average case of cud/cuu/vpa has 2 digits of parameters + * (strictly, (((10 * 1) + (14 * 2)) / 24) = 1.5833). + * + * All these averages depend on the assumption that all parameter values + * are equally probable. + */ + SP->_cup_cost = CostOf(tparm(SP->_address_cursor, 23, 23), 1); + SP->_cub_cost = CostOf(tparm(parm_left_cursor, 23), 1); + SP->_cuf_cost = CostOf(tparm(parm_right_cursor, 23), 1); + SP->_cud_cost = CostOf(tparm(parm_down_cursor, 23), 1); + SP->_cuu_cost = CostOf(tparm(parm_up_cursor, 23), 1); + SP->_hpa_cost = CostOf(tparm(column_address, 23), 1); + SP->_vpa_cost = CostOf(tparm(row_address, 23), 1); + + /* non-parameterized screen-update strings */ + SP->_ed_cost = NormalizedCost(clr_eos, 1); + SP->_el_cost = NormalizedCost(clr_eol, 1); + SP->_el1_cost = NormalizedCost(clr_bol, 1); + SP->_dch1_cost = NormalizedCost(delete_character, 1); + SP->_ich1_cost = NormalizedCost(insert_character, 1); + + /* parameterized screen-update strings */ + SP->_dch_cost = NormalizedCost(tparm(parm_dch, 23), 1); + SP->_ich_cost = NormalizedCost(tparm(parm_ich, 23), 1); + SP->_ech_cost = NormalizedCost(tparm(erase_chars, 23), 1); + SP->_rep_cost = NormalizedCost(tparm(repeat_char, ' ', 23), 1); + + SP->_cup_ch_cost = NormalizedCost(tparm(SP->_address_cursor, 23, 23), 1); + SP->_hpa_ch_cost = NormalizedCost(tparm(column_address, 23), 1); + + /* pre-compute some capability lengths */ + SP->_carriage_return_length = STRLEN(carriage_return); + SP->_cursor_home_length = STRLEN(cursor_home); + SP->_cursor_to_ll_length = STRLEN(cursor_to_ll); + + /* + * If save_cursor is used within enter_ca_mode, we should not use it for + * scrolling optimization, since the corresponding restore_cursor is not + * nested on the various terminals (vt100, xterm, etc.) which use this + * feature. + */ + if (save_cursor != 0 + && enter_ca_mode != 0 + && strstr(enter_ca_mode, save_cursor) != 0) { + T(("...suppressed sc/rc capability due to conflict with smcup/rmcup")); + save_cursor = 0; + restore_cursor = 0; + } + + /* + * A different, possibly better way to arrange this would be to set + * SP->_endwin = TRUE at window initialization time and let this be + * called by doupdate's return-from-shellout code. + */ + _nc_mvcur_resume(); +} + +void _nc_mvcur_wrap(void) +/* wrap up cursor-addressing mode */ +{ + /* leave cursor at screen bottom */ + mvcur(-1, -1, screen_lines - 1, 0); + + /* set cursor to normal mode */ + if (SP->_cursor != -1) + curs_set(1); + + if (exit_ca_mode) + { + TPUTS_TRACE("exit_ca_mode"); + putp(exit_ca_mode); + } + /* + * Reset terminal's tab counter. There's a long-time bug that + * if you exit a "curses" program such as vi or more, tab + * forward, and then backspace, the cursor doesn't go to the + * right place. The problem is that the kernel counts the + * escape sequences that reset things as column positions. + * Utter a \r to reset this invisibly. + */ + _nc_outch('\r'); +} + +/**************************************************************************** + * + * Optimized cursor movement + * + ****************************************************************************/ + +/* + * Perform repeated-append, returning cost + */ +static inline int +repeated_append (int total, int num, int repeat, char *dst, const char *src) +{ + register size_t src_len = strlen(src); + register size_t dst_len = STRLEN(dst); + + if ((dst_len + repeat * src_len) < OPT_SIZE-1) { + total += (num * repeat); + if (dst) { + dst += dst_len; + while (repeat-- > 0) { + (void) strcpy(dst, src); + dst += src_len; + } + } + } else { + total = INFINITY; + } + return total; +} + +#ifndef NO_OPTIMIZE +#define NEXTTAB(fr) (fr + init_tabs - (fr % init_tabs)) + +/* + * Assume back_tab (CBT) does not wrap backwards at the left margin, return + * a negative value at that point to simplify the loop. + */ +#define LASTTAB(fr) ((fr > 0) ? ((fr - 1) / init_tabs) * init_tabs : -1) + +/* Note: we'd like to inline this for speed, but GNU C barfs on the attempt. */ + +static int +relative_move(char *result, int from_y,int from_x,int to_y,int to_x, bool ovw) +/* move via local motions (cuu/cuu1/cud/cud1/cub1/cub/cuf1/cuf/vpa/hpa) */ +{ + int n, vcost = 0, hcost = 0; + + if (result) + result[0] = '\0'; + + if (to_y != from_y) + { + vcost = INFINITY; + + if (row_address) + { + if (result) + (void) strcpy(result, tparm(row_address, to_y)); + vcost = SP->_vpa_cost; + } + + if (to_y > from_y) + { + n = (to_y - from_y); + + if (parm_down_cursor && SP->_cud_cost < vcost) + { + if (result) + (void) strcpy(result, tparm(parm_down_cursor, n)); + vcost = SP->_cud_cost; + } + + if (cursor_down && (n * SP->_cud1_cost < vcost)) + { + if (result) + result[0] = '\0'; + vcost = repeated_append(0, SP->_cud1_cost, n, result, cursor_down); + } + } + else /* (to_y < from_y) */ + { + n = (from_y - to_y); + + if (parm_up_cursor && SP->_cup_cost < vcost) + { + if (result) + (void) strcpy(result, tparm(parm_up_cursor, n)); + vcost = SP->_cup_cost; + } + + if (cursor_up && (n * SP->_cuu1_cost < vcost)) + { + if (result) + result[0] = '\0'; + vcost = repeated_append(0, SP->_cuu1_cost, n, result, cursor_up); + } + } + + if (vcost == INFINITY) + return(INFINITY); + } + + if (result) + result += strlen(result); + + if (to_x != from_x) + { + char str[OPT_SIZE]; + + hcost = INFINITY; + + if (column_address) + { + if (result) + (void) strcpy(result, tparm(column_address, to_x)); + hcost = SP->_hpa_cost; + } + + if (to_x > from_x) + { + n = to_x - from_x; + + if (parm_right_cursor && SP->_cuf_cost < hcost) + { + if (result) + (void) strcpy(result, tparm(parm_right_cursor, n)); + hcost = SP->_cuf_cost; + } + + if (cursor_right) + { + int lhcost = 0; + + str[0] = '\0'; + +#if USE_HARD_TABS + /* use hard tabs, if we have them, to do as much as possible */ + if (init_tabs > 0 && tab) + { + int nxt, fr; + + for (fr = from_x; (nxt = NEXTTAB(fr)) <= to_x; fr = nxt) + { + lhcost = repeated_append(lhcost, SP->_ht_cost, 1, str, tab); + if (lhcost == INFINITY) + break; + } + + n = to_x - fr; + from_x = fr; + } +#endif /* USE_HARD_TABS */ + +#if defined(REAL_ATTR) && defined(WANT_CHAR) +#ifdef BSD_TPUTS + /* + * If we're allowing BSD-style padding in tputs, don't generate + * a string with a leading digit. Otherwise, that will be + * interpreted as a padding value rather than sent to the + * screen. + */ + if (ovw + && n > 0 + && vcost == 0 + && str[0] == '\0' + && isdigit(TextOf(WANT_CHAR(to_y, from_x)))) + ovw = FALSE; +#endif + /* + * If we have no attribute changes, overwrite is cheaper. + * Note: must suppress this by passing in ovw = FALSE whenever + * WANT_CHAR would return invalid data. In particular, this + * is true between the time a hardware scroll has been done + * and the time the structure WANT_CHAR would access has been + * updated. + */ + if (ovw) + { + int i; + + for (i = 0; i < n; i++) + if ((WANT_CHAR(to_y, from_x + i) & A_ATTRIBUTES) != CURRENT_ATTR) + { + ovw = FALSE; + break; + } + } + if (ovw) + { + char *sp; + int i; + + sp = str + strlen(str); + + for (i = 0; i < n; i++) + *sp++ = WANT_CHAR(to_y, from_x + i); + *sp = '\0'; + lhcost += n * SP->_char_padding; + } + else +#endif /* defined(REAL_ATTR) && defined(WANT_CHAR) */ + { + lhcost = repeated_append(lhcost, SP->_cuf1_cost, n, str, cursor_right); + } + + if (lhcost < hcost) + { + if (result) + (void) strcpy(result, str); + hcost = lhcost; + } + } + } + else /* (to_x < from_x) */ + { + n = from_x - to_x; + + if (parm_left_cursor && SP->_cub_cost < hcost) + { + if (result) + (void) strcpy(result, tparm(parm_left_cursor, n)); + hcost = SP->_cub_cost; + } + + if (cursor_left) + { + int lhcost = 0; + + str[0] = '\0'; + +#if USE_HARD_TABS + if (init_tabs > 0 && back_tab) + { + int nxt, fr; + + for (fr = from_x; (nxt = LASTTAB(fr)) >= to_x; fr = nxt) + { + lhcost = repeated_append(lhcost, SP->_cbt_cost, 1, str, back_tab); + if (lhcost == INFINITY) + break; + } + + n = fr - to_x; + } +#endif /* USE_HARD_TABS */ + + lhcost = repeated_append(lhcost, SP->_cub1_cost, n, str, cursor_left); + + if (lhcost < hcost) + { + if (result) + (void) strcpy(result, str); + hcost = lhcost; + } + } + } + + if (hcost == INFINITY) + return(INFINITY); + } + + return(vcost + hcost); +} +#endif /* !NO_OPTIMIZE */ + +/* + * With the machinery set up above, it's conceivable that + * onscreen_mvcur could be modified into a recursive function that does + * an alpha-beta search of motion space, as though it were a chess + * move tree, with the weight function being boolean and the search + * depth equated to length of string. However, this would jack up the + * computation cost a lot, especially on terminals without a cup + * capability constraining the search tree depth. So we settle for + * the simpler method below. + */ + +static inline int +onscreen_mvcur(int yold,int xold,int ynew,int xnew, bool ovw) +/* onscreen move from (yold, xold) to (ynew, xnew) */ +{ + char use[OPT_SIZE], *sp; + int tactic = 0, newcost, usecost = INFINITY; + +#if defined(MAIN) || defined(NCURSES_TEST) + struct timeval before, after; + + gettimeofday(&before, NULL); +#endif /* MAIN */ + + /* tactic #0: use direct cursor addressing */ + sp = tparm(SP->_address_cursor, ynew, xnew); + if (sp) + { + tactic = 0; + (void) strcpy(use, sp); + usecost = SP->_cup_cost; + +#if defined(TRACE) || defined(NCURSES_TEST) + if (!(_nc_optimize_enable & OPTIMIZE_MVCUR)) + goto nonlocal; +#endif /* TRACE */ + + /* + * We may be able to tell in advance that the full optimization + * will probably not be worth its overhead. Also, don't try to + * use local movement if the current attribute is anything but + * A_NORMAL...there are just too many ways this can screw up + * (like, say, local-movement \n getting mapped to some obscure + * character because A_ALTCHARSET is on). + */ + if (yold == -1 || xold == -1 || NOT_LOCAL(yold, xold, ynew, xnew)) + { +#if defined(MAIN) || defined(NCURSES_TEST) + if (!profiling) + { + (void) fputs("nonlocal\n", stderr); + goto nonlocal; /* always run the optimizer if profiling */ + } +#else + goto nonlocal; +#endif /* MAIN */ + } + } + +#ifndef NO_OPTIMIZE + /* tactic #1: use local movement */ + if (yold != -1 && xold != -1 + && ((newcost=relative_move(NULL, yold, xold, ynew, xnew, ovw))!=INFINITY) + && newcost < usecost) + { + tactic = 1; + usecost = newcost; + } + + /* tactic #2: use carriage-return + local movement */ + if (yold != -1 && carriage_return + && ((newcost=relative_move(NULL, yold,0,ynew,xnew, ovw)) != INFINITY) + && SP->_cr_cost + newcost < usecost) + { + tactic = 2; + usecost = SP->_cr_cost + newcost; + } + + /* tactic #3: use home-cursor + local movement */ + if (cursor_home + && ((newcost=relative_move(NULL, 0, 0, ynew, xnew, ovw)) != INFINITY) + && SP->_home_cost + newcost < usecost) + { + tactic = 3; + usecost = SP->_home_cost + newcost; + } + + /* tactic #4: use home-down + local movement */ + if (cursor_to_ll + && ((newcost=relative_move(NULL, screen_lines-1, 0, ynew, xnew, ovw)) != INFINITY) + && SP->_ll_cost + newcost < usecost) + { + tactic = 4; + usecost = SP->_ll_cost + newcost; + } + + /* + * tactic #5: use left margin for wrap to right-hand side, + * unless strange wrap behavior indicated by xenl might hose us. + */ + if (auto_left_margin && !eat_newline_glitch + && yold > 0 && cursor_left + && ((newcost=relative_move(NULL, yold-1, screen_columns-1, ynew, xnew, ovw)) != INFINITY) + && SP->_cr_cost + SP->_cub1_cost + newcost + newcost < usecost) + { + tactic = 5; + usecost = SP->_cr_cost + SP->_cub1_cost + newcost; + } + + /* + * These cases are ordered by estimated relative frequency. + */ + if (tactic) + { + if (tactic == 1) + (void) relative_move(use, yold, xold, ynew, xnew, ovw); + else if (tactic == 2) + { + (void) strcpy(use, carriage_return); + (void) relative_move(use + SP->_carriage_return_length, + yold,0,ynew,xnew, ovw); + } + else if (tactic == 3) + { + (void) strcpy(use, cursor_home); + (void) relative_move(use + SP->_cursor_home_length, + 0, 0, ynew, xnew, ovw); + } + else if (tactic == 4) + { + (void) strcpy(use, cursor_to_ll); + (void) relative_move(use + SP->_cursor_to_ll_length, + screen_lines-1, 0, ynew, xnew, ovw); + } + else /* if (tactic == 5) */ + { + use[0] = '\0'; + if (xold > 0) + (void) strcat(use, carriage_return); + (void) strcat(use, cursor_left); + (void) relative_move(use + strlen(use), + yold-1, screen_columns-1, ynew, xnew, ovw); + } + } +#endif /* !NO_OPTIMIZE */ + +#if defined(MAIN) || defined(NCURSES_TEST) + gettimeofday(&after, NULL); + diff = after.tv_usec - before.tv_usec + + (after.tv_sec - before.tv_sec) * 1000000; + if (!profiling) + (void) fprintf(stderr, "onscreen: %d msec, %f 28.8Kbps char-equivalents\n", + (int)diff, diff/288); +#endif /* MAIN */ + + nonlocal: + if (usecost != INFINITY) + { + TPUTS_TRACE("mvcur"); + tputs(use, 1, _nc_outch); + return(OK); + } + else + return(ERR); +} + +int mvcur(int yold, int xold, int ynew, int xnew) +/* optimized cursor move from (yold, xold) to (ynew, xnew) */ +{ + TR(TRACE_MOVE, ("mvcur(%d,%d,%d,%d) called", yold, xold, ynew, xnew)); + + if (yold == ynew && xold == xnew) + return(OK); + + /* + * Most work here is rounding for terminal boundaries getting the + * column position implied by wraparound or the lack thereof and + * rolling up the screen to get ynew on the screen. + */ + + if (xnew >= screen_columns) + { + ynew += xnew / screen_columns; + xnew %= screen_columns; + } + if (xold >= screen_columns) + { + int l; + + l = (xold + 1) / screen_columns; + yold += l; + if (yold >= screen_lines) + l -= (yold - screen_lines - 1); + + while (l > 0) { + if (newline) + { + TPUTS_TRACE("newline"); + tputs(newline, 0, _nc_outch); + } + else + putchar('\n'); + l--; + if (xold > 0) + { + if (carriage_return) + { + TPUTS_TRACE("carriage_return"); + tputs(carriage_return, 0, _nc_outch); + } + else + putchar('\r'); + xold = 0; + } + } + } + + if (yold > screen_lines - 1) + yold = screen_lines - 1; + if (ynew > screen_lines - 1) + ynew = screen_lines - 1; + + /* destination location is on screen now */ + return(onscreen_mvcur(yold, xold, ynew, xnew, TRUE)); +} + +#if defined(TRACE) || defined(NCURSES_TEST) +int _nc_optimize_enable = OPTIMIZE_ALL; +#endif + +#if defined(MAIN) || defined(NCURSES_TEST) +/**************************************************************************** + * + * Movement optimizer test code + * + ****************************************************************************/ + +#include <tic.h> +#include <dump_entry.h> + +const char *_nc_progname = "mvcur"; + +static unsigned long xmits; + +int tputs(const char *string, int affcnt GCC_UNUSED, int (*outc)(int) GCC_UNUSED) +/* stub tputs() that dumps sequences in a visible form */ +{ + if (profiling) + xmits += strlen(string); + else + (void) fputs(_nc_visbuf(string), stdout); + return(OK); +} + +int putp(const char *string) +{ + return(tputs(string, 1, _nc_outch)); +} + +int _nc_outch(int ch) +{ + putc(ch, stdout); + return OK; +} + +static char tname[MAX_ALIAS]; + +static void load_term(void) +{ + (void) setupterm(tname, STDOUT_FILENO, NULL); +} + +static int roll(int n) +{ + int i, j; + + i = (RAND_MAX / n) * n; + while ((j = rand()) >= i) + continue; + return (j % n); +} + +int main(int argc GCC_UNUSED, char *argv[] GCC_UNUSED) +{ + (void) strcpy(tname, termname()); + load_term(); + _nc_setupscreen(lines, columns, stdout); + baudrate(); + + _nc_mvcur_init(); + NC_BUFFERED(FALSE); + + (void) puts("The mvcur tester. Type ? for help"); + + fputs("smcup:", stdout); + putchar('\n'); + + for (;;) + { + int fy, fx, ty, tx, n, i; + char buf[BUFSIZ], capname[BUFSIZ]; + + (void) fputs("> ", stdout); + (void) fgets(buf, sizeof(buf), stdin); + + if (buf[0] == '?') + { +(void) puts("? -- display this help message"); +(void) puts("fy fx ty tx -- (4 numbers) display (fy,fx)->(ty,tx) move"); +(void) puts("s[croll] n t b m -- display scrolling sequence"); +(void) printf("r[eload] -- reload terminal info for %s\n", termname()); +(void) puts("l[oad] <term> -- load terminal info for type <term>"); +(void) puts("d[elete] <cap> -- delete named capability"); +(void) puts("i[nspect] -- display terminal capabilities"); +(void) puts("c[ost] -- dump cursor-optimization cost table"); +(void) puts("o[optimize] -- toggle movement optimization"); +(void) puts("t[orture] <num> -- torture-test with <num> random moves"); +(void) puts("q[uit] -- quit the program"); + } + else if (sscanf(buf, "%d %d %d %d", &fy, &fx, &ty, &tx) == 4) + { + struct timeval before, after; + + putchar('"'); + + gettimeofday(&before, NULL); + mvcur(fy, fx, ty, tx); + gettimeofday(&after, NULL); + + printf("\" (%ld msec)\n", + (long)(after.tv_usec - before.tv_usec + (after.tv_sec - before.tv_sec) * 1000000)); + } + else if (sscanf(buf, "s %d %d %d %d", &fy, &fx, &ty, &tx) == 4) + { + struct timeval before, after; + + putchar('"'); + + gettimeofday(&before, NULL); + _nc_scrolln(fy, fx, ty, tx); + gettimeofday(&after, NULL); + + printf("\" (%ld msec)\n", + (long)(after.tv_usec - before.tv_usec + (after.tv_sec - before.tv_sec) * 1000000)); + } + else if (buf[0] == 'r') + { + (void) strcpy(tname, termname()); + load_term(); + } + else if (sscanf(buf, "l %s", tname) == 1) + { + load_term(); + } + else if (sscanf(buf, "d %s", capname) == 1) + { + struct name_table_entry const *np = _nc_find_entry(capname, + _nc_info_hash_table); + + if (np == NULL) + (void) printf("No such capability as \"%s\"\n", capname); + else + { + switch(np->nte_type) + { + case BOOLEAN: + cur_term->type.Booleans[np->nte_index] = FALSE; + (void) printf("Boolean capability `%s' (%d) turned off.\n", + np->nte_name, np->nte_index); + break; + + case NUMBER: + cur_term->type.Numbers[np->nte_index] = -1; + (void) printf("Number capability `%s' (%d) set to -1.\n", + np->nte_name, np->nte_index); + break; + + case STRING: + cur_term->type.Strings[np->nte_index] = (char *)NULL; + (void) printf("String capability `%s' (%d) deleted.\n", + np->nte_name, np->nte_index); + break; + } + } + } + else if (buf[0] == 'i') + { + dump_init((char *)NULL, F_TERMINFO, S_TERMINFO, 70, 0, FALSE); + dump_entry(&cur_term->type, FALSE, TRUE, 0); + putchar('\n'); + } + else if (buf[0] == 'o') + { + if (_nc_optimize_enable & OPTIMIZE_MVCUR) + { + _nc_optimize_enable &=~ OPTIMIZE_MVCUR; + (void) puts("Optimization is now off."); + } + else + { + _nc_optimize_enable |= OPTIMIZE_MVCUR; + (void) puts("Optimization is now on."); + } + } + /* + * You can use the `t' test to profile and tune the movement + * optimizer. Use iteration values in three digits or more. + * At above 5000 iterations the profile timing averages are stable + * to within a millisecond or three. + * + * The `overhead' field of the report will help you pick a + * COMPUTE_OVERHEAD figure appropriate for your processor and + * expected line speed. The `total estimated time' is + * computation time plus a character-transmission time + * estimate computed from the number of transmits and the baud + * rate. + * + * Use this together with the `o' command to get a read on the + * optimizer's effectiveness. Compare the total estimated times + * for `t' runs of the same length in both optimized and un-optimized + * modes. As long as the optimized times are less, the optimizer + * is winning. + */ + else if (sscanf(buf, "t %d", &n) == 1) + { + float cumtime = 0, perchar; + int speeds[] = {2400, 9600, 14400, 19200, 28800, 38400, 0}; + + srand((unsigned)(getpid() + time((time_t *)0))); + profiling = TRUE; + xmits = 0; + for (i = 0; i < n; i++) + { + /* + * This does a move test between two random locations, + * Random moves probably short-change the optimizer, + * which will work better on the short moves probably + * typical of doupdate()'s usage pattern. Still, + * until we have better data... + */ +#ifdef FIND_COREDUMP + int from_y = roll(lines); + int to_y = roll(lines); + int from_x = roll(columns); + int to_x = roll(columns); + + printf("(%d,%d) -> (%d,%d)\n", from_y, from_x, to_y, to_x); + mvcur(from_y, from_x, to_y, to_x); +#else + mvcur(roll(lines), roll(columns), roll(lines), roll(columns)); +#endif /* FIND_COREDUMP */ + if (diff) + cumtime += diff; + } + profiling = FALSE; + + /* + * Average milliseconds per character optimization time. + * This is the key figure to watch when tuning the optimizer. + */ + perchar = cumtime / n; + + (void) printf("%d moves (%ld chars) in %d msec, %f msec each:\n", + n, xmits, (int)cumtime, perchar); + + for (i = 0; speeds[i]; i++) + { + /* + * Total estimated time for the moves, computation and + * transmission both. Transmission time is an estimate + * assuming 9 bits/char, 8 bits + 1 stop bit. + */ + float totalest = cumtime + xmits * 9 * 1e6 / speeds[i]; + + /* + * Per-character optimization overhead in character transmits + * at the current speed. Round this to the nearest integer + * to figure COMPUTE_OVERHEAD for the speed. + */ + float overhead = speeds[i] * perchar / 1e6; + + (void) printf("%6d bps: %3.2f char-xmits overhead; total estimated time %15.2f\n", + speeds[i], overhead, totalest); + } + } + else if (buf[0] == 'c') + { + (void) printf("char padding: %d\n", SP->_char_padding); + (void) printf("cr cost: %d\n", SP->_cr_cost); + (void) printf("cup cost: %d\n", SP->_cup_cost); + (void) printf("home cost: %d\n", SP->_home_cost); + (void) printf("ll cost: %d\n", SP->_ll_cost); +#if USE_HARD_TABS + (void) printf("ht cost: %d\n", SP->_ht_cost); + (void) printf("cbt cost: %d\n", SP->_cbt_cost); +#endif /* USE_HARD_TABS */ + (void) printf("cub1 cost: %d\n", SP->_cub1_cost); + (void) printf("cuf1 cost: %d\n", SP->_cuf1_cost); + (void) printf("cud1 cost: %d\n", SP->_cud1_cost); + (void) printf("cuu1 cost: %d\n", SP->_cuu1_cost); + (void) printf("cub cost: %d\n", SP->_cub_cost); + (void) printf("cuf cost: %d\n", SP->_cuf_cost); + (void) printf("cud cost: %d\n", SP->_cud_cost); + (void) printf("cuu cost: %d\n", SP->_cuu_cost); + (void) printf("hpa cost: %d\n", SP->_hpa_cost); + (void) printf("vpa cost: %d\n", SP->_vpa_cost); + } + else if (buf[0] == 'x' || buf[0] == 'q') + break; + else + (void) puts("Invalid command."); + } + + (void) fputs("rmcup:", stdout); + _nc_mvcur_wrap(); + putchar('\n'); + + return(0); +} + +#endif /* MAIN */ + +/* lib_mvcur.c ends here */ |