1 | /* |
2 | Copyright (c) 2000, 2012, Oracle and/or its affiliates. |
3 | Copyright (c) 2009, 2016, MariaDB |
4 | |
5 | This program is free software; you can redistribute it and/or modify |
6 | it under the terms of the GNU General Public License as published by |
7 | the Free Software Foundation; version 2 of the License. |
8 | |
9 | This program is distributed in the hope that it will be useful, |
10 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
11 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
12 | GNU General Public License for more details. |
13 | |
14 | You should have received a copy of the GNU General Public License |
15 | along with this program; if not, write to the Free Software |
16 | Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA */ |
17 | |
18 | |
19 | /** |
20 | @file |
21 | |
22 | @brief |
23 | This file defines all time functions |
24 | |
25 | @todo |
26 | Move month and days to language files |
27 | */ |
28 | |
29 | #ifdef USE_PRAGMA_IMPLEMENTATION |
30 | #pragma implementation // gcc: Class implementation |
31 | #endif |
32 | |
33 | #include "mariadb.h" |
34 | #include "sql_priv.h" |
35 | /* |
36 | It is necessary to include set_var.h instead of item.h because there |
37 | are dependencies on include order for set_var.h and item.h. This |
38 | will be resolved later. |
39 | */ |
40 | #include "sql_class.h" // set_var.h: THD |
41 | #include "set_var.h" |
42 | #include "sql_locale.h" // MY_LOCALE my_locale_en_US |
43 | #include "strfunc.h" // check_word |
44 | #include "sql_time.h" // make_truncated_value_warning, |
45 | // get_date_from_daynr, |
46 | // calc_weekday, calc_week, |
47 | // convert_month_to_period, |
48 | // convert_period_to_month, |
49 | // TIME_to_timestamp, |
50 | // calc_time_diff, |
51 | // calc_time_from_sec, |
52 | // get_date_time_format_str |
53 | #include "tztime.h" // struct Time_zone |
54 | #include "sql_class.h" // THD |
55 | #include <m_ctype.h> |
56 | #include <time.h> |
57 | |
58 | /** Day number for Dec 31st, 9999. */ |
59 | #define MAX_DAY_NUMBER 3652424L |
60 | |
61 | /* |
62 | Date formats corresponding to compound %r and %T conversion specifiers |
63 | |
64 | Note: We should init at least first element of "positions" array |
65 | (first member) or hpux11 compiler will die horribly. |
66 | */ |
67 | static DATE_TIME_FORMAT time_ampm_format= {{0}, '\0', 0, |
68 | {(char *)"%I:%i:%S %p" , 11}}; |
69 | static DATE_TIME_FORMAT time_24hrs_format= {{0}, '\0', 0, |
70 | {(char *)"%H:%i:%S" , 8}}; |
71 | |
72 | /** |
73 | Extract datetime value to MYSQL_TIME struct from string value |
74 | according to format string. |
75 | |
76 | @param format date/time format specification |
77 | @param val String to decode |
78 | @param length Length of string |
79 | @param l_time Store result here |
80 | @param cached_timestamp_type It uses to get an appropriate warning |
81 | in the case when the value is truncated. |
82 | @param sub_pattern_end if non-zero then we are parsing string which |
83 | should correspond compound specifier (like %T or |
84 | %r) and this parameter is pointer to place where |
85 | pointer to end of string matching this specifier |
86 | should be stored. |
87 | |
88 | @note |
89 | Possibility to parse strings matching to patterns equivalent to compound |
90 | specifiers is mainly intended for use from inside of this function in |
91 | order to understand %T and %r conversion specifiers, so number of |
92 | conversion specifiers that can be used in such sub-patterns is limited. |
93 | Also most of checks are skipped in this case. |
94 | |
95 | @note |
96 | If one adds new format specifiers to this function he should also |
97 | consider adding them to get_date_time_result_type() function. |
98 | |
99 | @retval |
100 | 0 ok |
101 | @retval |
102 | 1 error |
103 | */ |
104 | |
105 | static bool (DATE_TIME_FORMAT *format, |
106 | const char *val, uint length, MYSQL_TIME *l_time, |
107 | timestamp_type cached_timestamp_type, |
108 | const char **sub_pattern_end, |
109 | const char *date_time_type, |
110 | ulonglong fuzzy_date) |
111 | { |
112 | int weekday= 0, yearday= 0, daypart= 0; |
113 | int week_number= -1; |
114 | int error= 0; |
115 | int strict_week_number_year= -1; |
116 | int frac_part; |
117 | bool usa_time= 0; |
118 | bool UNINIT_VAR(sunday_first_n_first_week_non_iso); |
119 | bool UNINIT_VAR(strict_week_number); |
120 | bool UNINIT_VAR(strict_week_number_year_type); |
121 | const char *val_begin= val; |
122 | const char *val_end= val + length; |
123 | const char *ptr= format->format.str; |
124 | const char *end= ptr + format->format.length; |
125 | CHARSET_INFO *cs= &my_charset_bin; |
126 | DBUG_ENTER("extract_date_time" ); |
127 | |
128 | if (!sub_pattern_end) |
129 | bzero((char*) l_time, sizeof(*l_time)); |
130 | |
131 | l_time->time_type= cached_timestamp_type; |
132 | |
133 | for (; ptr != end && val != val_end; ptr++) |
134 | { |
135 | /* Skip pre-space between each argument */ |
136 | if ((val+= cs->cset->scan(cs, val, val_end, MY_SEQ_SPACES)) >= val_end) |
137 | break; |
138 | |
139 | if (*ptr == '%' && ptr+1 != end) |
140 | { |
141 | int val_len; |
142 | char *tmp; |
143 | |
144 | error= 0; |
145 | |
146 | val_len= (uint) (val_end - val); |
147 | switch (*++ptr) { |
148 | /* Year */ |
149 | case 'Y': |
150 | tmp= (char*) val + MY_MIN(4, val_len); |
151 | l_time->year= (int) my_strtoll10(val, &tmp, &error); |
152 | if ((int) (tmp-val) <= 2) |
153 | l_time->year= year_2000_handling(l_time->year); |
154 | val= tmp; |
155 | break; |
156 | case 'y': |
157 | tmp= (char*) val + MY_MIN(2, val_len); |
158 | l_time->year= (int) my_strtoll10(val, &tmp, &error); |
159 | val= tmp; |
160 | l_time->year= year_2000_handling(l_time->year); |
161 | break; |
162 | |
163 | /* Month */ |
164 | case 'm': |
165 | case 'c': |
166 | tmp= (char*) val + MY_MIN(2, val_len); |
167 | l_time->month= (int) my_strtoll10(val, &tmp, &error); |
168 | val= tmp; |
169 | break; |
170 | case 'M': |
171 | if ((l_time->month= check_word(my_locale_en_US.month_names, |
172 | val, val_end, &val)) <= 0) |
173 | goto err; |
174 | break; |
175 | case 'b': |
176 | if ((l_time->month= check_word(my_locale_en_US.ab_month_names, |
177 | val, val_end, &val)) <= 0) |
178 | goto err; |
179 | break; |
180 | /* Day */ |
181 | case 'd': |
182 | case 'e': |
183 | tmp= (char*) val + MY_MIN(2, val_len); |
184 | l_time->day= (int) my_strtoll10(val, &tmp, &error); |
185 | val= tmp; |
186 | break; |
187 | case 'D': |
188 | tmp= (char*) val + MY_MIN(2, val_len); |
189 | l_time->day= (int) my_strtoll10(val, &tmp, &error); |
190 | /* Skip 'st, 'nd, 'th .. */ |
191 | val= tmp + MY_MIN((int) (val_end-tmp), 2); |
192 | break; |
193 | |
194 | /* Hour */ |
195 | case 'h': |
196 | case 'I': |
197 | case 'l': |
198 | usa_time= 1; |
199 | /* fall through */ |
200 | case 'k': |
201 | case 'H': |
202 | tmp= (char*) val + MY_MIN(2, val_len); |
203 | l_time->hour= (int) my_strtoll10(val, &tmp, &error); |
204 | val= tmp; |
205 | break; |
206 | |
207 | /* Minute */ |
208 | case 'i': |
209 | tmp= (char*) val + MY_MIN(2, val_len); |
210 | l_time->minute= (int) my_strtoll10(val, &tmp, &error); |
211 | val= tmp; |
212 | break; |
213 | |
214 | /* Second */ |
215 | case 's': |
216 | case 'S': |
217 | tmp= (char*) val + MY_MIN(2, val_len); |
218 | l_time->second= (int) my_strtoll10(val, &tmp, &error); |
219 | val= tmp; |
220 | break; |
221 | |
222 | /* Second part */ |
223 | case 'f': |
224 | tmp= (char*) val_end; |
225 | if (tmp - val > 6) |
226 | tmp= (char*) val + 6; |
227 | l_time->second_part= (int) my_strtoll10(val, &tmp, &error); |
228 | frac_part= 6 - (int) (tmp - val); |
229 | if (frac_part > 0) |
230 | l_time->second_part*= (ulong) log_10_int[frac_part]; |
231 | val= tmp; |
232 | break; |
233 | |
234 | /* AM / PM */ |
235 | case 'p': |
236 | if (val_len < 2 || ! usa_time) |
237 | goto err; |
238 | if (!my_strnncoll(&my_charset_latin1, |
239 | (const uchar *) val, 2, |
240 | (const uchar *) "PM" , 2)) |
241 | daypart= 12; |
242 | else if (my_strnncoll(&my_charset_latin1, |
243 | (const uchar *) val, 2, |
244 | (const uchar *) "AM" , 2)) |
245 | goto err; |
246 | val+= 2; |
247 | break; |
248 | |
249 | /* Exotic things */ |
250 | case 'W': |
251 | if ((weekday= check_word(my_locale_en_US.day_names, val, val_end, &val)) <= 0) |
252 | goto err; |
253 | break; |
254 | case 'a': |
255 | if ((weekday= check_word(my_locale_en_US.ab_day_names, val, val_end, &val)) <= 0) |
256 | goto err; |
257 | break; |
258 | case 'w': |
259 | tmp= (char*) val + 1; |
260 | if (unlikely((weekday= (int) my_strtoll10(val, &tmp, &error)) < 0 || |
261 | weekday >= 7)) |
262 | goto err; |
263 | /* We should use the same 1 - 7 scale for %w as for %W */ |
264 | if (!weekday) |
265 | weekday= 7; |
266 | val= tmp; |
267 | break; |
268 | case 'j': |
269 | tmp= (char*) val + MY_MIN(val_len, 3); |
270 | yearday= (int) my_strtoll10(val, &tmp, &error); |
271 | val= tmp; |
272 | break; |
273 | |
274 | /* Week numbers */ |
275 | case 'V': |
276 | case 'U': |
277 | case 'v': |
278 | case 'u': |
279 | sunday_first_n_first_week_non_iso= (*ptr=='U' || *ptr== 'V'); |
280 | strict_week_number= (*ptr=='V' || *ptr=='v'); |
281 | tmp= (char*) val + MY_MIN(val_len, 2); |
282 | if (unlikely((week_number= |
283 | (int) my_strtoll10(val, &tmp, &error)) < 0 || |
284 | (strict_week_number && !week_number) || |
285 | week_number > 53)) |
286 | goto err; |
287 | val= tmp; |
288 | break; |
289 | |
290 | /* Year used with 'strict' %V and %v week numbers */ |
291 | case 'X': |
292 | case 'x': |
293 | strict_week_number_year_type= (*ptr=='X'); |
294 | tmp= (char*) val + MY_MIN(4, val_len); |
295 | strict_week_number_year= (int) my_strtoll10(val, &tmp, &error); |
296 | val= tmp; |
297 | break; |
298 | |
299 | /* Time in AM/PM notation */ |
300 | case 'r': |
301 | /* |
302 | We can't just set error here, as we don't want to generate two |
303 | warnings in case of errors |
304 | */ |
305 | if (extract_date_time(&time_ampm_format, val, |
306 | (uint)(val_end - val), l_time, |
307 | cached_timestamp_type, &val, "time" , fuzzy_date)) |
308 | DBUG_RETURN(1); |
309 | break; |
310 | |
311 | /* Time in 24-hour notation */ |
312 | case 'T': |
313 | if (extract_date_time(&time_24hrs_format, val, |
314 | (uint)(val_end - val), l_time, |
315 | cached_timestamp_type, &val, "time" , fuzzy_date)) |
316 | DBUG_RETURN(1); |
317 | break; |
318 | |
319 | /* Conversion specifiers that match classes of characters */ |
320 | case '.': |
321 | while (my_ispunct(cs, *val) && val != val_end) |
322 | val++; |
323 | break; |
324 | case '@': |
325 | while (my_isalpha(cs, *val) && val != val_end) |
326 | val++; |
327 | break; |
328 | case '#': |
329 | while (my_isdigit(cs, *val) && val != val_end) |
330 | val++; |
331 | break; |
332 | default: |
333 | goto err; |
334 | } |
335 | if (unlikely(error)) // Error from my_strtoll10 |
336 | goto err; |
337 | } |
338 | else if (!my_isspace(cs, *ptr)) |
339 | { |
340 | if (*val != *ptr) |
341 | goto err; |
342 | val++; |
343 | } |
344 | } |
345 | if (usa_time) |
346 | { |
347 | if (l_time->hour > 12 || l_time->hour < 1) |
348 | goto err; |
349 | l_time->hour= l_time->hour%12+daypart; |
350 | } |
351 | |
352 | /* |
353 | If we are recursively called for parsing string matching compound |
354 | specifiers we are already done. |
355 | */ |
356 | if (sub_pattern_end) |
357 | { |
358 | *sub_pattern_end= val; |
359 | DBUG_RETURN(0); |
360 | } |
361 | |
362 | if (yearday > 0) |
363 | { |
364 | uint days; |
365 | days= calc_daynr(l_time->year,1,1) + yearday - 1; |
366 | if (get_date_from_daynr(days,&l_time->year,&l_time->month,&l_time->day)) |
367 | goto err; |
368 | } |
369 | |
370 | if (week_number >= 0 && weekday) |
371 | { |
372 | int days; |
373 | uint weekday_b; |
374 | |
375 | /* |
376 | %V,%v require %X,%x resprectively, |
377 | %U,%u should be used with %Y and not %X or %x |
378 | */ |
379 | if ((strict_week_number && |
380 | (strict_week_number_year < 0 || |
381 | strict_week_number_year_type != |
382 | sunday_first_n_first_week_non_iso)) || |
383 | (!strict_week_number && strict_week_number_year >= 0)) |
384 | goto err; |
385 | |
386 | /* Number of days since year 0 till 1st Jan of this year */ |
387 | days= calc_daynr((strict_week_number ? strict_week_number_year : |
388 | l_time->year), |
389 | 1, 1); |
390 | /* Which day of week is 1st Jan of this year */ |
391 | weekday_b= calc_weekday(days, sunday_first_n_first_week_non_iso); |
392 | |
393 | /* |
394 | Below we are going to sum: |
395 | 1) number of days since year 0 till 1st day of 1st week of this year |
396 | 2) number of days between 1st week and our week |
397 | 3) and position of our day in the week |
398 | */ |
399 | if (sunday_first_n_first_week_non_iso) |
400 | { |
401 | days+= ((weekday_b == 0) ? 0 : 7) - weekday_b + |
402 | (week_number - 1) * 7 + |
403 | weekday % 7; |
404 | } |
405 | else |
406 | { |
407 | days+= ((weekday_b <= 3) ? 0 : 7) - weekday_b + |
408 | (week_number - 1) * 7 + |
409 | (weekday - 1); |
410 | } |
411 | |
412 | if (get_date_from_daynr(days,&l_time->year,&l_time->month,&l_time->day)) |
413 | goto err; |
414 | } |
415 | |
416 | if (l_time->month > 12 || l_time->day > 31 || l_time->hour > 23 || |
417 | l_time->minute > 59 || l_time->second > 59) |
418 | goto err; |
419 | |
420 | int was_cut; |
421 | if (check_date(l_time, fuzzy_date | TIME_INVALID_DATES, &was_cut)) |
422 | goto err; |
423 | |
424 | if (val != val_end) |
425 | { |
426 | do |
427 | { |
428 | if (!my_isspace(&my_charset_latin1,*val)) |
429 | { |
430 | make_truncated_value_warning(current_thd, |
431 | Sql_condition::WARN_LEVEL_WARN, |
432 | val_begin, length, |
433 | cached_timestamp_type, NullS); |
434 | break; |
435 | } |
436 | } while (++val != val_end); |
437 | } |
438 | DBUG_RETURN(0); |
439 | |
440 | err: |
441 | { |
442 | THD *thd= current_thd; |
443 | char buff[128]; |
444 | strmake(buff, val_begin, MY_MIN(length, sizeof(buff)-1)); |
445 | push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, |
446 | ER_WRONG_VALUE_FOR_TYPE, |
447 | ER_THD(thd, ER_WRONG_VALUE_FOR_TYPE), |
448 | date_time_type, buff, "str_to_date" ); |
449 | } |
450 | DBUG_RETURN(1); |
451 | } |
452 | |
453 | |
454 | /** |
455 | Create a formated date/time value in a string. |
456 | */ |
457 | |
458 | static bool make_date_time(DATE_TIME_FORMAT *format, MYSQL_TIME *l_time, |
459 | timestamp_type type, const MY_LOCALE *locale, |
460 | String *str) |
461 | { |
462 | char intbuff[15]; |
463 | uint hours_i; |
464 | uint weekday; |
465 | ulong length; |
466 | const char *ptr, *end; |
467 | |
468 | str->length(0); |
469 | |
470 | if (l_time->neg) |
471 | str->append('-'); |
472 | |
473 | end= (ptr= format->format.str) + format->format.length; |
474 | for (; ptr != end ; ptr++) |
475 | { |
476 | if (*ptr != '%' || ptr+1 == end) |
477 | str->append(*ptr); |
478 | else |
479 | { |
480 | switch (*++ptr) { |
481 | case 'M': |
482 | if (type == MYSQL_TIMESTAMP_TIME || !l_time->month) |
483 | return 1; |
484 | str->append(locale->month_names->type_names[l_time->month-1], |
485 | (uint) strlen(locale->month_names->type_names[l_time->month-1]), |
486 | system_charset_info); |
487 | break; |
488 | case 'b': |
489 | if (type == MYSQL_TIMESTAMP_TIME || !l_time->month) |
490 | return 1; |
491 | str->append(locale->ab_month_names->type_names[l_time->month-1], |
492 | (uint) strlen(locale->ab_month_names->type_names[l_time->month-1]), |
493 | system_charset_info); |
494 | break; |
495 | case 'W': |
496 | if (type == MYSQL_TIMESTAMP_TIME || !(l_time->month || l_time->year)) |
497 | return 1; |
498 | weekday= calc_weekday(calc_daynr(l_time->year,l_time->month, |
499 | l_time->day),0); |
500 | str->append(locale->day_names->type_names[weekday], |
501 | (uint) strlen(locale->day_names->type_names[weekday]), |
502 | system_charset_info); |
503 | break; |
504 | case 'a': |
505 | if (type == MYSQL_TIMESTAMP_TIME || !(l_time->month || l_time->year)) |
506 | return 1; |
507 | weekday=calc_weekday(calc_daynr(l_time->year,l_time->month, |
508 | l_time->day),0); |
509 | str->append(locale->ab_day_names->type_names[weekday], |
510 | (uint) strlen(locale->ab_day_names->type_names[weekday]), |
511 | system_charset_info); |
512 | break; |
513 | case 'D': |
514 | if (type == MYSQL_TIMESTAMP_TIME) |
515 | return 1; |
516 | length= (uint) (int10_to_str(l_time->day, intbuff, 10) - intbuff); |
517 | str->append_with_prefill(intbuff, length, 1, '0'); |
518 | if (l_time->day >= 10 && l_time->day <= 19) |
519 | str->append(STRING_WITH_LEN("th" )); |
520 | else |
521 | { |
522 | switch (l_time->day %10) { |
523 | case 1: |
524 | str->append(STRING_WITH_LEN("st" )); |
525 | break; |
526 | case 2: |
527 | str->append(STRING_WITH_LEN("nd" )); |
528 | break; |
529 | case 3: |
530 | str->append(STRING_WITH_LEN("rd" )); |
531 | break; |
532 | default: |
533 | str->append(STRING_WITH_LEN("th" )); |
534 | break; |
535 | } |
536 | } |
537 | break; |
538 | case 'Y': |
539 | if (type == MYSQL_TIMESTAMP_TIME) |
540 | return 1; |
541 | length= (uint) (int10_to_str(l_time->year, intbuff, 10) - intbuff); |
542 | str->append_with_prefill(intbuff, length, 4, '0'); |
543 | break; |
544 | case 'y': |
545 | if (type == MYSQL_TIMESTAMP_TIME) |
546 | return 1; |
547 | length= (uint) (int10_to_str(l_time->year%100, intbuff, 10) - intbuff); |
548 | str->append_with_prefill(intbuff, length, 2, '0'); |
549 | break; |
550 | case 'm': |
551 | if (type == MYSQL_TIMESTAMP_TIME) |
552 | return 1; |
553 | length= (uint) (int10_to_str(l_time->month, intbuff, 10) - intbuff); |
554 | str->append_with_prefill(intbuff, length, 2, '0'); |
555 | break; |
556 | case 'c': |
557 | if (type == MYSQL_TIMESTAMP_TIME) |
558 | return 1; |
559 | length= (uint) (int10_to_str(l_time->month, intbuff, 10) - intbuff); |
560 | str->append_with_prefill(intbuff, length, 1, '0'); |
561 | break; |
562 | case 'd': |
563 | if (type == MYSQL_TIMESTAMP_TIME) |
564 | return 1; |
565 | length= (uint) (int10_to_str(l_time->day, intbuff, 10) - intbuff); |
566 | str->append_with_prefill(intbuff, length, 2, '0'); |
567 | break; |
568 | case 'e': |
569 | if (type == MYSQL_TIMESTAMP_TIME) |
570 | return 1; |
571 | length= (uint) (int10_to_str(l_time->day, intbuff, 10) - intbuff); |
572 | str->append_with_prefill(intbuff, length, 1, '0'); |
573 | break; |
574 | case 'f': |
575 | length= (uint) (int10_to_str(l_time->second_part, intbuff, 10) - intbuff); |
576 | str->append_with_prefill(intbuff, length, 6, '0'); |
577 | break; |
578 | case 'H': |
579 | length= (uint) (int10_to_str(l_time->hour, intbuff, 10) - intbuff); |
580 | str->append_with_prefill(intbuff, length, 2, '0'); |
581 | break; |
582 | case 'h': |
583 | case 'I': |
584 | hours_i= (l_time->hour%24 + 11)%12+1; |
585 | length= (uint) (int10_to_str(hours_i, intbuff, 10) - intbuff); |
586 | str->append_with_prefill(intbuff, length, 2, '0'); |
587 | break; |
588 | case 'i': /* minutes */ |
589 | length= (uint) (int10_to_str(l_time->minute, intbuff, 10) - intbuff); |
590 | str->append_with_prefill(intbuff, length, 2, '0'); |
591 | break; |
592 | case 'j': |
593 | if (type == MYSQL_TIMESTAMP_TIME) |
594 | return 1; |
595 | length= (uint) (int10_to_str(calc_daynr(l_time->year,l_time->month, |
596 | l_time->day) - |
597 | calc_daynr(l_time->year,1,1) + 1, intbuff, 10) - intbuff); |
598 | str->append_with_prefill(intbuff, length, 3, '0'); |
599 | break; |
600 | case 'k': |
601 | length= (uint) (int10_to_str(l_time->hour, intbuff, 10) - intbuff); |
602 | str->append_with_prefill(intbuff, length, 1, '0'); |
603 | break; |
604 | case 'l': |
605 | hours_i= (l_time->hour%24 + 11)%12+1; |
606 | length= (uint) (int10_to_str(hours_i, intbuff, 10) - intbuff); |
607 | str->append_with_prefill(intbuff, length, 1, '0'); |
608 | break; |
609 | case 'p': |
610 | hours_i= l_time->hour%24; |
611 | str->append(hours_i < 12 ? "AM" : "PM" ,2); |
612 | break; |
613 | case 'r': |
614 | length= sprintf(intbuff, ((l_time->hour % 24) < 12) ? |
615 | "%02d:%02d:%02d AM" : "%02d:%02d:%02d PM" , |
616 | (l_time->hour+11)%12+1, |
617 | l_time->minute, |
618 | l_time->second); |
619 | str->append(intbuff, length); |
620 | break; |
621 | case 'S': |
622 | case 's': |
623 | length= (uint) (int10_to_str(l_time->second, intbuff, 10) - intbuff); |
624 | str->append_with_prefill(intbuff, length, 2, '0'); |
625 | break; |
626 | case 'T': |
627 | length= sprintf(intbuff, "%02d:%02d:%02d" , |
628 | l_time->hour, l_time->minute, l_time->second); |
629 | str->append(intbuff, length); |
630 | break; |
631 | case 'U': |
632 | case 'u': |
633 | { |
634 | uint year; |
635 | if (type == MYSQL_TIMESTAMP_TIME) |
636 | return 1; |
637 | length= (uint) (int10_to_str(calc_week(l_time, |
638 | (*ptr) == 'U' ? |
639 | WEEK_FIRST_WEEKDAY : WEEK_MONDAY_FIRST, |
640 | &year), |
641 | intbuff, 10) - intbuff); |
642 | str->append_with_prefill(intbuff, length, 2, '0'); |
643 | } |
644 | break; |
645 | case 'v': |
646 | case 'V': |
647 | { |
648 | uint year; |
649 | if (type == MYSQL_TIMESTAMP_TIME) |
650 | return 1; |
651 | length= (uint) (int10_to_str(calc_week(l_time, |
652 | ((*ptr) == 'V' ? |
653 | (WEEK_YEAR | WEEK_FIRST_WEEKDAY) : |
654 | (WEEK_YEAR | WEEK_MONDAY_FIRST)), |
655 | &year), |
656 | intbuff, 10) - intbuff); |
657 | str->append_with_prefill(intbuff, length, 2, '0'); |
658 | } |
659 | break; |
660 | case 'x': |
661 | case 'X': |
662 | { |
663 | uint year; |
664 | if (type == MYSQL_TIMESTAMP_TIME) |
665 | return 1; |
666 | (void) calc_week(l_time, |
667 | ((*ptr) == 'X' ? |
668 | WEEK_YEAR | WEEK_FIRST_WEEKDAY : |
669 | WEEK_YEAR | WEEK_MONDAY_FIRST), |
670 | &year); |
671 | length= (uint) (int10_to_str(year, intbuff, 10) - intbuff); |
672 | str->append_with_prefill(intbuff, length, 4, '0'); |
673 | } |
674 | break; |
675 | case 'w': |
676 | if (type == MYSQL_TIMESTAMP_TIME || !(l_time->month || l_time->year)) |
677 | return 1; |
678 | weekday=calc_weekday(calc_daynr(l_time->year,l_time->month, |
679 | l_time->day),1); |
680 | length= (uint) (int10_to_str(weekday, intbuff, 10) - intbuff); |
681 | str->append_with_prefill(intbuff, length, 1, '0'); |
682 | break; |
683 | |
684 | default: |
685 | str->append(*ptr); |
686 | break; |
687 | } |
688 | } |
689 | } |
690 | return 0; |
691 | } |
692 | |
693 | |
694 | /** |
695 | @details |
696 | Get a array of positive numbers from a string object. |
697 | Each number is separated by 1 non digit character |
698 | Return error if there is too many numbers. |
699 | If there is too few numbers, assume that the numbers are left out |
700 | from the high end. This allows one to give: |
701 | DAY_TO_SECOND as "D MM:HH:SS", "MM:HH:SS" "HH:SS" or as seconds. |
702 | |
703 | @param length: length of str |
704 | @param cs: charset of str |
705 | @param values: array of results |
706 | @param count: count of elements in result array |
707 | @param transform_msec: if value is true we suppose |
708 | that the last part of string value is microseconds |
709 | and we should transform value to six digit value. |
710 | For example, '1.1' -> '1.100000' |
711 | */ |
712 | |
713 | static bool get_interval_info(const char *str, size_t length,CHARSET_INFO *cs, size_t count, ulonglong *values, |
714 | bool transform_msec) |
715 | { |
716 | const char *end=str+length; |
717 | uint i; |
718 | size_t field_length= 0; |
719 | |
720 | while (str != end && !my_isdigit(cs,*str)) |
721 | str++; |
722 | |
723 | for (i=0 ; i < count ; i++) |
724 | { |
725 | longlong value; |
726 | const char *start= str; |
727 | for (value= 0; str != end && my_isdigit(cs, *str); str++) |
728 | value= value*10 + *str - '0'; |
729 | if ((field_length= (size_t)(str - start)) >= 20) |
730 | return true; |
731 | values[i]= value; |
732 | while (str != end && !my_isdigit(cs,*str)) |
733 | str++; |
734 | if (str == end && i != count-1) |
735 | { |
736 | i++; |
737 | /* Change values[0...i-1] -> values[0...count-1] */ |
738 | bmove_upp((uchar*) (values+count), (uchar*) (values+i), |
739 | sizeof(*values)*i); |
740 | bzero((uchar*) values, sizeof(*values)*(count-i)); |
741 | break; |
742 | } |
743 | } |
744 | |
745 | if (transform_msec && field_length > 0) |
746 | { |
747 | if (field_length < 6) |
748 | values[count - 1] *= log_10_int[6 - field_length]; |
749 | else if (field_length > 6) |
750 | values[count - 1] /= log_10_int[field_length - 6]; |
751 | } |
752 | |
753 | return (str != end); |
754 | } |
755 | |
756 | |
757 | longlong Item_func_period_add::val_int() |
758 | { |
759 | DBUG_ASSERT(fixed == 1); |
760 | ulong period=(ulong) args[0]->val_int(); |
761 | int months=(int) args[1]->val_int(); |
762 | |
763 | if ((null_value=args[0]->null_value || args[1]->null_value) || |
764 | period == 0L) |
765 | return 0; /* purecov: inspected */ |
766 | return (longlong) |
767 | convert_month_to_period((uint) ((int) convert_period_to_month(period)+ |
768 | months)); |
769 | } |
770 | |
771 | |
772 | longlong Item_func_period_diff::val_int() |
773 | { |
774 | DBUG_ASSERT(fixed == 1); |
775 | ulong period1=(ulong) args[0]->val_int(); |
776 | ulong period2=(ulong) args[1]->val_int(); |
777 | |
778 | if ((null_value=args[0]->null_value || args[1]->null_value)) |
779 | return 0; /* purecov: inspected */ |
780 | return (longlong) ((long) convert_period_to_month(period1)- |
781 | (long) convert_period_to_month(period2)); |
782 | } |
783 | |
784 | |
785 | |
786 | longlong Item_func_to_days::val_int() |
787 | { |
788 | DBUG_ASSERT(fixed == 1); |
789 | MYSQL_TIME ltime; |
790 | if (get_arg0_date(<ime, TIME_NO_ZERO_DATE | TIME_NO_ZERO_IN_DATE)) |
791 | return 0; |
792 | return (longlong) calc_daynr(ltime.year,ltime.month,ltime.day); |
793 | } |
794 | |
795 | |
796 | longlong Item_func_to_seconds::val_int_endpoint(bool left_endp, |
797 | bool *incl_endp) |
798 | { |
799 | DBUG_ASSERT(fixed == 1); |
800 | MYSQL_TIME ltime; |
801 | longlong seconds; |
802 | longlong days; |
803 | int dummy; /* unused */ |
804 | if (get_arg0_date(<ime, TIME_FUZZY_DATES)) |
805 | { |
806 | /* got NULL, leave the incl_endp intact */ |
807 | return LONGLONG_MIN; |
808 | } |
809 | seconds= ltime.hour * 3600L + ltime.minute * 60 + ltime.second; |
810 | seconds= ltime.neg ? -seconds : seconds; |
811 | days= (longlong) calc_daynr(ltime.year, ltime.month, ltime.day); |
812 | seconds+= days * 24L * 3600L; |
813 | /* Set to NULL if invalid date, but keep the value */ |
814 | null_value= check_date(<ime, |
815 | (ltime.year || ltime.month || ltime.day), |
816 | (TIME_NO_ZERO_IN_DATE | TIME_NO_ZERO_DATE), |
817 | &dummy); |
818 | /* |
819 | Even if the evaluation return NULL, seconds is useful for pruning |
820 | */ |
821 | return seconds; |
822 | } |
823 | |
824 | longlong Item_func_to_seconds::val_int() |
825 | { |
826 | DBUG_ASSERT(fixed == 1); |
827 | MYSQL_TIME ltime; |
828 | longlong seconds; |
829 | longlong days; |
830 | if (get_arg0_date(<ime, TIME_NO_ZERO_DATE | TIME_NO_ZERO_IN_DATE)) |
831 | return 0; |
832 | seconds= ltime.hour * 3600L + ltime.minute * 60 + ltime.second; |
833 | seconds=ltime.neg ? -seconds : seconds; |
834 | days= (longlong) calc_daynr(ltime.year, ltime.month, ltime.day); |
835 | return seconds + days * 24L * 3600L; |
836 | } |
837 | |
838 | /* |
839 | Get information about this Item tree monotonicity |
840 | |
841 | SYNOPSIS |
842 | Item_func_to_days::get_monotonicity_info() |
843 | |
844 | DESCRIPTION |
845 | Get information about monotonicity of the function represented by this item |
846 | tree. |
847 | |
848 | RETURN |
849 | See enum_monotonicity_info. |
850 | */ |
851 | |
852 | enum_monotonicity_info Item_func_to_days::get_monotonicity_info() const |
853 | { |
854 | if (args[0]->type() == Item::FIELD_ITEM) |
855 | { |
856 | if (args[0]->field_type() == MYSQL_TYPE_DATE) |
857 | return MONOTONIC_STRICT_INCREASING_NOT_NULL; |
858 | if (args[0]->field_type() == MYSQL_TYPE_DATETIME) |
859 | return MONOTONIC_INCREASING_NOT_NULL; |
860 | } |
861 | return NON_MONOTONIC; |
862 | } |
863 | |
864 | enum_monotonicity_info Item_func_to_seconds::get_monotonicity_info() const |
865 | { |
866 | if (args[0]->type() == Item::FIELD_ITEM) |
867 | { |
868 | if (args[0]->field_type() == MYSQL_TYPE_DATE || |
869 | args[0]->field_type() == MYSQL_TYPE_DATETIME) |
870 | return MONOTONIC_STRICT_INCREASING_NOT_NULL; |
871 | } |
872 | return NON_MONOTONIC; |
873 | } |
874 | |
875 | |
876 | longlong Item_func_to_days::val_int_endpoint(bool left_endp, bool *incl_endp) |
877 | { |
878 | DBUG_ASSERT(fixed == 1); |
879 | MYSQL_TIME ltime; |
880 | longlong res; |
881 | int dummy; /* unused */ |
882 | if (get_arg0_date(<ime, 0)) |
883 | { |
884 | /* got NULL, leave the incl_endp intact */ |
885 | return LONGLONG_MIN; |
886 | } |
887 | res=(longlong) calc_daynr(ltime.year,ltime.month,ltime.day); |
888 | /* Set to NULL if invalid date, but keep the value */ |
889 | null_value= check_date(<ime, |
890 | (TIME_NO_ZERO_IN_DATE | TIME_NO_ZERO_DATE), |
891 | &dummy); |
892 | if (null_value) |
893 | { |
894 | /* |
895 | Even if the evaluation return NULL, the calc_daynr is useful for pruning |
896 | */ |
897 | if (args[0]->field_type() != MYSQL_TYPE_DATE) |
898 | *incl_endp= TRUE; |
899 | return res; |
900 | } |
901 | |
902 | if (args[0]->field_type() == MYSQL_TYPE_DATE) |
903 | { |
904 | // TO_DAYS() is strictly monotonic for dates, leave incl_endp intact |
905 | return res; |
906 | } |
907 | |
908 | /* |
909 | Handle the special but practically useful case of datetime values that |
910 | point to day bound ("strictly less" comparison stays intact): |
911 | |
912 | col < '2007-09-15 00:00:00' -> TO_DAYS(col) < TO_DAYS('2007-09-15') |
913 | col > '2007-09-15 23:59:59' -> TO_DAYS(col) > TO_DAYS('2007-09-15') |
914 | |
915 | which is different from the general case ("strictly less" changes to |
916 | "less or equal"): |
917 | |
918 | col < '2007-09-15 12:34:56' -> TO_DAYS(col) <= TO_DAYS('2007-09-15') |
919 | */ |
920 | if ((!left_endp && !(ltime.hour || ltime.minute || ltime.second || |
921 | ltime.second_part)) || |
922 | (left_endp && ltime.hour == 23 && ltime.minute == 59 && |
923 | ltime.second == 59)) |
924 | /* do nothing */ |
925 | ; |
926 | else |
927 | *incl_endp= TRUE; |
928 | return res; |
929 | } |
930 | |
931 | |
932 | longlong Item_func_dayofyear::val_int() |
933 | { |
934 | DBUG_ASSERT(fixed == 1); |
935 | MYSQL_TIME ltime; |
936 | if (get_arg0_date(<ime, TIME_NO_ZERO_IN_DATE | TIME_NO_ZERO_DATE)) |
937 | return 0; |
938 | return (longlong) calc_daynr(ltime.year,ltime.month,ltime.day) - |
939 | calc_daynr(ltime.year,1,1) + 1; |
940 | } |
941 | |
942 | longlong Item_func_dayofmonth::val_int() |
943 | { |
944 | DBUG_ASSERT(fixed == 1); |
945 | MYSQL_TIME ltime; |
946 | return get_arg0_date(<ime, 0) ? 0 : (longlong) ltime.day; |
947 | } |
948 | |
949 | longlong Item_func_month::val_int() |
950 | { |
951 | DBUG_ASSERT(fixed == 1); |
952 | MYSQL_TIME ltime; |
953 | return get_arg0_date(<ime, 0) ? 0 : (longlong) ltime.month; |
954 | } |
955 | |
956 | |
957 | void Item_func_monthname::fix_length_and_dec() |
958 | { |
959 | THD* thd= current_thd; |
960 | CHARSET_INFO *cs= thd->variables.collation_connection; |
961 | locale= thd->variables.lc_time_names; |
962 | collation.set(cs, DERIVATION_COERCIBLE, locale->repertoire()); |
963 | decimals=0; |
964 | max_length= locale->max_month_name_length * collation.collation->mbmaxlen; |
965 | maybe_null=1; |
966 | } |
967 | |
968 | |
969 | String* Item_func_monthname::val_str(String* str) |
970 | { |
971 | DBUG_ASSERT(fixed == 1); |
972 | const char *month_name; |
973 | uint err; |
974 | MYSQL_TIME ltime; |
975 | |
976 | if ((null_value= (get_arg0_date(<ime, 0) || !ltime.month))) |
977 | return (String *) 0; |
978 | |
979 | month_name= locale->month_names->type_names[ltime.month - 1]; |
980 | str->copy(month_name, (uint) strlen(month_name), &my_charset_utf8_bin, |
981 | collation.collation, &err); |
982 | return str; |
983 | } |
984 | |
985 | |
986 | /** |
987 | Returns the quarter of the year. |
988 | */ |
989 | |
990 | longlong Item_func_quarter::val_int() |
991 | { |
992 | DBUG_ASSERT(fixed == 1); |
993 | MYSQL_TIME ltime; |
994 | if (get_arg0_date(<ime, 0)) |
995 | return 0; |
996 | return (longlong) ((ltime.month+2)/3); |
997 | } |
998 | |
999 | longlong Item_func_hour::val_int() |
1000 | { |
1001 | DBUG_ASSERT(fixed == 1); |
1002 | Time tm(args[0], Time::Options_for_cast()); |
1003 | return (null_value= !tm.is_valid_time()) ? 0 : tm.get_mysql_time()->hour; |
1004 | } |
1005 | |
1006 | longlong Item_func_minute::val_int() |
1007 | { |
1008 | DBUG_ASSERT(fixed == 1); |
1009 | Time tm(args[0], Time::Options_for_cast()); |
1010 | return (null_value= !tm.is_valid_time()) ? 0 : tm.get_mysql_time()->minute; |
1011 | } |
1012 | |
1013 | /** |
1014 | Returns the second in time_exp in the range of 0 - 59. |
1015 | */ |
1016 | longlong Item_func_second::val_int() |
1017 | { |
1018 | DBUG_ASSERT(fixed == 1); |
1019 | Time tm(args[0], Time::Options_for_cast()); |
1020 | return (null_value= !tm.is_valid_time()) ? 0 : tm.get_mysql_time()->second; |
1021 | } |
1022 | |
1023 | |
1024 | uint week_mode(uint mode) |
1025 | { |
1026 | uint week_format= (mode & 7); |
1027 | if (!(week_format & WEEK_MONDAY_FIRST)) |
1028 | week_format^= WEEK_FIRST_WEEKDAY; |
1029 | return week_format; |
1030 | } |
1031 | |
1032 | /** |
1033 | @verbatim |
1034 | The bits in week_format(for calc_week() function) has the following meaning: |
1035 | WEEK_MONDAY_FIRST (0) If not set Sunday is first day of week |
1036 | If set Monday is first day of week |
1037 | WEEK_YEAR (1) If not set Week is in range 0-53 |
1038 | |
1039 | Week 0 is returned for the the last week of the previous year (for |
1040 | a date at start of january) In this case one can get 53 for the |
1041 | first week of next year. This flag ensures that the week is |
1042 | relevant for the given year. Note that this flag is only |
1043 | releveant if WEEK_JANUARY is not set. |
1044 | |
1045 | If set Week is in range 1-53. |
1046 | |
1047 | In this case one may get week 53 for a date in January (when |
1048 | the week is that last week of previous year) and week 1 for a |
1049 | date in December. |
1050 | |
1051 | WEEK_FIRST_WEEKDAY (2) If not set Weeks are numbered according |
1052 | to ISO 8601:1988 |
1053 | If set The week that contains the first |
1054 | 'first-day-of-week' is week 1. |
1055 | |
1056 | ISO 8601:1988 means that if the week containing January 1 has |
1057 | four or more days in the new year, then it is week 1; |
1058 | Otherwise it is the last week of the previous year, and the |
1059 | next week is week 1. |
1060 | @endverbatim |
1061 | */ |
1062 | |
1063 | longlong Item_func_week::val_int() |
1064 | { |
1065 | DBUG_ASSERT(fixed == 1); |
1066 | uint year, week_format; |
1067 | MYSQL_TIME ltime; |
1068 | if (get_arg0_date(<ime, TIME_NO_ZERO_DATE | TIME_NO_ZERO_IN_DATE)) |
1069 | return 0; |
1070 | if (arg_count > 1) |
1071 | week_format= (uint)args[1]->val_int(); |
1072 | else |
1073 | week_format= current_thd->variables.default_week_format; |
1074 | return (longlong) calc_week(<ime, week_mode(week_format), &year); |
1075 | } |
1076 | |
1077 | |
1078 | longlong Item_func_yearweek::val_int() |
1079 | { |
1080 | DBUG_ASSERT(fixed == 1); |
1081 | uint year,week; |
1082 | MYSQL_TIME ltime; |
1083 | if (get_arg0_date(<ime, TIME_NO_ZERO_DATE | TIME_NO_ZERO_IN_DATE)) |
1084 | return 0; |
1085 | week= calc_week(<ime, |
1086 | (week_mode((uint) args[1]->val_int()) | WEEK_YEAR), |
1087 | &year); |
1088 | return week+year*100; |
1089 | } |
1090 | |
1091 | |
1092 | longlong Item_func_weekday::val_int() |
1093 | { |
1094 | DBUG_ASSERT(fixed == 1); |
1095 | MYSQL_TIME ltime; |
1096 | |
1097 | if (get_arg0_date(<ime, TIME_NO_ZERO_DATE | TIME_NO_ZERO_IN_DATE)) |
1098 | return 0; |
1099 | |
1100 | return (longlong) calc_weekday(calc_daynr(ltime.year, ltime.month, |
1101 | ltime.day), |
1102 | odbc_type) + MY_TEST(odbc_type); |
1103 | } |
1104 | |
1105 | void Item_func_dayname::fix_length_and_dec() |
1106 | { |
1107 | THD* thd= current_thd; |
1108 | CHARSET_INFO *cs= thd->variables.collation_connection; |
1109 | locale= thd->variables.lc_time_names; |
1110 | collation.set(cs, DERIVATION_COERCIBLE, locale->repertoire()); |
1111 | decimals=0; |
1112 | max_length= locale->max_day_name_length * collation.collation->mbmaxlen; |
1113 | maybe_null=1; |
1114 | } |
1115 | |
1116 | |
1117 | String* Item_func_dayname::val_str(String* str) |
1118 | { |
1119 | DBUG_ASSERT(fixed == 1); |
1120 | uint weekday=(uint) val_int(); // Always Item_func_weekday() |
1121 | const char *day_name; |
1122 | uint err; |
1123 | |
1124 | if (null_value) |
1125 | return (String*) 0; |
1126 | |
1127 | day_name= locale->day_names->type_names[weekday]; |
1128 | str->copy(day_name, (uint) strlen(day_name), &my_charset_utf8_bin, |
1129 | collation.collation, &err); |
1130 | return str; |
1131 | } |
1132 | |
1133 | |
1134 | longlong Item_func_year::val_int() |
1135 | { |
1136 | DBUG_ASSERT(fixed == 1); |
1137 | MYSQL_TIME ltime; |
1138 | return get_arg0_date(<ime, 0) ? 0 : (longlong) ltime.year; |
1139 | } |
1140 | |
1141 | |
1142 | /* |
1143 | Get information about this Item tree monotonicity |
1144 | |
1145 | SYNOPSIS |
1146 | Item_func_year::get_monotonicity_info() |
1147 | |
1148 | DESCRIPTION |
1149 | Get information about monotonicity of the function represented by this item |
1150 | tree. |
1151 | |
1152 | RETURN |
1153 | See enum_monotonicity_info. |
1154 | */ |
1155 | |
1156 | enum_monotonicity_info Item_func_year::get_monotonicity_info() const |
1157 | { |
1158 | if (args[0]->type() == Item::FIELD_ITEM && |
1159 | (args[0]->field_type() == MYSQL_TYPE_DATE || |
1160 | args[0]->field_type() == MYSQL_TYPE_DATETIME)) |
1161 | return MONOTONIC_INCREASING; |
1162 | return NON_MONOTONIC; |
1163 | } |
1164 | |
1165 | |
1166 | longlong Item_func_year::val_int_endpoint(bool left_endp, bool *incl_endp) |
1167 | { |
1168 | DBUG_ASSERT(fixed == 1); |
1169 | MYSQL_TIME ltime; |
1170 | if (get_arg0_date(<ime, 0)) |
1171 | { |
1172 | /* got NULL, leave the incl_endp intact */ |
1173 | return LONGLONG_MIN; |
1174 | } |
1175 | |
1176 | /* |
1177 | Handle the special but practically useful case of datetime values that |
1178 | point to year bound ("strictly less" comparison stays intact) : |
1179 | |
1180 | col < '2007-01-01 00:00:00' -> YEAR(col) < 2007 |
1181 | |
1182 | which is different from the general case ("strictly less" changes to |
1183 | "less or equal"): |
1184 | |
1185 | col < '2007-09-15 23:00:00' -> YEAR(col) <= 2007 |
1186 | */ |
1187 | if (!left_endp && ltime.day == 1 && ltime.month == 1 && |
1188 | !(ltime.hour || ltime.minute || ltime.second || ltime.second_part)) |
1189 | ; /* do nothing */ |
1190 | else |
1191 | *incl_endp= TRUE; |
1192 | return ltime.year; |
1193 | } |
1194 | |
1195 | |
1196 | bool Item_func_unix_timestamp::get_timestamp_value(my_time_t *seconds, |
1197 | ulong *second_part) |
1198 | { |
1199 | DBUG_ASSERT(fixed == 1); |
1200 | if (args[0]->type() == FIELD_ITEM) |
1201 | { // Optimize timestamp field |
1202 | Field *field=((Item_field*) args[0])->field; |
1203 | if (field->type() == MYSQL_TYPE_TIMESTAMP) |
1204 | { |
1205 | if ((null_value= field->is_null())) |
1206 | return 1; |
1207 | *seconds= ((Field_timestamp*)field)->get_timestamp(second_part); |
1208 | return 0; |
1209 | } |
1210 | } |
1211 | |
1212 | MYSQL_TIME ltime; |
1213 | if (get_arg0_date(<ime, TIME_NO_ZERO_IN_DATE)) |
1214 | return 1; |
1215 | |
1216 | uint error_code; |
1217 | *seconds= TIME_to_timestamp(current_thd, <ime, &error_code); |
1218 | *second_part= ltime.second_part; |
1219 | return (null_value= (error_code == ER_WARN_DATA_OUT_OF_RANGE)); |
1220 | } |
1221 | |
1222 | |
1223 | longlong Item_func_unix_timestamp::int_op() |
1224 | { |
1225 | if (arg_count == 0) |
1226 | return (longlong) current_thd->query_start(); |
1227 | |
1228 | ulong second_part; |
1229 | my_time_t seconds; |
1230 | if (get_timestamp_value(&seconds, &second_part)) |
1231 | return 0; |
1232 | |
1233 | return seconds; |
1234 | } |
1235 | |
1236 | |
1237 | my_decimal *Item_func_unix_timestamp::decimal_op(my_decimal* buf) |
1238 | { |
1239 | ulong second_part; |
1240 | my_time_t seconds; |
1241 | if (get_timestamp_value(&seconds, &second_part)) |
1242 | return 0; |
1243 | |
1244 | return seconds2my_decimal(seconds < 0, seconds < 0 ? -seconds : seconds, |
1245 | second_part, buf); |
1246 | } |
1247 | |
1248 | |
1249 | enum_monotonicity_info Item_func_unix_timestamp::get_monotonicity_info() const |
1250 | { |
1251 | if (args[0]->type() == Item::FIELD_ITEM && |
1252 | (args[0]->field_type() == MYSQL_TYPE_TIMESTAMP)) |
1253 | return MONOTONIC_INCREASING; |
1254 | return NON_MONOTONIC; |
1255 | } |
1256 | |
1257 | |
1258 | longlong Item_func_unix_timestamp::val_int_endpoint(bool left_endp, bool *incl_endp) |
1259 | { |
1260 | DBUG_ASSERT(fixed == 1); |
1261 | DBUG_ASSERT(arg_count == 1 && |
1262 | args[0]->type() == Item::FIELD_ITEM && |
1263 | args[0]->field_type() == MYSQL_TYPE_TIMESTAMP); |
1264 | Field_timestamp *field=(Field_timestamp *)(((Item_field*)args[0])->field); |
1265 | /* Leave the incl_endp intact */ |
1266 | ulong unused; |
1267 | my_time_t ts= field->get_timestamp(&unused); |
1268 | null_value= field->is_null(); |
1269 | return ts; |
1270 | } |
1271 | |
1272 | |
1273 | longlong Item_func_time_to_sec::int_op() |
1274 | { |
1275 | DBUG_ASSERT(fixed == 1); |
1276 | Time tm(args[0], Time::Options_for_cast()); |
1277 | return ((null_value= !tm.is_valid_time())) ? 0 : tm.to_seconds(); |
1278 | } |
1279 | |
1280 | |
1281 | my_decimal *Item_func_time_to_sec::decimal_op(my_decimal* buf) |
1282 | { |
1283 | DBUG_ASSERT(fixed == 1); |
1284 | Time tm(args[0], Time::Options_for_cast()); |
1285 | if ((null_value= !tm.is_valid_time())) |
1286 | return 0; |
1287 | const MYSQL_TIME *ltime= tm.get_mysql_time(); |
1288 | longlong seconds= tm.to_seconds_abs(); |
1289 | return seconds2my_decimal(ltime->neg, seconds, ltime->second_part, buf); |
1290 | } |
1291 | |
1292 | |
1293 | /** |
1294 | Convert a string to a interval value. |
1295 | |
1296 | To make code easy, allow interval objects without separators. |
1297 | */ |
1298 | |
1299 | bool get_interval_value(Item *args,interval_type int_type, INTERVAL *interval) |
1300 | { |
1301 | ulonglong array[5]; |
1302 | longlong UNINIT_VAR(value); |
1303 | const char *UNINIT_VAR(str); |
1304 | size_t UNINIT_VAR(length); |
1305 | CHARSET_INFO *UNINIT_VAR(cs); |
1306 | char buf[100]; |
1307 | String str_value(buf, sizeof(buf), &my_charset_bin); |
1308 | |
1309 | bzero((char*) interval,sizeof(*interval)); |
1310 | if (int_type == INTERVAL_SECOND && args->decimals) |
1311 | { |
1312 | my_decimal decimal_value, *val; |
1313 | ulonglong second; |
1314 | ulong second_part; |
1315 | if (!(val= args->val_decimal(&decimal_value))) |
1316 | return true; |
1317 | interval->neg= my_decimal2seconds(val, &second, &second_part); |
1318 | if (second == LONGLONG_MAX) |
1319 | { |
1320 | THD *thd= current_thd; |
1321 | ErrConvDecimal err(val); |
1322 | push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, |
1323 | ER_TRUNCATED_WRONG_VALUE, |
1324 | ER_THD(thd, ER_TRUNCATED_WRONG_VALUE), "DECIMAL" , |
1325 | err.ptr()); |
1326 | return true; |
1327 | } |
1328 | |
1329 | interval->second= second; |
1330 | interval->second_part= second_part; |
1331 | return false; |
1332 | } |
1333 | else if ((int) int_type <= INTERVAL_MICROSECOND) |
1334 | { |
1335 | value= args->val_int(); |
1336 | if (args->null_value) |
1337 | return 1; |
1338 | if (value < 0) |
1339 | { |
1340 | interval->neg=1; |
1341 | value= -value; |
1342 | } |
1343 | } |
1344 | else |
1345 | { |
1346 | String *res; |
1347 | if (!(res= args->val_str_ascii(&str_value))) |
1348 | return (1); |
1349 | |
1350 | /* record negative intervalls in interval->neg */ |
1351 | str=res->ptr(); |
1352 | cs= res->charset(); |
1353 | const char *end=str+res->length(); |
1354 | while (str != end && my_isspace(cs,*str)) |
1355 | str++; |
1356 | if (str != end && *str == '-') |
1357 | { |
1358 | interval->neg=1; |
1359 | str++; |
1360 | } |
1361 | length= (size_t) (end-str); // Set up pointers to new str |
1362 | } |
1363 | |
1364 | switch (int_type) { |
1365 | case INTERVAL_YEAR: |
1366 | interval->year= (ulong) value; |
1367 | break; |
1368 | case INTERVAL_QUARTER: |
1369 | interval->month= (ulong)(value*3); |
1370 | break; |
1371 | case INTERVAL_MONTH: |
1372 | interval->month= (ulong) value; |
1373 | break; |
1374 | case INTERVAL_WEEK: |
1375 | interval->day= (ulong)(value*7); |
1376 | break; |
1377 | case INTERVAL_DAY: |
1378 | interval->day= (ulong) value; |
1379 | break; |
1380 | case INTERVAL_HOUR: |
1381 | interval->hour= (ulong) value; |
1382 | break; |
1383 | case INTERVAL_MICROSECOND: |
1384 | interval->second_part=value; |
1385 | break; |
1386 | case INTERVAL_MINUTE: |
1387 | interval->minute=value; |
1388 | break; |
1389 | case INTERVAL_SECOND: |
1390 | interval->second=value; |
1391 | break; |
1392 | case INTERVAL_YEAR_MONTH: // Allow YEAR-MONTH YYYYYMM |
1393 | if (get_interval_info(str,length,cs,2,array,0)) |
1394 | return (1); |
1395 | interval->year= (ulong) array[0]; |
1396 | interval->month= (ulong) array[1]; |
1397 | break; |
1398 | case INTERVAL_DAY_HOUR: |
1399 | if (get_interval_info(str,length,cs,2,array,0)) |
1400 | return (1); |
1401 | interval->day= (ulong) array[0]; |
1402 | interval->hour= (ulong) array[1]; |
1403 | break; |
1404 | case INTERVAL_DAY_MICROSECOND: |
1405 | if (get_interval_info(str,length,cs,5,array,1)) |
1406 | return (1); |
1407 | interval->day= (ulong) array[0]; |
1408 | interval->hour= (ulong) array[1]; |
1409 | interval->minute= array[2]; |
1410 | interval->second= array[3]; |
1411 | interval->second_part= array[4]; |
1412 | break; |
1413 | case INTERVAL_DAY_MINUTE: |
1414 | if (get_interval_info(str,length,cs,3,array,0)) |
1415 | return (1); |
1416 | interval->day= (ulong) array[0]; |
1417 | interval->hour= (ulong) array[1]; |
1418 | interval->minute= array[2]; |
1419 | break; |
1420 | case INTERVAL_DAY_SECOND: |
1421 | if (get_interval_info(str,length,cs,4,array,0)) |
1422 | return (1); |
1423 | interval->day= (ulong) array[0]; |
1424 | interval->hour= (ulong) array[1]; |
1425 | interval->minute= array[2]; |
1426 | interval->second= array[3]; |
1427 | break; |
1428 | case INTERVAL_HOUR_MICROSECOND: |
1429 | if (get_interval_info(str,length,cs,4,array,1)) |
1430 | return (1); |
1431 | interval->hour= (ulong) array[0]; |
1432 | interval->minute= array[1]; |
1433 | interval->second= array[2]; |
1434 | interval->second_part= array[3]; |
1435 | break; |
1436 | case INTERVAL_HOUR_MINUTE: |
1437 | if (get_interval_info(str,length,cs,2,array,0)) |
1438 | return (1); |
1439 | interval->hour= (ulong) array[0]; |
1440 | interval->minute= array[1]; |
1441 | break; |
1442 | case INTERVAL_HOUR_SECOND: |
1443 | if (get_interval_info(str,length,cs,3,array,0)) |
1444 | return (1); |
1445 | interval->hour= (ulong) array[0]; |
1446 | interval->minute= array[1]; |
1447 | interval->second= array[2]; |
1448 | break; |
1449 | case INTERVAL_MINUTE_MICROSECOND: |
1450 | if (get_interval_info(str,length,cs,3,array,1)) |
1451 | return (1); |
1452 | interval->minute= array[0]; |
1453 | interval->second= array[1]; |
1454 | interval->second_part= array[2]; |
1455 | break; |
1456 | case INTERVAL_MINUTE_SECOND: |
1457 | if (get_interval_info(str,length,cs,2,array,0)) |
1458 | return (1); |
1459 | interval->minute= array[0]; |
1460 | interval->second= array[1]; |
1461 | break; |
1462 | case INTERVAL_SECOND_MICROSECOND: |
1463 | if (get_interval_info(str,length,cs,2,array,1)) |
1464 | return (1); |
1465 | interval->second= array[0]; |
1466 | interval->second_part= array[1]; |
1467 | break; |
1468 | case INTERVAL_LAST: /* purecov: begin deadcode */ |
1469 | DBUG_ASSERT(0); |
1470 | break; /* purecov: end */ |
1471 | } |
1472 | return 0; |
1473 | } |
1474 | |
1475 | |
1476 | String *Item_temporal_func::val_str(String *str) |
1477 | { |
1478 | DBUG_ASSERT(fixed == 1); |
1479 | return val_string_from_date(str); |
1480 | } |
1481 | |
1482 | |
1483 | bool Item_temporal_hybrid_func::fix_temporal_type(MYSQL_TIME *ltime) |
1484 | { |
1485 | if (ltime->time_type < 0) /* MYSQL_TIMESTAMP_NONE, MYSQL_TIMESTAMP_ERROR */ |
1486 | return false; |
1487 | |
1488 | if (ltime->time_type != MYSQL_TIMESTAMP_TIME) |
1489 | goto date_or_datetime_value; |
1490 | |
1491 | /* Convert TIME to DATE or DATETIME */ |
1492 | switch (field_type()) |
1493 | { |
1494 | case MYSQL_TYPE_DATE: |
1495 | case MYSQL_TYPE_DATETIME: |
1496 | case MYSQL_TYPE_TIMESTAMP: |
1497 | { |
1498 | MYSQL_TIME tmp; |
1499 | if (time_to_datetime_with_warn(current_thd, ltime, &tmp, 0)) |
1500 | return (null_value= true); |
1501 | *ltime= tmp; |
1502 | if (field_type() == MYSQL_TYPE_DATE) |
1503 | datetime_to_date(ltime); |
1504 | return false; |
1505 | } |
1506 | case MYSQL_TYPE_TIME: |
1507 | case MYSQL_TYPE_STRING: /* DATE_ADD, ADDTIME can return VARCHAR */ |
1508 | return false; |
1509 | default: |
1510 | DBUG_ASSERT(0); |
1511 | return (null_value= true); |
1512 | } |
1513 | |
1514 | date_or_datetime_value: |
1515 | /* Convert DATE or DATETIME to TIME, DATE, or DATETIME */ |
1516 | switch (field_type()) |
1517 | { |
1518 | case MYSQL_TYPE_TIME: |
1519 | datetime_to_time(ltime); |
1520 | return false; |
1521 | case MYSQL_TYPE_DATETIME: |
1522 | case MYSQL_TYPE_TIMESTAMP: |
1523 | date_to_datetime(ltime); |
1524 | return false; |
1525 | case MYSQL_TYPE_DATE: |
1526 | datetime_to_date(ltime); |
1527 | return false; |
1528 | case MYSQL_TYPE_STRING: /* DATE_ADD, ADDTIME can return VARCHAR */ |
1529 | return false; |
1530 | default: |
1531 | DBUG_ASSERT(0); |
1532 | return (null_value= true); |
1533 | } |
1534 | return false; |
1535 | } |
1536 | |
1537 | |
1538 | String *Item_temporal_hybrid_func::val_str_ascii(String *str) |
1539 | { |
1540 | DBUG_ASSERT(fixed == 1); |
1541 | MYSQL_TIME ltime; |
1542 | |
1543 | if (get_date(<ime, 0) || fix_temporal_type(<ime) || |
1544 | (null_value= my_TIME_to_str(<ime, str, decimals))) |
1545 | return (String *) 0; |
1546 | |
1547 | /* Check that the returned timestamp type matches to the function type */ |
1548 | DBUG_ASSERT(field_type() == MYSQL_TYPE_STRING || |
1549 | ltime.time_type == MYSQL_TIMESTAMP_NONE || |
1550 | ltime.time_type == mysql_timestamp_type()); |
1551 | return str; |
1552 | } |
1553 | |
1554 | |
1555 | bool Item_func_from_days::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
1556 | { |
1557 | longlong value=args[0]->val_int(); |
1558 | if ((null_value= (args[0]->null_value || |
1559 | ((fuzzy_date & TIME_NO_ZERO_DATE) && value == 0)))) |
1560 | return true; |
1561 | bzero(ltime, sizeof(MYSQL_TIME)); |
1562 | if (get_date_from_daynr((long) value, <ime->year, <ime->month, |
1563 | <ime->day)) |
1564 | return 0; |
1565 | |
1566 | ltime->time_type= MYSQL_TIMESTAMP_DATE; |
1567 | return 0; |
1568 | } |
1569 | |
1570 | |
1571 | /** |
1572 | Converts current time in my_time_t to MYSQL_TIME represenatation for local |
1573 | time zone. Defines time zone (local) used for whole CURDATE function. |
1574 | */ |
1575 | void Item_func_curdate_local::store_now_in_TIME(THD *thd, MYSQL_TIME *now_time) |
1576 | { |
1577 | thd->variables.time_zone->gmt_sec_to_TIME(now_time, thd->query_start()); |
1578 | thd->time_zone_used= 1; |
1579 | } |
1580 | |
1581 | |
1582 | /** |
1583 | Converts current time in my_time_t to MYSQL_TIME represenatation for UTC |
1584 | time zone. Defines time zone (UTC) used for whole UTC_DATE function. |
1585 | */ |
1586 | void Item_func_curdate_utc::store_now_in_TIME(THD *thd, MYSQL_TIME *now_time) |
1587 | { |
1588 | my_tz_UTC->gmt_sec_to_TIME(now_time, thd->query_start()); |
1589 | /* |
1590 | We are not flagging this query as using time zone, since it uses fixed |
1591 | UTC-SYSTEM time-zone. |
1592 | */ |
1593 | } |
1594 | |
1595 | |
1596 | bool Item_func_curdate::get_date(MYSQL_TIME *res, |
1597 | ulonglong fuzzy_date __attribute__((unused))) |
1598 | { |
1599 | THD *thd= current_thd; |
1600 | query_id_t query_id= thd->query_id; |
1601 | /* Cache value for this query */ |
1602 | if (last_query_id != query_id) |
1603 | { |
1604 | last_query_id= query_id; |
1605 | store_now_in_TIME(thd, <ime); |
1606 | /* We don't need to set second_part and neg because they already 0 */ |
1607 | ltime.hour= ltime.minute= ltime.second= 0; |
1608 | ltime.time_type= MYSQL_TIMESTAMP_DATE; |
1609 | } |
1610 | *res=ltime; |
1611 | return 0; |
1612 | } |
1613 | |
1614 | |
1615 | bool Item_func_curtime::fix_fields(THD *thd, Item **items) |
1616 | { |
1617 | if (decimals > TIME_SECOND_PART_DIGITS) |
1618 | { |
1619 | my_error(ER_TOO_BIG_PRECISION, MYF(0), static_cast<ulonglong>(decimals), |
1620 | func_name(), TIME_SECOND_PART_DIGITS); |
1621 | return 1; |
1622 | } |
1623 | return Item_timefunc::fix_fields(thd, items); |
1624 | } |
1625 | |
1626 | bool Item_func_curtime::get_date(MYSQL_TIME *res, |
1627 | ulonglong fuzzy_date __attribute__((unused))) |
1628 | { |
1629 | THD *thd= current_thd; |
1630 | query_id_t query_id= thd->query_id; |
1631 | /* Cache value for this query */ |
1632 | if (last_query_id != query_id) |
1633 | { |
1634 | last_query_id= query_id; |
1635 | store_now_in_TIME(thd, <ime); |
1636 | } |
1637 | *res= ltime; |
1638 | return 0; |
1639 | } |
1640 | |
1641 | void Item_func_curtime::print(String *str, enum_query_type query_type) |
1642 | { |
1643 | str->append(func_name()); |
1644 | str->append('('); |
1645 | if (decimals) |
1646 | str->append_ulonglong(decimals); |
1647 | str->append(')'); |
1648 | } |
1649 | |
1650 | static void set_sec_part(ulong sec_part, MYSQL_TIME *ltime, Item *item) |
1651 | { |
1652 | DBUG_ASSERT(item->decimals == AUTO_SEC_PART_DIGITS || |
1653 | item->decimals <= TIME_SECOND_PART_DIGITS); |
1654 | if (item->decimals) |
1655 | { |
1656 | ltime->second_part= sec_part; |
1657 | if (item->decimals < TIME_SECOND_PART_DIGITS) |
1658 | my_time_trunc(ltime, item->decimals); |
1659 | } |
1660 | } |
1661 | |
1662 | /** |
1663 | Converts current time in my_time_t to MYSQL_TIME represenatation for local |
1664 | time zone. Defines time zone (local) used for whole CURTIME function. |
1665 | */ |
1666 | void Item_func_curtime_local::store_now_in_TIME(THD *thd, MYSQL_TIME *now_time) |
1667 | { |
1668 | thd->variables.time_zone->gmt_sec_to_TIME(now_time, thd->query_start()); |
1669 | now_time->year= now_time->month= now_time->day= 0; |
1670 | now_time->time_type= MYSQL_TIMESTAMP_TIME; |
1671 | set_sec_part(thd->query_start_sec_part(), now_time, this); |
1672 | thd->time_zone_used= 1; |
1673 | } |
1674 | |
1675 | |
1676 | /** |
1677 | Converts current time in my_time_t to MYSQL_TIME represenatation for UTC |
1678 | time zone. Defines time zone (UTC) used for whole UTC_TIME function. |
1679 | */ |
1680 | void Item_func_curtime_utc::store_now_in_TIME(THD *thd, MYSQL_TIME *now_time) |
1681 | { |
1682 | my_tz_UTC->gmt_sec_to_TIME(now_time, thd->query_start()); |
1683 | now_time->year= now_time->month= now_time->day= 0; |
1684 | now_time->time_type= MYSQL_TIMESTAMP_TIME; |
1685 | set_sec_part(thd->query_start_sec_part(), now_time, this); |
1686 | /* |
1687 | We are not flagging this query as using time zone, since it uses fixed |
1688 | UTC-SYSTEM time-zone. |
1689 | */ |
1690 | } |
1691 | |
1692 | bool Item_func_now::fix_fields(THD *thd, Item **items) |
1693 | { |
1694 | if (decimals > TIME_SECOND_PART_DIGITS) |
1695 | { |
1696 | my_error(ER_TOO_BIG_PRECISION, MYF(0), static_cast<ulonglong>(decimals), |
1697 | func_name(), TIME_SECOND_PART_DIGITS); |
1698 | return 1; |
1699 | } |
1700 | return Item_temporal_func::fix_fields(thd, items); |
1701 | } |
1702 | |
1703 | void Item_func_now::print(String *str, enum_query_type query_type) |
1704 | { |
1705 | str->append(func_name()); |
1706 | str->append('('); |
1707 | if (decimals) |
1708 | str->append_ulonglong(decimals); |
1709 | str->append(')'); |
1710 | } |
1711 | |
1712 | |
1713 | int Item_func_now_local::save_in_field(Field *field, bool no_conversions) |
1714 | { |
1715 | if (field->type() == MYSQL_TYPE_TIMESTAMP) |
1716 | { |
1717 | THD *thd= field->get_thd(); |
1718 | my_time_t ts= thd->query_start(); |
1719 | uint dec= MY_MIN(decimals, field->decimals()); |
1720 | ulong sec_part= dec ? thd->query_start_sec_part() : 0; |
1721 | sec_part-= my_time_fraction_remainder(sec_part, dec); |
1722 | field->set_notnull(); |
1723 | ((Field_timestamp*)field)->store_TIME(ts, sec_part); |
1724 | return 0; |
1725 | } |
1726 | else |
1727 | return Item_temporal_func::save_in_field(field, no_conversions); |
1728 | } |
1729 | |
1730 | |
1731 | /** |
1732 | Converts current time in my_time_t to MYSQL_TIME represenatation for local |
1733 | time zone. Defines time zone (local) used for whole NOW function. |
1734 | */ |
1735 | void Item_func_now_local::store_now_in_TIME(THD *thd, MYSQL_TIME *now_time) |
1736 | { |
1737 | thd->variables.time_zone->gmt_sec_to_TIME(now_time, thd->query_start()); |
1738 | set_sec_part(thd->query_start_sec_part(), now_time, this); |
1739 | thd->time_zone_used= 1; |
1740 | } |
1741 | |
1742 | |
1743 | /** |
1744 | Converts current time in my_time_t to MYSQL_TIME represenatation for UTC |
1745 | time zone. Defines time zone (UTC) used for whole UTC_TIMESTAMP function. |
1746 | */ |
1747 | void Item_func_now_utc::store_now_in_TIME(THD *thd, MYSQL_TIME *now_time) |
1748 | { |
1749 | my_tz_UTC->gmt_sec_to_TIME(now_time, thd->query_start()); |
1750 | set_sec_part(thd->query_start_sec_part(), now_time, this); |
1751 | /* |
1752 | We are not flagging this query as using time zone, since it uses fixed |
1753 | UTC-SYSTEM time-zone. |
1754 | */ |
1755 | } |
1756 | |
1757 | |
1758 | bool Item_func_now::get_date(MYSQL_TIME *res, |
1759 | ulonglong fuzzy_date __attribute__((unused))) |
1760 | { |
1761 | THD *thd= current_thd; |
1762 | query_id_t query_id= thd->query_id; |
1763 | /* Cache value for this query */ |
1764 | if (last_query_id != query_id) |
1765 | { |
1766 | last_query_id= query_id; |
1767 | store_now_in_TIME(thd, <ime); |
1768 | } |
1769 | *res= ltime; |
1770 | return 0; |
1771 | } |
1772 | |
1773 | |
1774 | /** |
1775 | Converts current time in my_time_t to MYSQL_TIME represenatation for local |
1776 | time zone. Defines time zone (local) used for whole SYSDATE function. |
1777 | */ |
1778 | void Item_func_sysdate_local::store_now_in_TIME(THD *thd, MYSQL_TIME *now_time) |
1779 | { |
1780 | my_hrtime_t now= my_hrtime(); |
1781 | thd->variables.time_zone->gmt_sec_to_TIME(now_time, hrtime_to_my_time(now)); |
1782 | set_sec_part(hrtime_sec_part(now), now_time, this); |
1783 | thd->time_zone_used= 1; |
1784 | } |
1785 | |
1786 | |
1787 | bool Item_func_sysdate_local::get_date(MYSQL_TIME *res, |
1788 | ulonglong fuzzy_date __attribute__((unused))) |
1789 | { |
1790 | store_now_in_TIME(current_thd, res); |
1791 | return 0; |
1792 | } |
1793 | |
1794 | bool Item_func_sec_to_time::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
1795 | { |
1796 | DBUG_ASSERT(fixed == 1); |
1797 | bool sign; |
1798 | ulonglong sec; |
1799 | ulong sec_part; |
1800 | |
1801 | bzero((char *)ltime, sizeof(*ltime)); |
1802 | ltime->time_type= MYSQL_TIMESTAMP_TIME; |
1803 | |
1804 | sign= args[0]->get_seconds(&sec, &sec_part); |
1805 | |
1806 | if ((null_value= args[0]->null_value)) |
1807 | return 1; |
1808 | |
1809 | ltime->neg= sign; |
1810 | if (sec > TIME_MAX_VALUE_SECONDS) |
1811 | goto overflow; |
1812 | |
1813 | DBUG_ASSERT(sec_part <= TIME_MAX_SECOND_PART); |
1814 | |
1815 | ltime->hour= (uint) (sec/3600); |
1816 | ltime->minute= (uint) (sec % 3600) /60; |
1817 | ltime->second= (uint) sec % 60; |
1818 | ltime->second_part= sec_part; |
1819 | |
1820 | return 0; |
1821 | |
1822 | overflow: |
1823 | /* use check_time_range() to set ltime to the max value depending on dec */ |
1824 | int unused; |
1825 | char buf[100]; |
1826 | String tmp(buf, sizeof(buf), &my_charset_bin), *err= args[0]->val_str(&tmp); |
1827 | |
1828 | ltime->hour= TIME_MAX_HOUR+1; |
1829 | check_time_range(ltime, decimals, &unused); |
1830 | if (!err) |
1831 | { |
1832 | ErrConvInteger err2(sec, unsigned_flag); |
1833 | make_truncated_value_warning(current_thd, Sql_condition::WARN_LEVEL_WARN, |
1834 | &err2, MYSQL_TIMESTAMP_TIME, NullS); |
1835 | } |
1836 | else |
1837 | { |
1838 | ErrConvString err2(err); |
1839 | make_truncated_value_warning(current_thd, Sql_condition::WARN_LEVEL_WARN, |
1840 | &err2, MYSQL_TIMESTAMP_TIME, NullS); |
1841 | } |
1842 | return 0; |
1843 | } |
1844 | |
1845 | void Item_func_date_format::fix_length_and_dec() |
1846 | { |
1847 | THD* thd= current_thd; |
1848 | if (!is_time_format) |
1849 | { |
1850 | if (arg_count < 3) |
1851 | locale= thd->variables.lc_time_names; |
1852 | else |
1853 | if (args[2]->basic_const_item()) |
1854 | locale= args[2]->locale_from_val_str(); |
1855 | } |
1856 | |
1857 | /* |
1858 | Must use this_item() in case it's a local SP variable |
1859 | (for ->max_length and ->str_value) |
1860 | */ |
1861 | Item *arg1= args[1]->this_item(); |
1862 | |
1863 | decimals=0; |
1864 | CHARSET_INFO *cs= thd->variables.collation_connection; |
1865 | uint32 repertoire= arg1->collation.repertoire; |
1866 | if (!thd->variables.lc_time_names->is_ascii) |
1867 | repertoire|= MY_REPERTOIRE_EXTENDED; |
1868 | collation.set(cs, arg1->collation.derivation, repertoire); |
1869 | StringBuffer<STRING_BUFFER_USUAL_SIZE> buffer; |
1870 | String *str; |
1871 | if (args[1]->basic_const_item() && (str= args[1]->val_str(&buffer))) |
1872 | { // Optimize the normal case |
1873 | fixed_length=1; |
1874 | max_length= format_length(str) * collation.collation->mbmaxlen; |
1875 | } |
1876 | else |
1877 | { |
1878 | fixed_length=0; |
1879 | max_length=MY_MIN(arg1->max_length, MAX_BLOB_WIDTH) * 10 * |
1880 | collation.collation->mbmaxlen; |
1881 | set_if_smaller(max_length,MAX_BLOB_WIDTH); |
1882 | } |
1883 | maybe_null=1; // If wrong date |
1884 | } |
1885 | |
1886 | |
1887 | bool Item_func_date_format::eq(const Item *item, bool binary_cmp) const |
1888 | { |
1889 | Item_func_date_format *item_func; |
1890 | |
1891 | if (item->type() != FUNC_ITEM) |
1892 | return 0; |
1893 | if (func_name() != ((Item_func*) item)->func_name()) |
1894 | return 0; |
1895 | if (this == item) |
1896 | return 1; |
1897 | item_func= (Item_func_date_format*) item; |
1898 | if (arg_count != item_func->arg_count) |
1899 | return 0; |
1900 | if (!args[0]->eq(item_func->args[0], binary_cmp)) |
1901 | return 0; |
1902 | /* |
1903 | We must compare format string case sensitive. |
1904 | This needed because format modifiers with different case, |
1905 | for example %m and %M, have different meaning. |
1906 | */ |
1907 | if (!args[1]->eq(item_func->args[1], 1)) |
1908 | return 0; |
1909 | if (arg_count > 2 && !args[2]->eq(item_func->args[2], 1)) |
1910 | return 0; |
1911 | return 1; |
1912 | } |
1913 | |
1914 | |
1915 | |
1916 | uint Item_func_date_format::format_length(const String *format) |
1917 | { |
1918 | uint size=0; |
1919 | const char *ptr=format->ptr(); |
1920 | const char *end=ptr+format->length(); |
1921 | |
1922 | for (; ptr != end ; ptr++) |
1923 | { |
1924 | if (*ptr != '%' || ptr == end-1) |
1925 | size++; |
1926 | else |
1927 | { |
1928 | switch(*++ptr) { |
1929 | case 'M': /* month, textual */ |
1930 | case 'W': /* day (of the week), textual */ |
1931 | size += 64; /* large for UTF8 locale data */ |
1932 | break; |
1933 | case 'D': /* day (of the month), numeric plus english suffix */ |
1934 | case 'Y': /* year, numeric, 4 digits */ |
1935 | case 'x': /* Year, used with 'v' */ |
1936 | case 'X': /* Year, used with 'v, where week starts with Monday' */ |
1937 | size += 4; |
1938 | break; |
1939 | case 'a': /* locale's abbreviated weekday name (Sun..Sat) */ |
1940 | case 'b': /* locale's abbreviated month name (Jan.Dec) */ |
1941 | size += 32; /* large for UTF8 locale data */ |
1942 | break; |
1943 | case 'j': /* day of year (001..366) */ |
1944 | size += 3; |
1945 | break; |
1946 | case 'U': /* week (00..52) */ |
1947 | case 'u': /* week (00..52), where week starts with Monday */ |
1948 | case 'V': /* week 1..53 used with 'x' */ |
1949 | case 'v': /* week 1..53 used with 'x', where week starts with Monday */ |
1950 | case 'y': /* year, numeric, 2 digits */ |
1951 | case 'm': /* month, numeric */ |
1952 | case 'd': /* day (of the month), numeric */ |
1953 | case 'h': /* hour (01..12) */ |
1954 | case 'I': /* --||-- */ |
1955 | case 'i': /* minutes, numeric */ |
1956 | case 'l': /* hour ( 1..12) */ |
1957 | case 'p': /* locale's AM or PM */ |
1958 | case 'S': /* second (00..61) */ |
1959 | case 's': /* seconds, numeric */ |
1960 | case 'c': /* month (0..12) */ |
1961 | case 'e': /* day (0..31) */ |
1962 | size += 2; |
1963 | break; |
1964 | case 'k': /* hour ( 0..23) */ |
1965 | case 'H': /* hour (00..23; value > 23 OK, padding always 2-digit) */ |
1966 | size += 7; /* docs allow > 23, range depends on sizeof(unsigned int) */ |
1967 | break; |
1968 | case 'r': /* time, 12-hour (hh:mm:ss [AP]M) */ |
1969 | size += 11; |
1970 | break; |
1971 | case 'T': /* time, 24-hour (hh:mm:ss) */ |
1972 | size += 8; |
1973 | break; |
1974 | case 'f': /* microseconds */ |
1975 | size += 6; |
1976 | break; |
1977 | case 'w': /* day (of the week), numeric */ |
1978 | case '%': |
1979 | default: |
1980 | size++; |
1981 | break; |
1982 | } |
1983 | } |
1984 | } |
1985 | return size; |
1986 | } |
1987 | |
1988 | |
1989 | String *Item_func_date_format::val_str(String *str) |
1990 | { |
1991 | String *format; |
1992 | MYSQL_TIME l_time; |
1993 | uint size; |
1994 | const MY_LOCALE *lc= 0; |
1995 | DBUG_ASSERT(fixed == 1); |
1996 | |
1997 | if ((null_value= args[0]->get_date(&l_time, is_time_format ? TIME_TIME_ONLY : 0))) |
1998 | return 0; |
1999 | |
2000 | if (!(format = args[1]->val_str(str)) || !format->length()) |
2001 | goto null_date; |
2002 | |
2003 | if (!is_time_format && !(lc= locale) && !(lc= args[2]->locale_from_val_str())) |
2004 | goto null_date; // invalid locale |
2005 | |
2006 | if (fixed_length) |
2007 | size=max_length; |
2008 | else |
2009 | size=format_length(format); |
2010 | |
2011 | if (size < MAX_DATE_STRING_REP_LENGTH) |
2012 | size= MAX_DATE_STRING_REP_LENGTH; |
2013 | |
2014 | if (format == str) |
2015 | str= &value; // Save result here |
2016 | if (str->alloc(size)) |
2017 | goto null_date; |
2018 | |
2019 | DATE_TIME_FORMAT date_time_format; |
2020 | date_time_format.format.str= (char*) format->ptr(); |
2021 | date_time_format.format.length= format->length(); |
2022 | |
2023 | /* Create the result string */ |
2024 | str->set_charset(collation.collation); |
2025 | if (!make_date_time(&date_time_format, &l_time, |
2026 | is_time_format ? MYSQL_TIMESTAMP_TIME : |
2027 | MYSQL_TIMESTAMP_DATE, |
2028 | lc, str)) |
2029 | return str; |
2030 | |
2031 | null_date: |
2032 | null_value=1; |
2033 | return 0; |
2034 | } |
2035 | |
2036 | |
2037 | void Item_func_from_unixtime::fix_length_and_dec() |
2038 | { |
2039 | THD *thd= current_thd; |
2040 | thd->time_zone_used= 1; |
2041 | tz= thd->variables.time_zone; |
2042 | fix_attributes_datetime_not_fixed_dec(args[0]->decimals); |
2043 | maybe_null= true; |
2044 | } |
2045 | |
2046 | |
2047 | bool Item_func_from_unixtime::get_date(MYSQL_TIME *ltime, |
2048 | ulonglong fuzzy_date __attribute__((unused))) |
2049 | { |
2050 | bool sign; |
2051 | ulonglong sec; |
2052 | ulong sec_part; |
2053 | |
2054 | bzero((char *)ltime, sizeof(*ltime)); |
2055 | ltime->time_type= MYSQL_TIMESTAMP_TIME; |
2056 | |
2057 | sign= args[0]->get_seconds(&sec, &sec_part); |
2058 | |
2059 | if (args[0]->null_value || sign || sec > TIMESTAMP_MAX_VALUE) |
2060 | return (null_value= 1); |
2061 | |
2062 | tz->gmt_sec_to_TIME(ltime, (my_time_t)sec); |
2063 | |
2064 | ltime->second_part= sec_part; |
2065 | |
2066 | return (null_value= 0); |
2067 | } |
2068 | |
2069 | |
2070 | bool Item_func_convert_tz::get_date(MYSQL_TIME *ltime, |
2071 | ulonglong fuzzy_date __attribute__((unused))) |
2072 | { |
2073 | my_time_t my_time_tmp; |
2074 | String str; |
2075 | THD *thd= current_thd; |
2076 | |
2077 | if (!from_tz_cached) |
2078 | { |
2079 | from_tz= my_tz_find(thd, args[1]->val_str_ascii(&str)); |
2080 | from_tz_cached= args[1]->const_item(); |
2081 | } |
2082 | |
2083 | if (!to_tz_cached) |
2084 | { |
2085 | to_tz= my_tz_find(thd, args[2]->val_str_ascii(&str)); |
2086 | to_tz_cached= args[2]->const_item(); |
2087 | } |
2088 | |
2089 | if (from_tz==0 || to_tz==0 || |
2090 | get_arg0_date(ltime, TIME_NO_ZERO_DATE | TIME_NO_ZERO_IN_DATE)) |
2091 | return (null_value= 1); |
2092 | |
2093 | { |
2094 | uint not_used; |
2095 | my_time_tmp= from_tz->TIME_to_gmt_sec(ltime, ¬_used); |
2096 | ulong sec_part= ltime->second_part; |
2097 | /* my_time_tmp is guranteed to be in the allowed range */ |
2098 | if (my_time_tmp) |
2099 | to_tz->gmt_sec_to_TIME(ltime, my_time_tmp); |
2100 | /* we rely on the fact that no timezone conversion can change sec_part */ |
2101 | ltime->second_part= sec_part; |
2102 | } |
2103 | |
2104 | return (null_value= 0); |
2105 | } |
2106 | |
2107 | |
2108 | void Item_func_convert_tz::cleanup() |
2109 | { |
2110 | from_tz_cached= to_tz_cached= 0; |
2111 | Item_temporal_func::cleanup(); |
2112 | } |
2113 | |
2114 | |
2115 | void Item_date_add_interval::fix_length_and_dec() |
2116 | { |
2117 | enum_field_types arg0_field_type; |
2118 | |
2119 | if (!args[0]->type_handler()->is_traditional_type()) |
2120 | { |
2121 | my_error(ER_ILLEGAL_PARAMETER_DATA_TYPES2_FOR_OPERATION, MYF(0), |
2122 | args[0]->type_handler()->name().ptr(), |
2123 | "interval" , func_name()); |
2124 | return; |
2125 | } |
2126 | /* |
2127 | The field type for the result of an Item_datefunc is defined as |
2128 | follows: |
2129 | |
2130 | - If first arg is a MYSQL_TYPE_DATETIME result is MYSQL_TYPE_DATETIME |
2131 | - If first arg is a MYSQL_TYPE_DATE and the interval type uses hours, |
2132 | minutes or seconds then type is MYSQL_TYPE_DATETIME |
2133 | otherwise it's MYSQL_TYPE_DATE |
2134 | - if first arg is a MYSQL_TYPE_TIME and the interval type isn't using |
2135 | anything larger than days, then the result is MYSQL_TYPE_TIME, |
2136 | otherwise - MYSQL_TYPE_DATETIME. |
2137 | - Otherwise the result is MYSQL_TYPE_STRING |
2138 | (This is because you can't know if the string contains a DATE, |
2139 | MYSQL_TIME or DATETIME argument) |
2140 | */ |
2141 | arg0_field_type= args[0]->field_type(); |
2142 | uint interval_dec= 0; |
2143 | if (int_type == INTERVAL_MICROSECOND || |
2144 | (int_type >= INTERVAL_DAY_MICROSECOND && |
2145 | int_type <= INTERVAL_SECOND_MICROSECOND)) |
2146 | interval_dec= TIME_SECOND_PART_DIGITS; |
2147 | else if (int_type == INTERVAL_SECOND && args[1]->decimals > 0) |
2148 | interval_dec= MY_MIN(args[1]->decimals, TIME_SECOND_PART_DIGITS); |
2149 | |
2150 | if (arg0_field_type == MYSQL_TYPE_DATETIME || |
2151 | arg0_field_type == MYSQL_TYPE_TIMESTAMP) |
2152 | { |
2153 | uint dec= MY_MAX(args[0]->datetime_precision(), interval_dec); |
2154 | set_handler(&type_handler_datetime); |
2155 | fix_attributes_datetime(dec); |
2156 | } |
2157 | else if (arg0_field_type == MYSQL_TYPE_DATE) |
2158 | { |
2159 | if (int_type <= INTERVAL_DAY || int_type == INTERVAL_YEAR_MONTH) |
2160 | { |
2161 | set_handler(&type_handler_newdate); |
2162 | fix_attributes_date(); |
2163 | } |
2164 | else |
2165 | { |
2166 | set_handler(&type_handler_datetime2); |
2167 | fix_attributes_datetime(interval_dec); |
2168 | } |
2169 | } |
2170 | else if (arg0_field_type == MYSQL_TYPE_TIME) |
2171 | { |
2172 | uint dec= MY_MAX(args[0]->time_precision(), interval_dec); |
2173 | if (int_type >= INTERVAL_DAY && int_type != INTERVAL_YEAR_MONTH) |
2174 | { |
2175 | set_handler(&type_handler_time2); |
2176 | fix_attributes_time(dec); |
2177 | } |
2178 | else |
2179 | { |
2180 | set_handler(&type_handler_datetime2); |
2181 | fix_attributes_datetime(dec); |
2182 | } |
2183 | } |
2184 | else |
2185 | { |
2186 | uint dec= MY_MAX(args[0]->datetime_precision(), interval_dec); |
2187 | set_handler(&type_handler_string); |
2188 | collation.set(default_charset(), DERIVATION_COERCIBLE, MY_REPERTOIRE_ASCII); |
2189 | fix_char_length_temporal_not_fixed_dec(MAX_DATETIME_WIDTH, dec); |
2190 | } |
2191 | maybe_null= true; |
2192 | } |
2193 | |
2194 | |
2195 | bool Item_date_add_interval::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
2196 | { |
2197 | INTERVAL interval; |
2198 | |
2199 | if (args[0]->get_date(ltime, |
2200 | field_type() == MYSQL_TYPE_TIME ? |
2201 | TIME_TIME_ONLY : 0) || |
2202 | get_interval_value(args[1], int_type, &interval)) |
2203 | return (null_value=1); |
2204 | |
2205 | if (ltime->time_type != MYSQL_TIMESTAMP_TIME && |
2206 | check_date_with_warn(ltime, TIME_NO_ZERO_DATE | TIME_NO_ZERO_IN_DATE, |
2207 | MYSQL_TIMESTAMP_ERROR)) |
2208 | return (null_value=1); |
2209 | |
2210 | if (date_sub_interval) |
2211 | interval.neg = !interval.neg; |
2212 | |
2213 | if (date_add_interval(ltime, int_type, interval)) |
2214 | return (null_value=1); |
2215 | return (null_value= 0); |
2216 | } |
2217 | |
2218 | |
2219 | bool Item_date_add_interval::eq(const Item *item, bool binary_cmp) const |
2220 | { |
2221 | Item_date_add_interval *other= (Item_date_add_interval*) item; |
2222 | if (!Item_func::eq(item, binary_cmp)) |
2223 | return 0; |
2224 | return ((int_type == other->int_type) && |
2225 | (date_sub_interval == other->date_sub_interval)); |
2226 | } |
2227 | |
2228 | /* |
2229 | 'interval_names' reflects the order of the enumeration interval_type. |
2230 | See item_timefunc.h |
2231 | */ |
2232 | |
2233 | static const char *interval_names[]= |
2234 | { |
2235 | "year" , "quarter" , "month" , "week" , "day" , |
2236 | "hour" , "minute" , "second" , "microsecond" , |
2237 | "year_month" , "day_hour" , "day_minute" , |
2238 | "day_second" , "hour_minute" , "hour_second" , |
2239 | "minute_second" , "day_microsecond" , |
2240 | "hour_microsecond" , "minute_microsecond" , |
2241 | "second_microsecond" |
2242 | }; |
2243 | |
2244 | void Item_date_add_interval::print(String *str, enum_query_type query_type) |
2245 | { |
2246 | args[0]->print_parenthesised(str, query_type, ADDINTERVAL_PRECEDENCE); |
2247 | str->append(date_sub_interval?" - interval " :" + interval " ); |
2248 | args[1]->print_parenthesised(str, query_type, INTERVAL_PRECEDENCE); |
2249 | str->append(' '); |
2250 | str->append(interval_names[int_type]); |
2251 | } |
2252 | |
2253 | void Item_extract::(String *str, enum_query_type query_type) |
2254 | { |
2255 | str->append(STRING_WITH_LEN("extract(" )); |
2256 | str->append(interval_names[int_type]); |
2257 | str->append(STRING_WITH_LEN(" from " )); |
2258 | args[0]->print(str, query_type); |
2259 | str->append(')'); |
2260 | } |
2261 | |
2262 | void Item_extract::fix_length_and_dec() |
2263 | { |
2264 | maybe_null=1; // If wrong date |
2265 | switch (int_type) { |
2266 | case INTERVAL_YEAR: set_date_length(4); break; // YYYY |
2267 | case INTERVAL_YEAR_MONTH: set_date_length(6); break; // YYYYMM |
2268 | case INTERVAL_QUARTER: set_date_length(2); break; // 1..4 |
2269 | case INTERVAL_MONTH: set_date_length(2); break; // MM |
2270 | case INTERVAL_WEEK: set_date_length(2); break; // 0..52 |
2271 | case INTERVAL_DAY: set_date_length(2); break; // DD |
2272 | case INTERVAL_DAY_HOUR: set_time_length(4); break; // DDhh |
2273 | case INTERVAL_DAY_MINUTE: set_time_length(6); break; // DDhhmm |
2274 | case INTERVAL_DAY_SECOND: set_time_length(8); break; // DDhhmmss |
2275 | case INTERVAL_HOUR: set_time_length(2); break; // hh |
2276 | case INTERVAL_HOUR_MINUTE: set_time_length(4); break; // hhmm |
2277 | case INTERVAL_HOUR_SECOND: set_time_length(6); break; // hhmmss |
2278 | case INTERVAL_MINUTE: set_time_length(2); break; // mm |
2279 | case INTERVAL_MINUTE_SECOND: set_time_length(4); break; // mmss |
2280 | case INTERVAL_SECOND: set_time_length(2); break; // ss |
2281 | case INTERVAL_MICROSECOND: set_time_length(6); break; // ffffff |
2282 | case INTERVAL_DAY_MICROSECOND: set_time_length(14); break; // DDhhmmssffffff |
2283 | case INTERVAL_HOUR_MICROSECOND: set_time_length(12); break; // hhmmssffffff |
2284 | case INTERVAL_MINUTE_MICROSECOND: set_time_length(10); break; // mmssffffff |
2285 | case INTERVAL_SECOND_MICROSECOND: set_time_length(8); break; // ssffffff |
2286 | case INTERVAL_LAST: DBUG_ASSERT(0); break; /* purecov: deadcode */ |
2287 | } |
2288 | } |
2289 | |
2290 | |
2291 | longlong Item_extract::() |
2292 | { |
2293 | DBUG_ASSERT(fixed == 1); |
2294 | MYSQL_TIME ltime; |
2295 | uint year; |
2296 | ulong week_format; |
2297 | long neg; |
2298 | int is_time_flag = date_value ? 0 : TIME_TIME_ONLY; |
2299 | |
2300 | // Not using get_arg0_date to avoid automatic TIME to DATETIME conversion |
2301 | if ((null_value= args[0]->get_date(<ime, is_time_flag))) |
2302 | return 0; |
2303 | |
2304 | neg= ltime.neg ? -1 : 1; |
2305 | |
2306 | DBUG_ASSERT(ltime.time_type != MYSQL_TIMESTAMP_TIME || ltime.day == 0); |
2307 | if (ltime.time_type == MYSQL_TIMESTAMP_TIME) |
2308 | time_to_daytime_interval(<ime); |
2309 | |
2310 | switch (int_type) { |
2311 | case INTERVAL_YEAR: return ltime.year; |
2312 | case INTERVAL_YEAR_MONTH: return ltime.year*100L+ltime.month; |
2313 | case INTERVAL_QUARTER: return (ltime.month+2)/3; |
2314 | case INTERVAL_MONTH: return ltime.month; |
2315 | case INTERVAL_WEEK: |
2316 | { |
2317 | week_format= current_thd->variables.default_week_format; |
2318 | return calc_week(<ime, week_mode(week_format), &year); |
2319 | } |
2320 | case INTERVAL_DAY: return ltime.day; |
2321 | case INTERVAL_DAY_HOUR: return (long) (ltime.day*100L+ltime.hour)*neg; |
2322 | case INTERVAL_DAY_MINUTE: return (long) (ltime.day*10000L+ |
2323 | ltime.hour*100L+ |
2324 | ltime.minute)*neg; |
2325 | case INTERVAL_DAY_SECOND: return ((longlong) ltime.day*1000000L+ |
2326 | (longlong) (ltime.hour*10000L+ |
2327 | ltime.minute*100+ |
2328 | ltime.second))*neg; |
2329 | case INTERVAL_HOUR: return (long) ltime.hour*neg; |
2330 | case INTERVAL_HOUR_MINUTE: return (long) (ltime.hour*100+ltime.minute)*neg; |
2331 | case INTERVAL_HOUR_SECOND: return (long) (ltime.hour*10000+ltime.minute*100+ |
2332 | ltime.second)*neg; |
2333 | case INTERVAL_MINUTE: return (long) ltime.minute*neg; |
2334 | case INTERVAL_MINUTE_SECOND: return (long) (ltime.minute*100+ltime.second)*neg; |
2335 | case INTERVAL_SECOND: return (long) ltime.second*neg; |
2336 | case INTERVAL_MICROSECOND: return (long) ltime.second_part*neg; |
2337 | case INTERVAL_DAY_MICROSECOND: return (((longlong)ltime.day*1000000L + |
2338 | (longlong)ltime.hour*10000L + |
2339 | ltime.minute*100 + |
2340 | ltime.second)*1000000L + |
2341 | ltime.second_part)*neg; |
2342 | case INTERVAL_HOUR_MICROSECOND: return (((longlong)ltime.hour*10000L + |
2343 | ltime.minute*100 + |
2344 | ltime.second)*1000000L + |
2345 | ltime.second_part)*neg; |
2346 | case INTERVAL_MINUTE_MICROSECOND: return (((longlong)(ltime.minute*100+ |
2347 | ltime.second))*1000000L+ |
2348 | ltime.second_part)*neg; |
2349 | case INTERVAL_SECOND_MICROSECOND: return ((longlong)ltime.second*1000000L+ |
2350 | ltime.second_part)*neg; |
2351 | case INTERVAL_LAST: DBUG_ASSERT(0); break; /* purecov: deadcode */ |
2352 | } |
2353 | return 0; // Impossible |
2354 | } |
2355 | |
2356 | bool Item_extract::(const Item *item, bool binary_cmp) const |
2357 | { |
2358 | if (this == item) |
2359 | return 1; |
2360 | if (item->type() != FUNC_ITEM || |
2361 | functype() != ((Item_func*)item)->functype()) |
2362 | return 0; |
2363 | |
2364 | Item_extract* ie= (Item_extract*)item; |
2365 | if (ie->int_type != int_type) |
2366 | return 0; |
2367 | |
2368 | if (!args[0]->eq(ie->args[0], binary_cmp)) |
2369 | return 0; |
2370 | return 1; |
2371 | } |
2372 | |
2373 | |
2374 | bool Item_char_typecast::eq(const Item *item, bool binary_cmp) const |
2375 | { |
2376 | if (this == item) |
2377 | return 1; |
2378 | if (item->type() != FUNC_ITEM || |
2379 | functype() != ((Item_func*)item)->functype()) |
2380 | return 0; |
2381 | |
2382 | Item_char_typecast *cast= (Item_char_typecast*)item; |
2383 | if (cast_length != cast->cast_length || |
2384 | cast_cs != cast->cast_cs) |
2385 | return 0; |
2386 | |
2387 | if (!args[0]->eq(cast->args[0], binary_cmp)) |
2388 | return 0; |
2389 | return 1; |
2390 | } |
2391 | |
2392 | void Item_temporal_typecast::print(String *str, enum_query_type query_type) |
2393 | { |
2394 | char buf[32]; |
2395 | str->append(STRING_WITH_LEN("cast(" )); |
2396 | args[0]->print(str, query_type); |
2397 | str->append(STRING_WITH_LEN(" as " )); |
2398 | str->append(cast_type()); |
2399 | if (decimals && decimals != NOT_FIXED_DEC) |
2400 | { |
2401 | str->append('('); |
2402 | str->append(llstr(decimals, buf)); |
2403 | str->append(')'); |
2404 | } |
2405 | str->append(')'); |
2406 | } |
2407 | |
2408 | |
2409 | void Item_char_typecast::print(String *str, enum_query_type query_type) |
2410 | { |
2411 | str->append(STRING_WITH_LEN("cast(" )); |
2412 | args[0]->print(str, query_type); |
2413 | str->append(STRING_WITH_LEN(" as char" )); |
2414 | if (cast_length != ~0U) |
2415 | { |
2416 | str->append('('); |
2417 | char buffer[20]; |
2418 | // my_charset_bin is good enough for numbers |
2419 | String st(buffer, sizeof(buffer), &my_charset_bin); |
2420 | st.set(static_cast<ulonglong>(cast_length), &my_charset_bin); |
2421 | str->append(st); |
2422 | str->append(')'); |
2423 | } |
2424 | if (cast_cs) |
2425 | { |
2426 | str->append(STRING_WITH_LEN(" charset " )); |
2427 | str->append(cast_cs->csname); |
2428 | } |
2429 | str->append(')'); |
2430 | } |
2431 | |
2432 | |
2433 | void Item_char_typecast::check_truncation_with_warn(String *src, size_t dstlen) |
2434 | { |
2435 | if (dstlen < src->length()) |
2436 | { |
2437 | THD *thd= current_thd; |
2438 | char char_type[40]; |
2439 | ErrConvString err(src); |
2440 | bool save_abort_on_warning= thd->abort_on_warning; |
2441 | thd->abort_on_warning&= !m_suppress_warning_to_error_escalation; |
2442 | my_snprintf(char_type, sizeof(char_type), "%s(%lu)" , |
2443 | cast_cs == &my_charset_bin ? "BINARY" : "CHAR" , |
2444 | (ulong) cast_length); |
2445 | push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, |
2446 | ER_TRUNCATED_WRONG_VALUE, |
2447 | ER_THD(thd, ER_TRUNCATED_WRONG_VALUE), char_type, |
2448 | err.ptr()); |
2449 | thd->abort_on_warning= save_abort_on_warning; |
2450 | } |
2451 | } |
2452 | |
2453 | |
2454 | String *Item_char_typecast::reuse(String *src, size_t length) |
2455 | { |
2456 | DBUG_ASSERT(length <= src->length()); |
2457 | check_truncation_with_warn(src, length); |
2458 | tmp_value.set(src->ptr(), length, cast_cs); |
2459 | return &tmp_value; |
2460 | } |
2461 | |
2462 | |
2463 | /* |
2464 | Make a copy, to handle conversion or fix bad bytes. |
2465 | */ |
2466 | String *Item_char_typecast::copy(String *str, CHARSET_INFO *strcs) |
2467 | { |
2468 | String_copier_for_item copier(current_thd); |
2469 | if (copier.copy_with_warn(cast_cs, &tmp_value, strcs, |
2470 | str->ptr(), str->length(), cast_length)) |
2471 | { |
2472 | null_value= 1; // EOM |
2473 | return 0; |
2474 | } |
2475 | check_truncation_with_warn(str, (uint)(copier.source_end_pos() - str->ptr())); |
2476 | return &tmp_value; |
2477 | } |
2478 | |
2479 | |
2480 | uint Item_char_typecast::adjusted_length_with_warn(uint length) |
2481 | { |
2482 | if (length <= current_thd->variables.max_allowed_packet) |
2483 | return length; |
2484 | |
2485 | THD *thd= current_thd; |
2486 | push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, |
2487 | ER_WARN_ALLOWED_PACKET_OVERFLOWED, |
2488 | ER_THD(thd, ER_WARN_ALLOWED_PACKET_OVERFLOWED), |
2489 | cast_cs == &my_charset_bin ? |
2490 | "cast_as_binary" : func_name(), |
2491 | thd->variables.max_allowed_packet); |
2492 | return thd->variables.max_allowed_packet; |
2493 | } |
2494 | |
2495 | |
2496 | String *Item_char_typecast::val_str(String *str) |
2497 | { |
2498 | DBUG_ASSERT(fixed == 1); |
2499 | String *res; |
2500 | |
2501 | if (has_explicit_length()) |
2502 | cast_length= adjusted_length_with_warn(cast_length); |
2503 | |
2504 | if (!(res= args[0]->val_str(str))) |
2505 | { |
2506 | null_value= 1; |
2507 | return 0; |
2508 | } |
2509 | |
2510 | if (cast_cs == &my_charset_bin && |
2511 | has_explicit_length() && |
2512 | cast_length > res->length()) |
2513 | { |
2514 | // Special case: pad binary value with trailing 0x00 |
2515 | DBUG_ASSERT(cast_length <= current_thd->variables.max_allowed_packet); |
2516 | if (res->alloced_length() < cast_length) |
2517 | { |
2518 | str_value.alloc(cast_length); |
2519 | str_value.copy(*res); |
2520 | res= &str_value; |
2521 | } |
2522 | bzero((char*) res->ptr() + res->length(), cast_length - res->length()); |
2523 | res->length(cast_length); |
2524 | res->set_charset(&my_charset_bin); |
2525 | } |
2526 | else |
2527 | { |
2528 | /* |
2529 | from_cs is 0 in the case where the result set may vary between calls, |
2530 | for example with dynamic columns. |
2531 | */ |
2532 | CHARSET_INFO *cs= from_cs ? from_cs : res->charset(); |
2533 | if (!charset_conversion) |
2534 | { |
2535 | // Try to reuse the original string (if well formed). |
2536 | Well_formed_prefix prefix(cs, res->ptr(), res->end(), cast_length); |
2537 | if (!prefix.well_formed_error_pos()) |
2538 | res= reuse(res, prefix.length()); |
2539 | goto end; |
2540 | } |
2541 | // Character set conversion, or bad bytes were found. |
2542 | if (!(res= copy(res, cs))) |
2543 | return 0; |
2544 | } |
2545 | |
2546 | end: |
2547 | return ((null_value= (res->length() > |
2548 | adjusted_length_with_warn(res->length())))) ? 0 : res; |
2549 | } |
2550 | |
2551 | |
2552 | void Item_char_typecast::fix_length_and_dec_numeric() |
2553 | { |
2554 | fix_length_and_dec_internal(from_cs= cast_cs->mbminlen == 1 ? |
2555 | cast_cs : |
2556 | &my_charset_latin1); |
2557 | } |
2558 | |
2559 | |
2560 | void Item_char_typecast::fix_length_and_dec_generic() |
2561 | { |
2562 | fix_length_and_dec_internal(from_cs= args[0]->dynamic_result() ? |
2563 | 0 : |
2564 | args[0]->collation.collation); |
2565 | } |
2566 | |
2567 | |
2568 | void Item_char_typecast::fix_length_and_dec_internal(CHARSET_INFO *from_cs) |
2569 | { |
2570 | uint32 char_length; |
2571 | /* |
2572 | We always force character set conversion if cast_cs |
2573 | is a multi-byte character set. It garantees that the |
2574 | result of CAST is a well-formed string. |
2575 | For single-byte character sets we allow just to copy |
2576 | from the argument. A single-byte character sets string |
2577 | is always well-formed. |
2578 | |
2579 | There is a special trick to convert form a number to ucs2. |
2580 | As numbers have my_charset_bin as their character set, |
2581 | it wouldn't do conversion to ucs2 without an additional action. |
2582 | To force conversion, we should pretend to be non-binary. |
2583 | Let's choose from_cs this way: |
2584 | - If the argument in a number and cast_cs is ucs2 (i.e. mbminlen > 1), |
2585 | then from_cs is set to latin1, to perform latin1 -> ucs2 conversion. |
2586 | - If the argument is a number and cast_cs is ASCII-compatible |
2587 | (i.e. mbminlen == 1), then from_cs is set to cast_cs, |
2588 | which allows just to take over the args[0]->val_str() result |
2589 | and thus avoid unnecessary character set conversion. |
2590 | - If the argument is not a number, then from_cs is set to |
2591 | the argument's charset. |
2592 | - If argument has a dynamic collation (can change from call to call) |
2593 | we set from_cs to 0 as a marker that we have to take the collation |
2594 | from the result string. |
2595 | |
2596 | Note (TODO): we could use repertoire technique here. |
2597 | */ |
2598 | charset_conversion= !from_cs || (cast_cs->mbmaxlen > 1) || |
2599 | (!my_charset_same(from_cs, cast_cs) && |
2600 | from_cs != &my_charset_bin && |
2601 | cast_cs != &my_charset_bin); |
2602 | collation.set(cast_cs, DERIVATION_IMPLICIT); |
2603 | char_length= ((cast_length != ~0U) ? cast_length : |
2604 | args[0]->max_length / |
2605 | (cast_cs == &my_charset_bin ? 1 : |
2606 | args[0]->collation.collation->mbmaxlen)); |
2607 | max_length= char_length * cast_cs->mbmaxlen; |
2608 | } |
2609 | |
2610 | |
2611 | bool Item_time_typecast::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
2612 | { |
2613 | Time tm(args[0], Time::Options_for_cast()); |
2614 | if ((null_value= !tm.is_valid_time())) |
2615 | return true; |
2616 | tm.copy_to_mysql_time(ltime); |
2617 | if (decimals < TIME_SECOND_PART_DIGITS) |
2618 | my_time_trunc(ltime, decimals); |
2619 | return (fuzzy_date & TIME_TIME_ONLY) ? 0 : |
2620 | (null_value= check_date_with_warn(ltime, fuzzy_date, |
2621 | MYSQL_TIMESTAMP_ERROR)); |
2622 | } |
2623 | |
2624 | |
2625 | bool Item_date_typecast::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
2626 | { |
2627 | fuzzy_date |= sql_mode_for_dates(current_thd); |
2628 | if (get_arg0_date(ltime, fuzzy_date & ~TIME_TIME_ONLY)) |
2629 | return 1; |
2630 | |
2631 | if (make_date_with_warn(ltime, fuzzy_date, MYSQL_TIMESTAMP_DATE)) |
2632 | return (null_value= 1); |
2633 | |
2634 | return 0; |
2635 | } |
2636 | |
2637 | |
2638 | bool Item_datetime_typecast::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
2639 | { |
2640 | fuzzy_date |= sql_mode_for_dates(current_thd); |
2641 | if (get_arg0_date(ltime, fuzzy_date & ~TIME_TIME_ONLY)) |
2642 | return 1; |
2643 | |
2644 | if (decimals < TIME_SECOND_PART_DIGITS) |
2645 | my_time_trunc(ltime, decimals); |
2646 | |
2647 | DBUG_ASSERT(ltime->time_type != MYSQL_TIMESTAMP_TIME); |
2648 | ltime->time_type= MYSQL_TIMESTAMP_DATETIME; |
2649 | return 0; |
2650 | } |
2651 | |
2652 | |
2653 | /** |
2654 | MAKEDATE(a,b) is a date function that creates a date value |
2655 | from a year and day value. |
2656 | |
2657 | NOTES: |
2658 | As arguments are integers, we can't know if the year is a 2 digit |
2659 | or 4 digit year. In this case we treat all years < 100 as 2 digit |
2660 | years. Ie, this is not safe for dates between 0000-01-01 and |
2661 | 0099-12-31 |
2662 | */ |
2663 | |
2664 | bool Item_func_makedate::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
2665 | { |
2666 | DBUG_ASSERT(fixed == 1); |
2667 | long daynr= (long) args[1]->val_int(); |
2668 | long year= (long) args[0]->val_int(); |
2669 | long days; |
2670 | |
2671 | if (args[0]->null_value || args[1]->null_value || |
2672 | year < 0 || year > 9999 || daynr <= 0) |
2673 | goto err; |
2674 | |
2675 | if (year < 100) |
2676 | year= year_2000_handling(year); |
2677 | |
2678 | days= calc_daynr(year,1,1) + daynr - 1; |
2679 | if (get_date_from_daynr(days, <ime->year, <ime->month, <ime->day)) |
2680 | goto err; |
2681 | ltime->time_type= MYSQL_TIMESTAMP_DATE; |
2682 | ltime->neg= 0; |
2683 | ltime->hour= ltime->minute= ltime->second= ltime->second_part= 0; |
2684 | return (null_value= 0); |
2685 | |
2686 | err: |
2687 | return (null_value= 1); |
2688 | } |
2689 | |
2690 | |
2691 | void Item_func_add_time::fix_length_and_dec() |
2692 | { |
2693 | enum_field_types arg0_field_type; |
2694 | |
2695 | if (!args[0]->type_handler()->is_traditional_type() || |
2696 | !args[1]->type_handler()->is_traditional_type()) |
2697 | { |
2698 | my_error(ER_ILLEGAL_PARAMETER_DATA_TYPES2_FOR_OPERATION, MYF(0), |
2699 | args[0]->type_handler()->name().ptr(), |
2700 | args[1]->type_handler()->name().ptr(), func_name()); |
2701 | return; |
2702 | } |
2703 | /* |
2704 | The field type for the result of an Item_func_add_time function is defined |
2705 | as follows: |
2706 | |
2707 | - If first arg is a MYSQL_TYPE_DATETIME or MYSQL_TYPE_TIMESTAMP |
2708 | result is MYSQL_TYPE_DATETIME |
2709 | - If first arg is a MYSQL_TYPE_TIME result is MYSQL_TYPE_TIME |
2710 | - Otherwise the result is MYSQL_TYPE_STRING |
2711 | */ |
2712 | |
2713 | arg0_field_type= args[0]->field_type(); |
2714 | if (arg0_field_type == MYSQL_TYPE_DATE || |
2715 | arg0_field_type == MYSQL_TYPE_DATETIME || |
2716 | arg0_field_type == MYSQL_TYPE_TIMESTAMP || |
2717 | is_date) |
2718 | { |
2719 | uint dec= MY_MAX(args[0]->datetime_precision(), args[1]->time_precision()); |
2720 | set_handler(&type_handler_datetime2); |
2721 | fix_attributes_datetime(dec); |
2722 | } |
2723 | else if (arg0_field_type == MYSQL_TYPE_TIME) |
2724 | { |
2725 | uint dec= MY_MAX(args[0]->time_precision(), args[1]->time_precision()); |
2726 | set_handler(&type_handler_time2); |
2727 | fix_attributes_time(dec); |
2728 | } |
2729 | else |
2730 | { |
2731 | uint dec= MY_MAX(args[0]->decimals, args[1]->decimals); |
2732 | set_handler(&type_handler_string); |
2733 | collation.set(default_charset(), DERIVATION_COERCIBLE, MY_REPERTOIRE_ASCII); |
2734 | fix_char_length_temporal_not_fixed_dec(MAX_DATETIME_WIDTH, dec); |
2735 | } |
2736 | maybe_null= true; |
2737 | } |
2738 | |
2739 | /** |
2740 | ADDTIME(t,a) and SUBTIME(t,a) are time functions that calculate a |
2741 | time/datetime value |
2742 | |
2743 | t: time_or_datetime_expression |
2744 | a: time_expression |
2745 | |
2746 | Result: Time value or datetime value |
2747 | */ |
2748 | |
2749 | bool Item_func_add_time::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
2750 | { |
2751 | DBUG_ASSERT(fixed == 1); |
2752 | MYSQL_TIME l_time1, l_time2; |
2753 | bool is_time= 0; |
2754 | long days, microseconds; |
2755 | longlong seconds; |
2756 | int l_sign= sign; |
2757 | |
2758 | if (Item_func_add_time::field_type() == MYSQL_TYPE_DATETIME) |
2759 | { |
2760 | // TIMESTAMP function OR the first argument is DATE/DATETIME/TIMESTAMP |
2761 | if (get_arg0_date(&l_time1, 0) || |
2762 | args[1]->get_time(&l_time2) || |
2763 | l_time1.time_type == MYSQL_TIMESTAMP_TIME || |
2764 | l_time2.time_type != MYSQL_TIMESTAMP_TIME) |
2765 | return (null_value= 1); |
2766 | } |
2767 | else |
2768 | { |
2769 | // ADDTIME function AND the first argument is TIME |
2770 | if (args[0]->get_time(&l_time1) || |
2771 | args[1]->get_time(&l_time2) || |
2772 | l_time2.time_type != MYSQL_TIMESTAMP_TIME) |
2773 | return (null_value= 1); |
2774 | is_time= (l_time1.time_type == MYSQL_TIMESTAMP_TIME); |
2775 | } |
2776 | if (l_time1.neg != l_time2.neg) |
2777 | l_sign= -l_sign; |
2778 | |
2779 | bzero(ltime, sizeof(*ltime)); |
2780 | |
2781 | ltime->neg= calc_time_diff(&l_time1, &l_time2, -l_sign, |
2782 | &seconds, µseconds); |
2783 | |
2784 | /* |
2785 | If first argument was negative and diff between arguments |
2786 | is non-zero we need to swap sign to get proper result. |
2787 | */ |
2788 | if (l_time1.neg && (seconds || microseconds)) |
2789 | ltime->neg= 1-ltime->neg; // Swap sign of result |
2790 | |
2791 | if (!is_time && ltime->neg) |
2792 | return (null_value= 1); |
2793 | |
2794 | days= (long) (seconds / SECONDS_IN_24H); |
2795 | |
2796 | calc_time_from_sec(ltime, (long)(seconds % SECONDS_IN_24H), microseconds); |
2797 | |
2798 | ltime->time_type= is_time ? MYSQL_TIMESTAMP_TIME : MYSQL_TIMESTAMP_DATETIME; |
2799 | |
2800 | if (!is_time) |
2801 | { |
2802 | if (get_date_from_daynr(days,<ime->year,<ime->month,<ime->day) || |
2803 | !ltime->day) |
2804 | return (null_value= 1); |
2805 | return (null_value= 0); |
2806 | } |
2807 | |
2808 | ltime->hour+= days*24; |
2809 | return (null_value= adjust_time_range_with_warn(ltime, decimals)); |
2810 | } |
2811 | |
2812 | |
2813 | void Item_func_add_time::print(String *str, enum_query_type query_type) |
2814 | { |
2815 | if (is_date) |
2816 | { |
2817 | DBUG_ASSERT(sign > 0); |
2818 | str->append(STRING_WITH_LEN("timestamp(" )); |
2819 | } |
2820 | else |
2821 | { |
2822 | if (sign > 0) |
2823 | str->append(STRING_WITH_LEN("addtime(" )); |
2824 | else |
2825 | str->append(STRING_WITH_LEN("subtime(" )); |
2826 | } |
2827 | args[0]->print(str, query_type); |
2828 | str->append(','); |
2829 | args[1]->print(str, query_type); |
2830 | str->append(')'); |
2831 | } |
2832 | |
2833 | |
2834 | /** |
2835 | TIMEDIFF(t,s) is a time function that calculates the |
2836 | time value between a start and end time. |
2837 | |
2838 | t and s: time_or_datetime_expression |
2839 | Result: Time value |
2840 | */ |
2841 | |
2842 | bool Item_func_timediff::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
2843 | { |
2844 | DBUG_ASSERT(fixed == 1); |
2845 | int l_sign= 1; |
2846 | MYSQL_TIME l_time1,l_time2,l_time3; |
2847 | ErrConvTime str(&l_time3); |
2848 | |
2849 | /* the following may be true in, for example, date_add(timediff(...), ... */ |
2850 | if (fuzzy_date & TIME_NO_ZERO_IN_DATE) |
2851 | return (null_value= 1); |
2852 | |
2853 | if (args[0]->get_time(&l_time1) || |
2854 | args[1]->get_time(&l_time2) || |
2855 | l_time1.time_type != l_time2.time_type) |
2856 | return (null_value= 1); |
2857 | |
2858 | if (l_time1.neg != l_time2.neg) |
2859 | l_sign= -l_sign; |
2860 | |
2861 | if (calc_time_diff(&l_time1, &l_time2, l_sign, &l_time3, fuzzy_date)) |
2862 | return (null_value= 1); |
2863 | |
2864 | *ltime= l_time3; |
2865 | return (null_value= adjust_time_range_with_warn(ltime, decimals)); |
2866 | } |
2867 | |
2868 | /** |
2869 | MAKETIME(h,m,s) is a time function that calculates a time value |
2870 | from the total number of hours, minutes, and seconds. |
2871 | Result: Time value |
2872 | */ |
2873 | |
2874 | bool Item_func_maketime::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
2875 | { |
2876 | DBUG_ASSERT(fixed == 1); |
2877 | bool overflow= 0; |
2878 | longlong hour= args[0]->val_int(); |
2879 | longlong minute= args[1]->val_int(); |
2880 | ulonglong second; |
2881 | ulong microsecond; |
2882 | bool neg= args[2]->get_seconds(&second, µsecond); |
2883 | |
2884 | if (args[0]->null_value || args[1]->null_value || args[2]->null_value || |
2885 | minute < 0 || minute > 59 || neg || second > 59) |
2886 | return (null_value= 1); |
2887 | |
2888 | bzero(ltime, sizeof(*ltime)); |
2889 | ltime->time_type= MYSQL_TIMESTAMP_TIME; |
2890 | |
2891 | /* Check for integer overflows */ |
2892 | if (hour < 0) |
2893 | { |
2894 | if (args[0]->unsigned_flag) |
2895 | overflow= 1; |
2896 | else |
2897 | ltime->neg= 1; |
2898 | } |
2899 | if (-hour > TIME_MAX_HOUR || hour > TIME_MAX_HOUR) |
2900 | overflow= 1; |
2901 | |
2902 | if (!overflow) |
2903 | { |
2904 | ltime->hour= (uint) ((hour < 0 ? -hour : hour)); |
2905 | ltime->minute= (uint) minute; |
2906 | ltime->second= (uint) second; |
2907 | ltime->second_part= microsecond; |
2908 | } |
2909 | else |
2910 | { |
2911 | ltime->hour= TIME_MAX_HOUR; |
2912 | ltime->minute= TIME_MAX_MINUTE; |
2913 | ltime->second= TIME_MAX_SECOND; |
2914 | char buf[28]; |
2915 | char *ptr= longlong10_to_str(hour, buf, args[0]->unsigned_flag ? 10 : -10); |
2916 | int len = (int)(ptr - buf) + sprintf(ptr, ":%02u:%02u" , (uint)minute, (uint)second); |
2917 | make_truncated_value_warning(current_thd, Sql_condition::WARN_LEVEL_WARN, |
2918 | buf, len, MYSQL_TIMESTAMP_TIME, |
2919 | NullS); |
2920 | } |
2921 | |
2922 | return (null_value= 0); |
2923 | } |
2924 | |
2925 | |
2926 | /** |
2927 | MICROSECOND(a) is a function ( extraction) that extracts the microseconds |
2928 | from a. |
2929 | |
2930 | a: Datetime or time value |
2931 | Result: int value |
2932 | */ |
2933 | |
2934 | longlong Item_func_microsecond::val_int() |
2935 | { |
2936 | DBUG_ASSERT(fixed == 1); |
2937 | Time tm(args[0], Time::Options_for_cast()); |
2938 | return ((null_value= !tm.is_valid_time())) ? |
2939 | 0 : tm.get_mysql_time()->second_part; |
2940 | } |
2941 | |
2942 | |
2943 | longlong Item_func_timestamp_diff::val_int() |
2944 | { |
2945 | MYSQL_TIME ltime1, ltime2; |
2946 | longlong seconds; |
2947 | long microseconds; |
2948 | long months= 0; |
2949 | int neg= 1; |
2950 | THD *thd= current_thd; |
2951 | ulonglong fuzzydate= TIME_NO_ZERO_DATE | TIME_NO_ZERO_IN_DATE; |
2952 | |
2953 | null_value= 0; |
2954 | |
2955 | if (Datetime(thd, args[0], fuzzydate).copy_to_mysql_time(<ime1) || |
2956 | Datetime(thd, args[1], fuzzydate).copy_to_mysql_time(<ime2)) |
2957 | goto null_date; |
2958 | |
2959 | if (calc_time_diff(<ime2,<ime1, 1, |
2960 | &seconds, µseconds)) |
2961 | neg= -1; |
2962 | |
2963 | if (int_type == INTERVAL_YEAR || |
2964 | int_type == INTERVAL_QUARTER || |
2965 | int_type == INTERVAL_MONTH) |
2966 | { |
2967 | uint year_beg, year_end, month_beg, month_end, day_beg, day_end; |
2968 | uint years= 0; |
2969 | uint second_beg, second_end, microsecond_beg, microsecond_end; |
2970 | |
2971 | if (neg == -1) |
2972 | { |
2973 | year_beg= ltime2.year; |
2974 | year_end= ltime1.year; |
2975 | month_beg= ltime2.month; |
2976 | month_end= ltime1.month; |
2977 | day_beg= ltime2.day; |
2978 | day_end= ltime1.day; |
2979 | second_beg= ltime2.hour * 3600 + ltime2.minute * 60 + ltime2.second; |
2980 | second_end= ltime1.hour * 3600 + ltime1.minute * 60 + ltime1.second; |
2981 | microsecond_beg= ltime2.second_part; |
2982 | microsecond_end= ltime1.second_part; |
2983 | } |
2984 | else |
2985 | { |
2986 | year_beg= ltime1.year; |
2987 | year_end= ltime2.year; |
2988 | month_beg= ltime1.month; |
2989 | month_end= ltime2.month; |
2990 | day_beg= ltime1.day; |
2991 | day_end= ltime2.day; |
2992 | second_beg= ltime1.hour * 3600 + ltime1.minute * 60 + ltime1.second; |
2993 | second_end= ltime2.hour * 3600 + ltime2.minute * 60 + ltime2.second; |
2994 | microsecond_beg= ltime1.second_part; |
2995 | microsecond_end= ltime2.second_part; |
2996 | } |
2997 | |
2998 | /* calc years */ |
2999 | years= year_end - year_beg; |
3000 | if (month_end < month_beg || (month_end == month_beg && day_end < day_beg)) |
3001 | years-= 1; |
3002 | |
3003 | /* calc months */ |
3004 | months= 12*years; |
3005 | if (month_end < month_beg || (month_end == month_beg && day_end < day_beg)) |
3006 | months+= 12 - (month_beg - month_end); |
3007 | else |
3008 | months+= (month_end - month_beg); |
3009 | |
3010 | if (day_end < day_beg) |
3011 | months-= 1; |
3012 | else if ((day_end == day_beg) && |
3013 | ((second_end < second_beg) || |
3014 | (second_end == second_beg && microsecond_end < microsecond_beg))) |
3015 | months-= 1; |
3016 | } |
3017 | |
3018 | switch (int_type) { |
3019 | case INTERVAL_YEAR: |
3020 | return months/12*neg; |
3021 | case INTERVAL_QUARTER: |
3022 | return months/3*neg; |
3023 | case INTERVAL_MONTH: |
3024 | return months*neg; |
3025 | case INTERVAL_WEEK: |
3026 | return seconds / SECONDS_IN_24H / 7L * neg; |
3027 | case INTERVAL_DAY: |
3028 | return seconds / SECONDS_IN_24H * neg; |
3029 | case INTERVAL_HOUR: |
3030 | return seconds/3600L*neg; |
3031 | case INTERVAL_MINUTE: |
3032 | return seconds/60L*neg; |
3033 | case INTERVAL_SECOND: |
3034 | return seconds*neg; |
3035 | case INTERVAL_MICROSECOND: |
3036 | /* |
3037 | In MySQL difference between any two valid datetime values |
3038 | in microseconds fits into longlong. |
3039 | */ |
3040 | return (seconds*1000000L+microseconds)*neg; |
3041 | default: |
3042 | break; |
3043 | } |
3044 | |
3045 | null_date: |
3046 | null_value=1; |
3047 | return 0; |
3048 | } |
3049 | |
3050 | |
3051 | void Item_func_timestamp_diff::print(String *str, enum_query_type query_type) |
3052 | { |
3053 | str->append(func_name()); |
3054 | str->append('('); |
3055 | |
3056 | switch (int_type) { |
3057 | case INTERVAL_YEAR: |
3058 | str->append(STRING_WITH_LEN("YEAR" )); |
3059 | break; |
3060 | case INTERVAL_QUARTER: |
3061 | str->append(STRING_WITH_LEN("QUARTER" )); |
3062 | break; |
3063 | case INTERVAL_MONTH: |
3064 | str->append(STRING_WITH_LEN("MONTH" )); |
3065 | break; |
3066 | case INTERVAL_WEEK: |
3067 | str->append(STRING_WITH_LEN("WEEK" )); |
3068 | break; |
3069 | case INTERVAL_DAY: |
3070 | str->append(STRING_WITH_LEN("DAY" )); |
3071 | break; |
3072 | case INTERVAL_HOUR: |
3073 | str->append(STRING_WITH_LEN("HOUR" )); |
3074 | break; |
3075 | case INTERVAL_MINUTE: |
3076 | str->append(STRING_WITH_LEN("MINUTE" )); |
3077 | break; |
3078 | case INTERVAL_SECOND: |
3079 | str->append(STRING_WITH_LEN("SECOND" )); |
3080 | break; |
3081 | case INTERVAL_MICROSECOND: |
3082 | str->append(STRING_WITH_LEN("MICROSECOND" )); |
3083 | break; |
3084 | default: |
3085 | break; |
3086 | } |
3087 | |
3088 | for (uint i=0 ; i < 2 ; i++) |
3089 | { |
3090 | str->append(','); |
3091 | args[i]->print(str, query_type); |
3092 | } |
3093 | str->append(')'); |
3094 | } |
3095 | |
3096 | |
3097 | String *Item_func_get_format::val_str_ascii(String *str) |
3098 | { |
3099 | DBUG_ASSERT(fixed == 1); |
3100 | const char *format_name; |
3101 | KNOWN_DATE_TIME_FORMAT *format; |
3102 | String *val= args[0]->val_str_ascii(str); |
3103 | ulong val_len; |
3104 | |
3105 | if ((null_value= args[0]->null_value)) |
3106 | return 0; |
3107 | |
3108 | val_len= val->length(); |
3109 | for (format= &known_date_time_formats[0]; |
3110 | (format_name= format->format_name); |
3111 | format++) |
3112 | { |
3113 | uint format_name_len; |
3114 | format_name_len= (uint) strlen(format_name); |
3115 | if (val_len == format_name_len && |
3116 | !my_strnncoll(&my_charset_latin1, |
3117 | (const uchar *) val->ptr(), val_len, |
3118 | (const uchar *) format_name, val_len)) |
3119 | { |
3120 | const char *format_str= get_date_time_format_str(format, type); |
3121 | str->set(format_str, (uint) strlen(format_str), &my_charset_numeric); |
3122 | return str; |
3123 | } |
3124 | } |
3125 | |
3126 | null_value= 1; |
3127 | return 0; |
3128 | } |
3129 | |
3130 | |
3131 | void Item_func_get_format::print(String *str, enum_query_type query_type) |
3132 | { |
3133 | str->append(func_name()); |
3134 | str->append('('); |
3135 | |
3136 | switch (type) { |
3137 | case MYSQL_TIMESTAMP_DATE: |
3138 | str->append(STRING_WITH_LEN("DATE, " )); |
3139 | break; |
3140 | case MYSQL_TIMESTAMP_DATETIME: |
3141 | str->append(STRING_WITH_LEN("DATETIME, " )); |
3142 | break; |
3143 | case MYSQL_TIMESTAMP_TIME: |
3144 | str->append(STRING_WITH_LEN("TIME, " )); |
3145 | break; |
3146 | default: |
3147 | DBUG_ASSERT(0); |
3148 | } |
3149 | args[0]->print(str, query_type); |
3150 | str->append(')'); |
3151 | } |
3152 | |
3153 | |
3154 | /** |
3155 | Get type of datetime value (DATE/TIME/...) which will be produced |
3156 | according to format string. |
3157 | |
3158 | @param format format string |
3159 | @param length length of format string |
3160 | |
3161 | @note |
3162 | We don't process day format's characters('D', 'd', 'e') because day |
3163 | may be a member of all date/time types. |
3164 | |
3165 | @note |
3166 | Format specifiers supported by this function should be in sync with |
3167 | specifiers supported by extract_date_time() function. |
3168 | |
3169 | @return |
3170 | One of date_time_format_types values: |
3171 | - DATE_TIME_MICROSECOND |
3172 | - DATE_TIME |
3173 | - DATE_ONLY |
3174 | - TIME_MICROSECOND |
3175 | - TIME_ONLY |
3176 | */ |
3177 | |
3178 | static date_time_format_types |
3179 | get_date_time_result_type(const char *format, uint length) |
3180 | { |
3181 | const char *time_part_frms= "HISThiklrs" ; |
3182 | const char *date_part_frms= "MVUXYWabcjmvuxyw" ; |
3183 | bool date_part_used= 0, time_part_used= 0, frac_second_used= 0; |
3184 | |
3185 | const char *val= format; |
3186 | const char *end= format + length; |
3187 | |
3188 | for (; val != end; val++) |
3189 | { |
3190 | if (*val == '%' && val+1 != end) |
3191 | { |
3192 | val++; |
3193 | if (*val == 'f') |
3194 | frac_second_used= time_part_used= 1; |
3195 | else if (!time_part_used && strchr(time_part_frms, *val)) |
3196 | time_part_used= 1; |
3197 | else if (!date_part_used && strchr(date_part_frms, *val)) |
3198 | date_part_used= 1; |
3199 | if (date_part_used && frac_second_used) |
3200 | { |
3201 | /* |
3202 | frac_second_used implies time_part_used, and thus we already |
3203 | have all types of date-time components and can end our search. |
3204 | */ |
3205 | return DATE_TIME_MICROSECOND; |
3206 | } |
3207 | } |
3208 | } |
3209 | |
3210 | /* We don't have all three types of date-time components */ |
3211 | if (frac_second_used) |
3212 | return TIME_MICROSECOND; |
3213 | if (time_part_used) |
3214 | { |
3215 | if (date_part_used) |
3216 | return DATE_TIME; |
3217 | return TIME_ONLY; |
3218 | } |
3219 | return DATE_ONLY; |
3220 | } |
3221 | |
3222 | |
3223 | void Item_func_str_to_date::fix_length_and_dec() |
3224 | { |
3225 | if (!args[0]->type_handler()->is_traditional_type() || |
3226 | !args[1]->type_handler()->is_traditional_type()) |
3227 | { |
3228 | my_error(ER_ILLEGAL_PARAMETER_DATA_TYPES2_FOR_OPERATION, MYF(0), |
3229 | args[0]->type_handler()->name().ptr(), |
3230 | args[1]->type_handler()->name().ptr(), func_name()); |
3231 | return; |
3232 | } |
3233 | if (agg_arg_charsets(collation, args, 2, MY_COLL_ALLOW_CONV, 1)) |
3234 | return; |
3235 | if (collation.collation->mbminlen > 1) |
3236 | internal_charset= &my_charset_utf8mb4_general_ci; |
3237 | |
3238 | maybe_null= true; |
3239 | set_handler(&type_handler_datetime2); |
3240 | fix_attributes_datetime(TIME_SECOND_PART_DIGITS); |
3241 | |
3242 | if ((const_item= args[1]->const_item())) |
3243 | { |
3244 | char format_buff[64]; |
3245 | String format_str(format_buff, sizeof(format_buff), &my_charset_bin); |
3246 | String *format= args[1]->val_str(&format_str, &format_converter, |
3247 | internal_charset); |
3248 | decimals= 0; |
3249 | if (!args[1]->null_value) |
3250 | { |
3251 | date_time_format_types cached_format_type= |
3252 | get_date_time_result_type(format->ptr(), format->length()); |
3253 | switch (cached_format_type) { |
3254 | case DATE_ONLY: |
3255 | set_handler(&type_handler_newdate); |
3256 | fix_attributes_date(); |
3257 | break; |
3258 | case TIME_MICROSECOND: |
3259 | set_handler(&type_handler_time2); |
3260 | fix_attributes_time(TIME_SECOND_PART_DIGITS); |
3261 | break; |
3262 | case TIME_ONLY: |
3263 | set_handler(&type_handler_time2); |
3264 | fix_attributes_time(0); |
3265 | break; |
3266 | case DATE_TIME_MICROSECOND: |
3267 | set_handler(&type_handler_datetime2); |
3268 | fix_attributes_datetime(TIME_SECOND_PART_DIGITS); |
3269 | break; |
3270 | case DATE_TIME: |
3271 | set_handler(&type_handler_datetime2); |
3272 | fix_attributes_datetime(0); |
3273 | break; |
3274 | } |
3275 | } |
3276 | } |
3277 | cached_timestamp_type= mysql_timestamp_type(); |
3278 | } |
3279 | |
3280 | |
3281 | bool Item_func_str_to_date::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
3282 | { |
3283 | DATE_TIME_FORMAT date_time_format; |
3284 | char val_buff[64], format_buff[64]; |
3285 | String val_string(val_buff, sizeof(val_buff), &my_charset_bin), *val; |
3286 | String format_str(format_buff, sizeof(format_buff), &my_charset_bin), |
3287 | *format; |
3288 | |
3289 | val= args[0]->val_str(&val_string, &subject_converter, internal_charset); |
3290 | format= args[1]->val_str(&format_str, &format_converter, internal_charset); |
3291 | if (args[0]->null_value || args[1]->null_value) |
3292 | return (null_value=1); |
3293 | |
3294 | date_time_format.format.str= (char*) format->ptr(); |
3295 | date_time_format.format.length= format->length(); |
3296 | if (extract_date_time(&date_time_format, val->ptr(), val->length(), |
3297 | ltime, cached_timestamp_type, 0, "datetime" , |
3298 | fuzzy_date | sql_mode_for_dates(current_thd))) |
3299 | return (null_value=1); |
3300 | if (cached_timestamp_type == MYSQL_TIMESTAMP_TIME && ltime->day) |
3301 | { |
3302 | /* |
3303 | Day part for time type can be nonzero value and so |
3304 | we should add hours from day part to hour part to |
3305 | keep valid time value. |
3306 | */ |
3307 | ltime->hour+= ltime->day*24; |
3308 | ltime->day= 0; |
3309 | } |
3310 | return (null_value= 0); |
3311 | } |
3312 | |
3313 | |
3314 | bool Item_func_last_day::get_date(MYSQL_TIME *ltime, ulonglong fuzzy_date) |
3315 | { |
3316 | if (get_arg0_date(ltime, fuzzy_date & ~TIME_TIME_ONLY) || |
3317 | (ltime->month == 0)) |
3318 | return (null_value=1); |
3319 | uint month_idx= ltime->month-1; |
3320 | ltime->day= days_in_month[month_idx]; |
3321 | if ( month_idx == 1 && calc_days_in_year(ltime->year) == 366) |
3322 | ltime->day= 29; |
3323 | ltime->hour= ltime->minute= ltime->second= 0; |
3324 | ltime->second_part= 0; |
3325 | ltime->time_type= MYSQL_TIMESTAMP_DATE; |
3326 | return (null_value= 0); |
3327 | } |
3328 | |