void
move_cursor_to_start_of_prompt(int erase)
{
int termwidth = winsize.ws_col;
int promptlen_on_screen, number_of_lines_in_prompt, curpos, count;
int cooked = (saved_rl_state.cooked_prompt != NULL);
DPRINTF2(DEBUG_READLINE,"prompt_is_still_uncooked: %d, impatient_prompt: %d", prompt_is_still_uncooked, impatient_prompt);
if (prompt_is_still_uncooked && ! impatient_prompt)
return; /* @@@ is this necessary ?*/
promptlen_on_screen = colourless_strlen_unmarked(saved_rl_state.cooked_prompt ? saved_rl_state.cooked_prompt : saved_rl_state.raw_prompt, termwidth);
curpos = (within_line_edit ? 1 : 0); /* if the user has pressed a key the cursor will be 1 past the current prompt */
assert(termwidth > 0);
number_of_lines_in_prompt = 1 + ((promptlen_on_screen + curpos -1) / termwidth); /* integer arithmetic! (e.g. 171/80 = 2) */
cr();
for (count = 0; count < number_of_lines_in_prompt -1; count++) {
if (erase)
clear_line();
curs_up();
}
clear_line();
DPRINTF4(DEBUG_READLINE,"moved cursor up %d lines (erase = %d, len=%d, cooked=%d)", number_of_lines_in_prompt - 1, erase, promptlen_on_screen, cooked);
}
int
colourless_strlen(const char *str, char **pcopy_without_ignore_markers, int termwidth)
{
int visible = TRUE;
int column = 0;
int length = strlen(str);
const char *p;
char *q, *copy_without_ignore_markers;
assert(termwidth >= 0);
q = copy_without_ignore_markers = mymalloc(length + 1);
for(p = str; *p; p++) {
assert (q < copy_without_ignore_markers + length);
switch (*p) {
case RL_PROMPT_START_IGNORE:
visible = FALSE;
continue;
case RL_PROMPT_END_IGNORE:
visible = TRUE;
continue;
case '\r':
if (visible) {
q -= column;
column = 0;
continue;
}
break;
case '\b':
if ((visible && q > copy_without_ignore_markers)) {
q -= 1;
column -= 1;
if (termwidth && column < 0)
column += termwidth;
continue;
}
break;
}
if (visible) {
*q++ = *p;
column +=1;
if (termwidth && column >= termwidth)
column -= termwidth;
}
}
*q = '\0';
DPRINTF4(DEBUG_READLINE, "colourless_strlen(\"%s\", 0x%lx, %d) = %ld",
mangle_string_for_debug_log(str, MANGLE_LENGTH), (long) pcopy_without_ignore_markers,
termwidth, q - copy_without_ignore_markers);
if (pcopy_without_ignore_markers)
*pcopy_without_ignore_markers = copy_without_ignore_markers;
else
free (copy_without_ignore_markers);
return q - copy_without_ignore_markers;
}
static void bindkey(int key, rl_command_func_t *function, const char *maplist) {
char *mapnames[] = {"emacs-standard","emacs-ctlx","emacs-meta","vi-insert","vi-move","vi-command",NULL};
char **mapname;
for (mapname = mapnames; *mapname; mapname++)
if(strstr(maplist, *mapname)) {
Keymap kmap = getmap(*mapname);
DPRINTF4(DEBUG_READLINE,"Binding key %d (%s) in keymap '%s' to <0x%lx>", key, mangle_char_for_debug_log(key,TRUE), *mapname, (long) function);
if (rl_bind_key_in_map(key, function, kmap))
myerror(FATAL|NOERRNO, "Could not bind key %d (%s) in keymap '%s'", key, mangle_char_for_debug_log(key,TRUE), *mapname);
}
}
void addSymbol(S_symbol *pp, S_symTab *tab) {
/* a bug can produce, if you add a symbol into old table, and the same
symbol exists in a newer one. Then it will be deleted from the newer
one. All this story is about storing information in trail. It should
containt, both table and pointer !!!!
*/
DPRINTF4("adding symbol %s: %s %s\n",pp->name, typesName[pp->b.symType], storagesName[pp->b.storage]);
assert(pp->b.npointers==0);
AddSymbolNoTrail(pp,tab);
addToTrail(deleteSymDef, pp /* AND ALSO!!! , tab */ );
//if (WORK_NEST_LEVEL0()) {static int c=0;fprintf(dumpOut,"addsym0#%d\n",c++);}
}
void
cleanup_rlwrap_and_exit(int status)
{
unblock_all_signals();
DPRINTF0(DEBUG_TERMIO, "Cleaning up");
if (write_histfile && (histsize==0 || history_total_bytes() > 0)) /* avoid creating empty .speling_eror_history file after typo */
write_history(history_filename); /* ignore errors */
close_logfile();
DPRINTF4(DEBUG_SIGNALS, "command_pid: %d, commands_exit_status: %x, filter_pid: %d, filters_exit_status: %x",
command_pid, commands_exit_status, filter_pid, filters_exit_status);
mymicrosleep(10); /* we may have got an EOF or EPIPE because the filter or command died, but this doesn't mean that
SIGCHLD has been caught already. Taking a little nap now improves the chance that we will catch it
(no grave problem if we miss it, but diagnostics, exit status and transparent signal handling depend on it) */
if (filter_pid)
kill_filter();
else if (filter_is_dead) {
int filters_killer = killed_by(filters_exit_status);
errno = 0;
mywarn((filters_killer ? "filter was killed by signal %d (%s)": WEXITSTATUS(filters_exit_status) ? "filter died" : "filter exited"), filters_killer, signal_name(filters_killer));
}
if (debug)
debug_postmortem();
if (terminal_settings_saved)
if (tcsetattr(STDIN_FILENO, TCSAFLUSH, &saved_terminal_settings) < 0) /* ignore errors (almost dead anyway) */
; /* fprintf(stderr, "Arggh\n"); don't use myerror!!*/
if (!newline_came_last) /* print concluding newline, if necessary */
my_putstr("\n");
if (status != EXIT_SUCCESS) /* rlwrap itself has failed, rather than the wrapped command */
exit(status);
else {
int commands_killer = killed_by(commands_exit_status);
if (commands_killer)
suicide_by(commands_killer, commands_exit_status); /* command terminated by signal, make rlwrap's
parent believe rlwrap was killed by it */
else
exit(WEXITSTATUS(commands_exit_status)); /* propagate command's exit status */
}
}
int
adapt_tty_winsize(int from_fd, int to_fd)
{
int ret;
struct winsize old_winsize = winsize;
ret = ioctl(from_fd, TIOCGWINSZ, &winsize);
DPRINTF1(DEBUG_SIGNALS, "ioctl (..., TIOCGWINSZ) = %d", ret);
if (winsize.ws_col != old_winsize.ws_col || winsize.ws_row != old_winsize.ws_row) {
DPRINTF4(DEBUG_SIGNALS, "ws.col: %d -> %d, ws.row: %d -> %d", old_winsize.ws_col, winsize.ws_col, old_winsize.ws_row, winsize.ws_row);
if (always_readline &&!dont_wrap_command_waits()) /* if --always_readline option is set, the client will probably spew a */
deferred_adapt_commands_window_size = TRUE; /* volley of control chars at us when its terminal is resized. Hence we don't do it now */
else {
ret = ioctl(to_fd, TIOCSWINSZ, &winsize);
DPRINTF1(DEBUG_SIGNALS, "ioctl (..., TIOCSWINSZ) = %d", ret);
}
DPRINTF2(DEBUG_READLINE, "rl_prompt: %s, line_buffer: %s", mangle_string_for_debug_log(rl_prompt, 100), rl_line_buffer);
rl_set_screen_size(winsize.ws_row, winsize.ws_col); /* this is safe: we know that right now rlwrap is not calling
the readline library because we keep SIGWINCH blocked all the time */
if (!within_line_edit && !skip_rlwrap()) {
wipe_textarea(&old_winsize);
received_WINCH = TRUE; /* we can't start line edit in signal handler, so we only set a flag */
} else if (within_line_edit) { /* try to keep displayed line tidy */
wipe_textarea(&old_winsize);
rl_on_new_line();
rl_redisplay();
}
return (!always_readline || dont_wrap_command_waits()); /* pass the signal on (except when always_readline is set and command is not waiting) */
} else { /* window size has not changed */
return FALSE;
}
}
/* format line and add it to history list, avoiding duplicates if necessary */
static void
my_add_history(char *line)
{
int lookback, count, here;
char *new_entry, *filtered_line;
filtered_line = pass_through_filter(TAG_HISTORY, line);
switch (history_duplicate_avoidance_policy) {
case KEEP_ALL_DOUBLES:
lookback = 0; break;
case ELIMINATE_SUCCESIVE_DOUBLES:
lookback = 1; break;
case ELIMINATE_ALL_DOUBLES:
lookback = history_length; break;
}
new_entry = filtered_line;
lookback = min(history_length, lookback);
for (count = 0, here = history_length - 1;
count < lookback ;
count++, here--) {
DPRINTF4(DEBUG_READLINE, "strcmping <%s> and <%s> (count = %d, here = %d)", line, history_get(history_base + here)->line ,count, here);
if (strncmp(new_entry, history_get(history_base + here) -> line, 10000) == 0) { /* history_get uses the logical offset history_base .. */
HIST_ENTRY *entry = remove_history (here); /* .. but remove_history doesn't! */
DPRINTF2(DEBUG_READLINE, "removing duplicate entry #%d (%s)", here, entry->line);
free_foreign(entry->line);
free_foreign(entry);
}
}
add_history(new_entry);
free(new_entry);
}
/*
Given the actual data set <dataSet>, estimates the values for
algorithm parameters that would give the optimal running time of a
query.
The set <sampleQueries> is a set with query sample points
(R-NN DS's parameters are optimized for query points from the set
<sampleQueries>). <sampleQueries> could be a sample of points from the
actual query set or from the data set. When computing the estimated
number of collisions of a sample query point <q> with the data set
points, if there is a point in the data set with the same pointer
with <q> (that is when <q> is a data set point), then the
corresponding point (<q>) is not considered in the data set (for the
purpose of computing the respective #collisions estimation).
//如果查询样本点来自数据集，则为了考虑碰撞估计，不讲对应的样本查询点考虑进数据集合
The return value is the estimate of the optimal parameters.
*/
RNNParametersT computeOptimalParameters(RealT R, //半径
RealT successProbability, //成功率
IntT nPoints, //数据点的个数
IntT dimension, //维度
PPointT *dataSet, //数据集合
IntT nSampleQueries, //样本查询点的个数
PPointT *sampleQueries, //样本查询点的数据集
MemVarT memoryUpperBound){ //每个半径需要使用内存的上界，多个半径的时候Qboundry才会使用到，用一个半径的时候，是使用不到的。
ASSERT(nSampleQueries > 0);
initializeLSHGlobal(); //设置了一个时间参数，把时间函数所花的时间求出来，得出精确操作所花费的时间
//用来评估算法效率
RNNParametersT optParameters;
optParameters.successProbability = successProbability;
optParameters.dimension = dimension;
optParameters.parameterR = R;
#ifdef USE_L1_DISTANCE
optParameters.parameterR2 = R; //L1 norm下R2 与R相等， //R与R2的作用？
#else
optParameters.parameterR2 = SQR(R); //？R2的作用是什么
#endif
optParameters.useUfunctions = TRUE; // TODO: could optimize here:
// maybe sometimes, the old way
// was better.
optParameters.parameterW = PARAMETER_W_DEFAULT; //W用于计算L2 norm下的LSH函数的参数，4
optParameters.parameterT = nPoints; //数据点的个数
optParameters.typeHT = HT_HYBRID_CHAINS; //hashing表构建的类型
// Compute the run-time parameters (timings of different parts of the algorithm).
IntT nReps = 10; // # number of repetitiions //重复的数量为10个
RealT lshPrecomp = 0, uhashOver = 0, distComp = 0; //LSH预处理的时间，从Hash表中取得桶的时间 ，计算两点距离的时间
for(IntT i = 0; i < nReps; i++){ //重复nReps次，计算LSH 3个阶段的时间，然后在计算出来的平均时间。
RealT lP, uO, dC; //并且进行了哈希表的构建，哈希函数的计算，等等
determineRTCoefficients(optParameters.parameterR,
optParameters.successProbability,
optParameters.useUfunctions,
optParameters.typeHT,
optParameters.dimension,
nPoints,
dataSet,
lP,
uO,
dC);
lshPrecomp += lP; //累计重复nReps次的时间。 因为随机数产生是不同的，所以每次处理的时间也不同
uhashOver += uO;
distComp += dC;
DPRINTF4("Coefs: lP = %0.9lf\tuO = %0.9lf\tdC = %0.9lf\n", lP, uO, dC);
}
lshPrecomp /= nReps; //平均处理的时间
uhashOver /= nReps;
distComp /= nReps;
DPRINTF("Coefs (final): lshPrecomp = %0.9lf\n", lshPrecomp);
DPRINTF("Coefs (final): uhashOver = %0.9lf\n", uhashOver);
DPRINTF("Coefs (final): distComp = %0.9lf\n", distComp);
DPRINTF("Remaining memory: %lld\n", memoryUpperBound);
// Try all possible <k>s and choose the one for which the time
// estimate of a query is minimal.
IntT k;
RealT timeLSH, timeUH, timeCycling;
//IntT queryIndex = genRandomInt(0, nPoints);
//PPointT query = dataSet[queryIndex]; // query points = a random points from the data set. //查询点取为从数据集中取的随机点
IntT bestK = 0;
RealT bestTime = 0;
for(k = 2; ; k += 2){ //之前进行计算的时候，将k预设为16
DPRINTF("ST. k = %d\n", k);
IntT m = computeMForULSH(k, successProbability); //通过k和p计算出m
IntT L = m * (m-1) / 2;
//DPRINTF("Available memory: %lld\n", getAvailableMemory());
if (L * nPoints > memoryUpperBound / 12){ //??为什么除以12呢？？ 12是一个常数，用于对内存的估算 //4×tableSize+8×#buckets+8n<20n
break;
}
timeLSH = m * k * lshPrecomp; //m次，k维预处理时间，之前有计算好平均每次处理的时间
timeUH = L * uhashOver; //L个hash表
//RealT nCollisions = estimateNCollisionsFromDSPoint(nPoints, dimension, dataSet, queryIndex, k, L, R);
// Compute the mean number of collisions for the points from the sample query set.
RealT nCollisions = 0; //计算平均冲突的点的数量
for(IntT i = 0; i < nSampleQueries; i++){
nCollisions += estimateNDistinctCollisions(nPoints, dimension, dataSet, sampleQueries[i], TRUE, k, m, R);
}
nCollisions /= nSampleQueries; //每个查询点能查询到近邻点的平均数量
timeCycling = nCollisions * distComp; //查询点查到所有近邻的时间？？
DPRINTF3("ST.m=%d L=%d \n", m, L);
DPRINTF("ST.Estimated # distinct collisions = %0.6lf\n", (double)nCollisions);
DPRINTF("ST.TimeLSH = %0.6lf\n", timeLSH);
DPRINTF("ST.TimeUH = %0.6lf\n", timeUH);
DPRINTF("ST.TimeCycling = %0.6lf\n", timeCycling);
DPRINTF("ST.Sum = %0.6lf\n", timeLSH + timeUH + timeCycling);
if (bestK == 0 || (timeLSH + timeUH + timeCycling) < bestTime) { //第一轮循环k=0,此时bestTime是比较大的，所以随着k的增大，得到最优时间
bestK = k;
bestTime = timeLSH + timeUH + timeCycling;
}
ASSERT(k < 100); // otherwise, k reached 100 -- which, from
// experience, should never happen for reasonable
// data set & available memory amount.
}
DPRINTF("STO.Optimal k = %d\n", bestK);
IntT m = computeMForULSH(bestK, successProbability); //计算L2 norm下的m
IntT L = m * (m-1) / 2;
timeLSH = m * bestK * lshPrecomp;
timeUH = L * uhashOver;
// Compute the mean number of collisions for the points from the sample query set.
RealT nCollisions = 0;
for(IntT i = 0; i < nSampleQueries; i++){
nCollisions += estimateNDistinctCollisions(nPoints, dimension, dataSet, sampleQueries[i], TRUE, k, m, R);
}
nCollisions /= nSampleQueries;
// timeCycling = estimateNCollisionsFromDSPoint(nPoints, dimension, dataSet, queryIndex, bestK, L, R) * distComp;
timeCycling = nCollisions * distComp;
DPRINTF("STO.TimeLSH = %0.6lf\n", timeLSH);
DPRINTF("STO.TimeUH = %0.6lf\n", timeUH);
DPRINTF("STO.TimeCycling = %0.6lf\n", timeCycling);
DPRINTF("STO.Sum = %0.6lf\n", timeLSH + timeUH + timeCycling);
optParameters.parameterK = bestK;
optParameters.parameterM = m;
optParameters.parameterL = L;
return optParameters;
}
/*
Given the actual data set <dataSet>, estimates the values for
algorithm parameters that would give the optimal running time of a
query.
The set <sampleQueries> is a set with query sample points
(R-NN DS's parameters are optimized for query points from the set
<sampleQueries>). <sampleQueries> could be a sample of points from the
actual query set or from the data set. When computing the estimated
number of collisions of a sample query point <q> with the data set
points, if there is a point in the data set with the same pointer
with <q> (that is when <q> is a data set point), then the
corresponding point (<q>) is not considered in the data set (for the
purpose of computing the respective #collisions estimation).
The return value is the estimate of the optimal parameters.
*/
RNNParametersT computeOptimalParameters(RealT R,
RealT successProbability,
IntT nPoints,
IntT dimension,
PPointT *dataSet,
IntT nSampleQueries,
PPointT *sampleQueries,
MemVarT memoryUpperBound){
ASSERT(nSampleQueries > 0);
initializeLSHGlobal();
RNNParametersT optParameters;
optParameters.successProbability = successProbability;
optParameters.dimension = dimension;
optParameters.parameterR = R;
#ifdef USE_L1_DISTANCE
optParameters.parameterR2 = R;
#else
optParameters.parameterR2 = SQR(R);
#endif
optParameters.useUfunctions = TRUE; // TODO: could optimize here:
// maybe sometimes, the old way
// was better.
optParameters.parameterW = PARAMETER_W_DEFAULT;
optParameters.parameterT = nPoints;
optParameters.typeHT = HT_HYBRID_CHAINS;
// Compute the run-time parameters (timings of different parts of the algorithm).
IntT nReps = 10; // # number of repetions
RealT lshPrecomp = 0, uhashOver = 0, distComp = 0;
for(IntT i = 0; i < nReps; i++){
RealT lP, uO, dC;
determineRTCoefficients(optParameters.parameterR,
optParameters.successProbability,
optParameters.useUfunctions,
optParameters.typeHT,
optParameters.dimension,
nPoints,
dataSet,
lP,
uO,
dC);
lshPrecomp += lP;
uhashOver += uO;
distComp += dC;
DPRINTF4("Coefs: lP = %0.9lf\tuO = %0.9lf\tdC = %0.9lf\n", lP, uO, dC);
}
lshPrecomp /= nReps;
uhashOver /= nReps;
distComp /= nReps;
DPRINTF("Coefs (final): lshPrecomp = %0.9lf\n", lshPrecomp);
DPRINTF("Coefs (final): uhashOver = %0.9lf\n", uhashOver);
DPRINTF("Coefs (final): distComp = %0.9lf\n", distComp);
DPRINTF("Remaining memory: %lld\n", memoryUpperBound);
// Try all possible <k>s and choose the one for which the time
// estimate of a query is minimal.
IntT k;
RealT timeLSH, timeUH, timeCycling;
//IntT queryIndex = genRandomInt(0, nPoints);
//PPointT query = dataSet[queryIndex]; // query points = a random points from the data set.
IntT bestK = 0;
RealT bestTime = 0;
for(k = 2; ; k += 2){
DPRINTF("ST. k = %d\n", k);
IntT m = computeMForULSH(k, successProbability);
IntT L = m * (m-1) / 2;
//DPRINTF("Available memory: %lld\n", getAvailableMemory());
if (L * nPoints > memoryUpperBound / 12){
break;
}
timeLSH = m * k * lshPrecomp;
timeUH = L * uhashOver;
//RealT nCollisions = estimateNCollisionsFromDSPoint(nPoints, dimension, dataSet, queryIndex, k, L, R);
// Compute the mean number of collisions for the points from the sample query set.
RealT nCollisions = 0;
for(IntT i = 0; i < nSampleQueries; i++){
nCollisions += estimateNDistinctCollisions(nPoints, dimension, dataSet, sampleQueries[i], TRUE, k, m, R);
}
nCollisions /= nSampleQueries;
timeCycling = nCollisions * distComp;
DPRINTF3("ST.m=%d L=%d \n", m, L);
DPRINTF("ST.Estimated # distinct collisions = %0.6lf\n", (double)nCollisions);
DPRINTF("ST.TimeLSH = %0.6lf\n", timeLSH);
DPRINTF("ST.TimeUH = %0.6lf\n", timeUH);
DPRINTF("ST.TimeCycling = %0.6lf\n", timeCycling);
DPRINTF("ST.Sum = %0.6lf\n", timeLSH + timeUH + timeCycling);
if (bestK == 0 || (timeLSH + timeUH + timeCycling) < bestTime) {
bestK = k;
bestTime = timeLSH + timeUH + timeCycling;
}
ASSERT(k < 100); // otherwise, k reached 100 -- which, from
// experience, should never happen for reasonable
// data set & available memory amount.
}
DPRINTF("STO.Optimal k = %d\n", bestK);
IntT m = computeMForULSH(bestK, successProbability);
IntT L = m * (m-1) / 2;
timeLSH = m * bestK * lshPrecomp;
timeUH = L * uhashOver;
// Compute the mean number of collisions for the points from the sample query set.
RealT nCollisions = 0;
for(IntT i = 0; i < nSampleQueries; i++){
nCollisions += estimateNDistinctCollisions(nPoints, dimension, dataSet, sampleQueries[i], TRUE, k, m, R);
}
nCollisions /= nSampleQueries;
// timeCycling = estimateNCollisionsFromDSPoint(nPoints, dimension, dataSet, queryIndex, bestK, L, R) * distComp;
timeCycling = nCollisions * distComp;
DPRINTF("STO.TimeLSH = %0.6lf\n", timeLSH);
DPRINTF("STO.TimeUH = %0.6lf\n", timeUH);
DPRINTF("STO.TimeCycling = %0.6lf\n", timeCycling);
DPRINTF("STO.Sum = %0.6lf\n", timeLSH + timeUH + timeCycling);
optParameters.parameterK = bestK;
optParameters.parameterM = m;
optParameters.parameterL = L;
return optParameters;
}
