1 // © 2016 and later: Unicode, Inc. and others.
2 // License & terms of use: http://www.unicode.org/copyright.html
3 /*
4 ******************************************************************************
5 *
6 * Copyright (C) 1999-2016, International Business Machines
7 * Corporation and others. All Rights Reserved.
8 *
9 ******************************************************************************
10 * file name: udata.cpp
11 * encoding: UTF-8
12 * tab size: 8 (not used)
13 * indentation:4
14 *
15 * created on: 1999oct25
16 * created by: Markus W. Scherer
17 */
18
19 #include "unicode/utypes.h" /* U_PLATFORM etc. */
20
21 #ifdef __GNUC__
22 /* if gcc
23 #define ATTRIBUTE_WEAK __attribute__ ((weak))
24 might have to #include some other header
25 */
26 #endif
27
28 #include "unicode/putil.h"
29 #include "unicode/udata.h"
30 #include "unicode/uversion.h"
31 #include "charstr.h"
32 #include "cmemory.h"
33 #include "cstring.h"
34 #include "mutex.h"
35 #include "putilimp.h"
36 #include "restrace.h"
37 #include "uassert.h"
38 #include "ucln_cmn.h"
39 #include "ucmndata.h"
40 #include "udatamem.h"
41 #include "uhash.h"
42 #include "umapfile.h"
43 #include "umutex.h"
44
45 /***********************************************************************
46 *
47 * Notes on the organization of the ICU data implementation
48 *
49 * All of the public API is defined in udata.h
50 *
51 * The implementation is split into several files...
52 *
53 * - udata.c (this file) contains higher level code that knows about
54 * the search paths for locating data, caching opened data, etc.
55 *
56 * - umapfile.c contains the low level platform-specific code for actually loading
57 * (memory mapping, file reading, whatever) data into memory.
58 *
59 * - ucmndata.c deals with the tables of contents of ICU data items within
60 * an ICU common format data file. The implementation includes
61 * an abstract interface and support for multiple TOC formats.
62 * All knowledge of any specific TOC format is encapsulated here.
63 *
64 * - udatamem.c has code for managing UDataMemory structs. These are little
65 * descriptor objects for blocks of memory holding ICU data of
66 * various types.
67 */
68
69 /* configuration ---------------------------------------------------------- */
70
71 /* If you are excruciatingly bored turn this on .. */
72 /* #define UDATA_DEBUG 1 */
73
74 #if defined(UDATA_DEBUG)
75 # include <stdio.h>
76 #endif
77
78 U_NAMESPACE_USE
79
80 /*
81 * Forward declarations
82 */
83 static UDataMemory *udata_findCachedData(const char *path, UErrorCode &err);
84
85 /***********************************************************************
86 *
87 * static (Global) data
88 *
89 ************************************************************************/
90
91 /*
92 * Pointers to the common ICU data.
93 *
94 * We store multiple pointers to ICU data packages and iterate through them
95 * when looking for a data item.
96 *
97 * It is possible to combine this with dependency inversion:
98 * One or more data package libraries may export
99 * functions that each return a pointer to their piece of the ICU data,
100 * and this file would import them as weak functions, without a
101 * strong linker dependency from the common library on the data library.
102 *
103 * Then we can have applications depend on only that part of ICU's data
104 * that they really need, reducing the size of binaries that take advantage
105 * of this.
106 */
107 static UDataMemory *gCommonICUDataArray[10] = { NULL }; // Access protected by icu global mutex.
108
109 static u_atomic_int32_t gHaveTriedToLoadCommonData = ATOMIC_INT32_T_INITIALIZER(0); // See extendICUData().
110
111 static UHashtable *gCommonDataCache = NULL; /* Global hash table of opened ICU data files. */
112 static icu::UInitOnce gCommonDataCacheInitOnce = U_INITONCE_INITIALIZER;
113
114 // Android-changed: On Android, use our patched version of openCommonData() to load the data,
115 // and do not try to load ICU data from other files.
116 #if U_PLATFORM == U_PF_ANDROID
117 static UDataFileAccess gDataFileAccess = UDATA_NO_FILES;
118 #elif !defined(ICU_DATA_DIR_WINDOWS)
119 static UDataFileAccess gDataFileAccess = UDATA_DEFAULT_ACCESS; // Access not synchronized.
120 // Modifying is documented as thread-unsafe.
121 #else
122 // If we are using the Windows data directory, then look in one spot only.
123 static UDataFileAccess gDataFileAccess = UDATA_NO_FILES;
124 #endif
125
126 // BEGIN Android-added: Include android/host-linux-specific headers and variables.
127 #ifdef ANDROID // if using the AOSP build system, e.g. Soong, but not the normal GNU make used by ./updateicudata.py
128 #if U_PLATFORM == U_PF_ANDROID || U_PLATFORM == U_PF_LINUX // if targeting Android or host linux
129 #define AOSP_ICU_INIT 1
130 #endif
131 #endif
132
133 #ifdef AOSP_ICU_INIT
134 #include "androidicuinit/android_icu_init.h"
135 static icu::UInitOnce gAospInitOnce = U_INITONCE_INITIALIZER;
136 #endif
137 // END Android-added: Include android/host-linux-specific headers and variables.
138
139
140 static UBool U_CALLCONV
udata_cleanup(void)141 udata_cleanup(void)
142 {
143 int32_t i;
144
145 if (gCommonDataCache) { /* Delete the cache of user data mappings. */
146 uhash_close(gCommonDataCache); /* Table owns the contents, and will delete them. */
147 gCommonDataCache = NULL; /* Cleanup is not thread safe. */
148 }
149 gCommonDataCacheInitOnce.reset();
150
151 for (i = 0; i < UPRV_LENGTHOF(gCommonICUDataArray) && gCommonICUDataArray[i] != NULL; ++i) {
152 udata_close(gCommonICUDataArray[i]);
153 gCommonICUDataArray[i] = NULL;
154 }
155 gHaveTriedToLoadCommonData = 0;
156
157 // BEGIN Android-added: Use specialized libandroidicuinit to unload the data on Android/ART host.
158 #ifdef AOSP_ICU_INIT
159 android_icu_cleanup();
160 gAospInitOnce.reset();
161 #endif
162 // END Android-added: Use specialized libandroidicuinit to unload the data on Android/ART host.
163
164 return TRUE; /* Everything was cleaned up */
165 }
166
167 static UBool U_CALLCONV
findCommonICUDataByName(const char * inBasename,UErrorCode & err)168 findCommonICUDataByName(const char *inBasename, UErrorCode &err)
169 {
170 UBool found = FALSE;
171 int32_t i;
172
173 UDataMemory *pData = udata_findCachedData(inBasename, err);
174 if (U_FAILURE(err) || pData == NULL)
175 return FALSE;
176
177 {
178 Mutex lock;
179 for (i = 0; i < UPRV_LENGTHOF(gCommonICUDataArray); ++i) {
180 if ((gCommonICUDataArray[i] != NULL) && (gCommonICUDataArray[i]->pHeader == pData->pHeader)) {
181 /* The data pointer is already in the array. */
182 found = TRUE;
183 break;
184 }
185 }
186 }
187 return found;
188 }
189
190
191 /*
192 * setCommonICUData. Set a UDataMemory to be the global ICU Data
193 */
194 static UBool
setCommonICUData(UDataMemory * pData,UBool warn,UErrorCode * pErr)195 setCommonICUData(UDataMemory *pData, /* The new common data. Belongs to caller, we copy it. */
196 UBool warn, /* If true, set USING_DEFAULT warning if ICUData was */
197 /* changed by another thread before we got to it. */
198 UErrorCode *pErr)
199 {
200 UDataMemory *newCommonData = UDataMemory_createNewInstance(pErr);
201 int32_t i;
202 UBool didUpdate = FALSE;
203 if (U_FAILURE(*pErr)) {
204 return FALSE;
205 }
206
207 /* For the assignment, other threads must cleanly see either the old */
208 /* or the new, not some partially initialized new. The old can not be */
209 /* deleted - someone may still have a pointer to it lying around in */
210 /* their locals. */
211 UDatamemory_assign(newCommonData, pData);
212 umtx_lock(NULL);
213 for (i = 0; i < UPRV_LENGTHOF(gCommonICUDataArray); ++i) {
214 if (gCommonICUDataArray[i] == NULL) {
215 gCommonICUDataArray[i] = newCommonData;
216 didUpdate = TRUE;
217 break;
218 } else if (gCommonICUDataArray[i]->pHeader == pData->pHeader) {
219 /* The same data pointer is already in the array. */
220 break;
221 }
222 }
223 umtx_unlock(NULL);
224
225 if (i == UPRV_LENGTHOF(gCommonICUDataArray) && warn) {
226 *pErr = U_USING_DEFAULT_WARNING;
227 }
228 if (didUpdate) {
229 ucln_common_registerCleanup(UCLN_COMMON_UDATA, udata_cleanup);
230 } else {
231 uprv_free(newCommonData);
232 }
233 return didUpdate;
234 }
235
236 #if !defined(ICU_DATA_DIR_WINDOWS)
237
238 static UBool
setCommonICUDataPointer(const void * pData,UBool,UErrorCode * pErrorCode)239 setCommonICUDataPointer(const void *pData, UBool /*warn*/, UErrorCode *pErrorCode) {
240 UDataMemory tData;
241 UDataMemory_init(&tData);
242 UDataMemory_setData(&tData, pData);
243 udata_checkCommonData(&tData, pErrorCode);
244 return setCommonICUData(&tData, FALSE, pErrorCode);
245 }
246
247 #endif
248
249 static const char *
findBasename(const char * path)250 findBasename(const char *path) {
251 const char *basename=uprv_strrchr(path, U_FILE_SEP_CHAR);
252 if(basename==NULL) {
253 return path;
254 } else {
255 return basename+1;
256 }
257 }
258
259 #ifdef UDATA_DEBUG
260 static const char *
packageNameFromPath(const char * path)261 packageNameFromPath(const char *path)
262 {
263 if((path == NULL) || (*path == 0)) {
264 return U_ICUDATA_NAME;
265 }
266
267 path = findBasename(path);
268
269 if((path == NULL) || (*path == 0)) {
270 return U_ICUDATA_NAME;
271 }
272
273 return path;
274 }
275 #endif
276
277 /*----------------------------------------------------------------------*
278 * *
279 * Cache for common data *
280 * Functions for looking up or adding entries to a cache of *
281 * data that has been previously opened. Avoids a potentially *
282 * expensive operation of re-opening the data for subsequent *
283 * uses. *
284 * *
285 * Data remains cached for the duration of the process. *
286 * *
287 *----------------------------------------------------------------------*/
288
289 typedef struct DataCacheElement {
290 char *name;
291 UDataMemory *item;
292 } DataCacheElement;
293
294
295
296 /*
297 * Deleter function for DataCacheElements.
298 * udata cleanup function closes the hash table; hash table in turn calls back to
299 * here for each entry.
300 */
DataCacheElement_deleter(void * pDCEl)301 static void U_CALLCONV DataCacheElement_deleter(void *pDCEl) {
302 DataCacheElement *p = (DataCacheElement *)pDCEl;
303 udata_close(p->item); /* unmaps storage */
304 uprv_free(p->name); /* delete the hash key string. */
305 uprv_free(pDCEl); /* delete 'this' */
306 }
307
udata_initHashTable(UErrorCode & err)308 static void U_CALLCONV udata_initHashTable(UErrorCode &err) {
309 U_ASSERT(gCommonDataCache == NULL);
310 gCommonDataCache = uhash_open(uhash_hashChars, uhash_compareChars, NULL, &err);
311 if (U_FAILURE(err)) {
312 return;
313 }
314 U_ASSERT(gCommonDataCache != NULL);
315 uhash_setValueDeleter(gCommonDataCache, DataCacheElement_deleter);
316 ucln_common_registerCleanup(UCLN_COMMON_UDATA, udata_cleanup);
317 }
318
319 /* udata_getCacheHashTable()
320 * Get the hash table used to store the data cache entries.
321 * Lazy create it if it doesn't yet exist.
322 */
udata_getHashTable(UErrorCode & err)323 static UHashtable *udata_getHashTable(UErrorCode &err) {
324 umtx_initOnce(gCommonDataCacheInitOnce, &udata_initHashTable, err);
325 return gCommonDataCache;
326 }
327
328
329
udata_findCachedData(const char * path,UErrorCode & err)330 static UDataMemory *udata_findCachedData(const char *path, UErrorCode &err)
331 {
332 UHashtable *htable;
333 UDataMemory *retVal = NULL;
334 DataCacheElement *el;
335 const char *baseName;
336
337 htable = udata_getHashTable(err);
338 if (U_FAILURE(err)) {
339 return NULL;
340 }
341
342 baseName = findBasename(path); /* Cache remembers only the base name, not the full path. */
343 umtx_lock(NULL);
344 el = (DataCacheElement *)uhash_get(htable, baseName);
345 umtx_unlock(NULL);
346 if (el != NULL) {
347 retVal = el->item;
348 }
349 #ifdef UDATA_DEBUG
350 fprintf(stderr, "Cache: [%s] -> %p\n", baseName, (void*) retVal);
351 #endif
352 return retVal;
353 }
354
355
udata_cacheDataItem(const char * path,UDataMemory * item,UErrorCode * pErr)356 static UDataMemory *udata_cacheDataItem(const char *path, UDataMemory *item, UErrorCode *pErr) {
357 DataCacheElement *newElement;
358 const char *baseName;
359 int32_t nameLen;
360 UHashtable *htable;
361 DataCacheElement *oldValue = NULL;
362 UErrorCode subErr = U_ZERO_ERROR;
363
364 htable = udata_getHashTable(*pErr);
365 if (U_FAILURE(*pErr)) {
366 return NULL;
367 }
368
369 /* Create a new DataCacheElement - the thingy we store in the hash table -
370 * and copy the supplied path and UDataMemoryItems into it.
371 */
372 newElement = (DataCacheElement *)uprv_malloc(sizeof(DataCacheElement));
373 if (newElement == NULL) {
374 *pErr = U_MEMORY_ALLOCATION_ERROR;
375 return NULL;
376 }
377 newElement->item = UDataMemory_createNewInstance(pErr);
378 if (U_FAILURE(*pErr)) {
379 uprv_free(newElement);
380 return NULL;
381 }
382 UDatamemory_assign(newElement->item, item);
383
384 baseName = findBasename(path);
385 nameLen = (int32_t)uprv_strlen(baseName);
386 newElement->name = (char *)uprv_malloc(nameLen+1);
387 if (newElement->name == NULL) {
388 *pErr = U_MEMORY_ALLOCATION_ERROR;
389 uprv_free(newElement->item);
390 uprv_free(newElement);
391 return NULL;
392 }
393 uprv_strcpy(newElement->name, baseName);
394
395 /* Stick the new DataCacheElement into the hash table.
396 */
397 umtx_lock(NULL);
398 oldValue = (DataCacheElement *)uhash_get(htable, path);
399 if (oldValue != NULL) {
400 subErr = U_USING_DEFAULT_WARNING;
401 }
402 else {
403 uhash_put(
404 htable,
405 newElement->name, /* Key */
406 newElement, /* Value */
407 &subErr);
408 }
409 umtx_unlock(NULL);
410
411 #ifdef UDATA_DEBUG
412 fprintf(stderr, "Cache: [%s] <<< %p : %s. vFunc=%p\n", newElement->name,
413 (void*) newElement->item, u_errorName(subErr), (void*) newElement->item->vFuncs);
414 #endif
415
416 if (subErr == U_USING_DEFAULT_WARNING || U_FAILURE(subErr)) {
417 *pErr = subErr; /* copy sub err unto fillin ONLY if something happens. */
418 uprv_free(newElement->name);
419 uprv_free(newElement->item);
420 uprv_free(newElement);
421 return oldValue ? oldValue->item : NULL;
422 }
423
424 return newElement->item;
425 }
426
427 /*----------------------------------------------------------------------*==============
428 * *
429 * Path management. Could be shared with other tools/etc if need be *
430 * later on. *
431 * *
432 *----------------------------------------------------------------------*/
433
434 U_NAMESPACE_BEGIN
435
436 class UDataPathIterator
437 {
438 public:
439 UDataPathIterator(const char *path, const char *pkg,
440 const char *item, const char *suffix, UBool doCheckLastFour,
441 UErrorCode *pErrorCode);
442 const char *next(UErrorCode *pErrorCode);
443
444 private:
445 const char *path; /* working path (u_icudata_Dir) */
446 const char *nextPath; /* path following this one */
447 const char *basename; /* item's basename (icudt22e_mt.res)*/
448
449 StringPiece suffix; /* item suffix (can be null) */
450
451 uint32_t basenameLen; /* length of basename */
452
453 CharString itemPath; /* path passed in with item name */
454 CharString pathBuffer; /* output path for this it'ion */
455 CharString packageStub; /* example: "/icudt28b". Will ignore that leaf in set paths. */
456
457 UBool checkLastFour; /* if TRUE then allow paths such as '/foo/myapp.dat'
458 * to match, checks last 4 chars of suffix with
459 * last 4 of path, then previous chars. */
460 };
461
462 /**
463 * @param iter The iterator to be initialized. Its current state does not matter.
464 * @param inPath The full pathname to be iterated over. If NULL, defaults to U_ICUDATA_NAME
465 * @param pkg Package which is being searched for, ex "icudt28l". Will ignore leaf directories such as /icudt28l
466 * @param item Item to be searched for. Can include full path, such as /a/b/foo.dat
467 * @param inSuffix Optional item suffix, if not-null (ex. ".dat") then 'path' can contain 'item' explicitly.
468 * Ex: 'stuff.dat' would be found in '/a/foo:/tmp/stuff.dat:/bar/baz' as item #2.
469 * '/blarg/stuff.dat' would also be found.
470 * Note: inSuffix may also be the 'item' being searched for as well, (ex: "ibm-5348_P100-1997.cnv"), in which case
471 * the 'item' parameter is often the same as pkg. (Though sometimes might have a tree part as well, ex: "icudt62l-curr").
472 */
UDataPathIterator(const char * inPath,const char * pkg,const char * item,const char * inSuffix,UBool doCheckLastFour,UErrorCode * pErrorCode)473 UDataPathIterator::UDataPathIterator(const char *inPath, const char *pkg,
474 const char *item, const char *inSuffix, UBool doCheckLastFour,
475 UErrorCode *pErrorCode)
476 {
477 #ifdef UDATA_DEBUG
478 fprintf(stderr, "SUFFIX1=%s PATH=%s\n", inSuffix, inPath);
479 #endif
480 /** Path **/
481 if(inPath == NULL) {
482 path = u_getDataDirectory();
483 } else {
484 path = inPath;
485 }
486
487 /** Package **/
488 if(pkg != NULL) {
489 packageStub.append(U_FILE_SEP_CHAR, *pErrorCode).append(pkg, *pErrorCode);
490 #ifdef UDATA_DEBUG
491 fprintf(stderr, "STUB=%s [%d]\n", packageStub.data(), packageStub.length());
492 #endif
493 }
494
495 /** Item **/
496 basename = findBasename(item);
497 basenameLen = (int32_t)uprv_strlen(basename);
498
499 /** Item path **/
500 if(basename == item) {
501 nextPath = path;
502 } else {
503 itemPath.append(item, (int32_t)(basename-item), *pErrorCode);
504 nextPath = itemPath.data();
505 }
506 #ifdef UDATA_DEBUG
507 fprintf(stderr, "SUFFIX=%s [%p]\n", inSuffix, (void*) inSuffix);
508 #endif
509
510 /** Suffix **/
511 if(inSuffix != NULL) {
512 suffix = inSuffix;
513 } else {
514 suffix = "";
515 }
516
517 checkLastFour = doCheckLastFour;
518
519 /* pathBuffer will hold the output path strings returned by this iterator */
520
521 #ifdef UDATA_DEBUG
522 fprintf(stderr, "0: init %s -> [path=%s], [base=%s], [suff=%s], [itempath=%s], [nextpath=%s], [checklast4=%s]\n",
523 item,
524 path,
525 basename,
526 suffix.data(),
527 itemPath.data(),
528 nextPath,
529 checkLastFour?"TRUE":"false");
530 #endif
531 }
532
533 /**
534 * Get the next path on the list.
535 *
536 * @param iter The Iter to be used
537 * @param len If set, pointer to the length of the returned path, for convenience.
538 * @return Pointer to the next path segment, or NULL if there are no more.
539 */
next(UErrorCode * pErrorCode)540 const char *UDataPathIterator::next(UErrorCode *pErrorCode)
541 {
542 if(U_FAILURE(*pErrorCode)) {
543 return NULL;
544 }
545
546 const char *currentPath = NULL;
547 int32_t pathLen = 0;
548 const char *pathBasename;
549
550 do
551 {
552 if( nextPath == NULL ) {
553 break;
554 }
555 currentPath = nextPath;
556
557 if(nextPath == itemPath.data()) { /* we were processing item's path. */
558 nextPath = path; /* start with regular path next tm. */
559 pathLen = (int32_t)uprv_strlen(currentPath);
560 } else {
561 /* fix up next for next time */
562 nextPath = uprv_strchr(currentPath, U_PATH_SEP_CHAR);
563 if(nextPath == NULL) {
564 /* segment: entire path */
565 pathLen = (int32_t)uprv_strlen(currentPath);
566 } else {
567 /* segment: until next segment */
568 pathLen = (int32_t)(nextPath - currentPath);
569 /* skip divider */
570 nextPath ++;
571 }
572 }
573
574 if(pathLen == 0) {
575 continue;
576 }
577
578 #ifdef UDATA_DEBUG
579 fprintf(stderr, "rest of path (IDD) = %s\n", currentPath);
580 fprintf(stderr, " ");
581 {
582 int32_t qqq;
583 for(qqq=0;qqq<pathLen;qqq++)
584 {
585 fprintf(stderr, " ");
586 }
587
588 fprintf(stderr, "^\n");
589 }
590 #endif
591 pathBuffer.clear().append(currentPath, pathLen, *pErrorCode);
592
593 /* check for .dat files */
594 pathBasename = findBasename(pathBuffer.data());
595
596 if(checkLastFour == TRUE &&
597 (pathLen>=4) &&
598 uprv_strncmp(pathBuffer.data() +(pathLen-4), suffix.data(), 4)==0 && /* suffix matches */
599 uprv_strncmp(findBasename(pathBuffer.data()), basename, basenameLen)==0 && /* base matches */
600 uprv_strlen(pathBasename)==(basenameLen+4)) { /* base+suffix = full len */
601
602 #ifdef UDATA_DEBUG
603 fprintf(stderr, "Have %s file on the path: %s\n", suffix.data(), pathBuffer.data());
604 #endif
605 /* do nothing */
606 }
607 else
608 { /* regular dir path */
609 if(pathBuffer[pathLen-1] != U_FILE_SEP_CHAR) {
610 if((pathLen>=4) &&
611 uprv_strncmp(pathBuffer.data()+(pathLen-4), ".dat", 4) == 0)
612 {
613 #ifdef UDATA_DEBUG
614 fprintf(stderr, "skipping non-directory .dat file %s\n", pathBuffer.data());
615 #endif
616 continue;
617 }
618
619 /* Check if it is a directory with the same name as our package */
620 if(!packageStub.isEmpty() &&
621 (pathLen > packageStub.length()) &&
622 !uprv_strcmp(pathBuffer.data() + pathLen - packageStub.length(), packageStub.data())) {
623 #ifdef UDATA_DEBUG
624 fprintf(stderr, "Found stub %s (will add package %s of len %d)\n", packageStub.data(), basename, basenameLen);
625 #endif
626 pathBuffer.truncate(pathLen - packageStub.length());
627 }
628 pathBuffer.append(U_FILE_SEP_CHAR, *pErrorCode);
629 }
630
631 /* + basename */
632 pathBuffer.append(packageStub.data()+1, packageStub.length()-1, *pErrorCode);
633
634 if (!suffix.empty()) /* tack on suffix */
635 {
636 if (suffix.length() > 4) {
637 // If the suffix is actually an item ("ibm-5348_P100-1997.cnv") and not an extension (".res")
638 // then we need to ensure that the path ends with a separator.
639 pathBuffer.ensureEndsWithFileSeparator(*pErrorCode);
640 }
641 pathBuffer.append(suffix, *pErrorCode);
642 }
643 }
644
645 #ifdef UDATA_DEBUG
646 fprintf(stderr, " --> %s\n", pathBuffer.data());
647 #endif
648
649 return pathBuffer.data();
650
651 } while(path);
652
653 /* fell way off the end */
654 return NULL;
655 }
656
657 U_NAMESPACE_END
658
659 /* ==================================================================================*/
660
661
662 /*----------------------------------------------------------------------*
663 * *
664 * Add a static reference to the common data library *
665 * Unless overridden by an explicit udata_setCommonData, this will be *
666 * our common data. *
667 * *
668 *----------------------------------------------------------------------*/
669 #if !defined(ICU_DATA_DIR_WINDOWS)
670 // When using the Windows system data, we expect only a single data file.
671 extern "C" const DataHeader U_DATA_API U_ICUDATA_ENTRY_POINT;
672 #endif
673
674 /*
675 * This would be a good place for weak-linkage declarations of
676 * partial-data-library access functions where each returns a pointer
677 * to its data package, if it is linked in.
678 */
679 /*
680 extern const void *uprv_getICUData_collation(void) ATTRIBUTE_WEAK;
681 extern const void *uprv_getICUData_conversion(void) ATTRIBUTE_WEAK;
682 */
683
684 /*----------------------------------------------------------------------*
685 * *
686 * openCommonData Attempt to open a common format (.dat) file *
687 * Map it into memory (if it's not there already) *
688 * and return a UDataMemory object for it. *
689 * *
690 * If the requested data is already open and cached *
691 * just return the cached UDataMem object. *
692 * *
693 *----------------------------------------------------------------------*/
694 static UDataMemory *
openCommonData(const char * path,int32_t commonDataIndex,UErrorCode * pErrorCode)695 openCommonData(const char *path, /* Path from OpenChoice? */
696 int32_t commonDataIndex, /* ICU Data (index >= 0) if path == NULL */
697 UErrorCode *pErrorCode)
698 {
699 UDataMemory tData;
700 const char *pathBuffer;
701 const char *inBasename;
702
703 if (U_FAILURE(*pErrorCode)) {
704 return NULL;
705 }
706
707 UDataMemory_init(&tData);
708
709 /* ??????? TODO revisit this */
710 if (commonDataIndex >= 0) {
711 /* "mini-cache" for common ICU data */
712 if(commonDataIndex >= UPRV_LENGTHOF(gCommonICUDataArray)) {
713 return NULL;
714 }
715 {
716 Mutex lock;
717 if(gCommonICUDataArray[commonDataIndex] != NULL) {
718 return gCommonICUDataArray[commonDataIndex];
719 }
720 #if !defined(ICU_DATA_DIR_WINDOWS)
721 // When using the Windows system data, we expect only a single data file.
722 int32_t i;
723 for(i = 0; i < commonDataIndex; ++i) {
724 if(gCommonICUDataArray[i]->pHeader == &U_ICUDATA_ENTRY_POINT) {
725 /* The linked-in data is already in the list. */
726 return NULL;
727 }
728 }
729 #endif
730 }
731
732 // BEGIN Android-added: Use specialized libandroidicuinit to load the data on Android/ART host.
733 #ifdef AOSP_ICU_INIT // Do nothing on other platforms, e.g. Windows
734 // android_icu_init() is only called once.
735 umtx_initOnce(gAospInitOnce, &android_icu_init);
736 #endif // AOSP_ICU_INIT
737 // END Android-added: Use specialized libandroidicuinit to load the data on Android/ART host.
738
739 /* Add the linked-in data to the list. */
740 /*
741 * This is where we would check and call weakly linked partial-data-library
742 * access functions.
743 */
744 /*
745 if (uprv_getICUData_collation) {
746 setCommonICUDataPointer(uprv_getICUData_collation(), FALSE, pErrorCode);
747 }
748 if (uprv_getICUData_conversion) {
749 setCommonICUDataPointer(uprv_getICUData_conversion(), FALSE, pErrorCode);
750 }
751 */
752 #if !defined(ICU_DATA_DIR_WINDOWS)
753 // When using the Windows system data, we expect only a single data file.
754 setCommonICUDataPointer(&U_ICUDATA_ENTRY_POINT, FALSE, pErrorCode);
755 {
756 Mutex lock;
757 return gCommonICUDataArray[commonDataIndex];
758 }
759 #endif
760 }
761
762
763 /* request is NOT for ICU Data. */
764
765 /* Find the base name portion of the supplied path. */
766 /* inBasename will be left pointing somewhere within the original path string. */
767 inBasename = findBasename(path);
768 #ifdef UDATA_DEBUG
769 fprintf(stderr, "inBasename = %s\n", inBasename);
770 #endif
771
772 if(*inBasename==0) {
773 /* no basename. This will happen if the original path was a directory name, */
774 /* like "a/b/c/". (Fallback to separate files will still work.) */
775 #ifdef UDATA_DEBUG
776 fprintf(stderr, "ocd: no basename in %s, bailing.\n", path);
777 #endif
778 if (U_SUCCESS(*pErrorCode)) {
779 *pErrorCode=U_FILE_ACCESS_ERROR;
780 }
781 return NULL;
782 }
783
784 /* Is the requested common data file already open and cached? */
785 /* Note that the cache is keyed by the base name only. The rest of the path, */
786 /* if any, is not considered. */
787 UDataMemory *dataToReturn = udata_findCachedData(inBasename, *pErrorCode);
788 if (dataToReturn != NULL || U_FAILURE(*pErrorCode)) {
789 return dataToReturn;
790 }
791
792 /* Requested item is not in the cache.
793 * Hunt it down, trying all the path locations
794 */
795
796 UDataPathIterator iter(u_getDataDirectory(), inBasename, path, ".dat", TRUE, pErrorCode);
797
798 while ((UDataMemory_isLoaded(&tData)==FALSE) && (pathBuffer = iter.next(pErrorCode)) != NULL)
799 {
800 #ifdef UDATA_DEBUG
801 fprintf(stderr, "ocd: trying path %s - ", pathBuffer);
802 #endif
803 uprv_mapFile(&tData, pathBuffer, pErrorCode);
804 #ifdef UDATA_DEBUG
805 fprintf(stderr, "%s\n", UDataMemory_isLoaded(&tData)?"LOADED":"not loaded");
806 #endif
807 }
808 if (U_FAILURE(*pErrorCode)) {
809 return NULL;
810 }
811
812 #if defined(OS390_STUBDATA) && defined(OS390BATCH)
813 if (!UDataMemory_isLoaded(&tData)) {
814 char ourPathBuffer[1024];
815 /* One more chance, for extendCommonData() */
816 uprv_strncpy(ourPathBuffer, path, 1019);
817 ourPathBuffer[1019]=0;
818 uprv_strcat(ourPathBuffer, ".dat");
819 uprv_mapFile(&tData, ourPathBuffer, pErrorCode);
820 }
821 #endif
822
823 if (U_FAILURE(*pErrorCode)) {
824 return NULL;
825 }
826 if (!UDataMemory_isLoaded(&tData)) {
827 /* no common data */
828 *pErrorCode=U_FILE_ACCESS_ERROR;
829 return NULL;
830 }
831
832 /* we have mapped a file, check its header */
833 udata_checkCommonData(&tData, pErrorCode);
834
835
836 /* Cache the UDataMemory struct for this .dat file,
837 * so we won't need to hunt it down and map it again next time
838 * something is needed from it. */
839 return udata_cacheDataItem(inBasename, &tData, pErrorCode);
840 }
841
842
843 /*----------------------------------------------------------------------*
844 * *
845 * extendICUData If the full set of ICU data was not loaded at *
846 * program startup, load it now. This function will *
847 * be called when the lookup of an ICU data item in *
848 * the common ICU data fails. *
849 * *
850 * return true if new data is loaded, false otherwise.*
851 * *
852 *----------------------------------------------------------------------*/
extendICUData(UErrorCode * pErr)853 static UBool extendICUData(UErrorCode *pErr)
854 {
855 UDataMemory *pData;
856 UDataMemory copyPData;
857 UBool didUpdate = FALSE;
858
859 /*
860 * There is a chance for a race condition here.
861 * Normally, ICU data is loaded from a DLL or via mmap() and
862 * setCommonICUData() will detect if the same address is set twice.
863 * If ICU is built with data loading via fread() then the address will
864 * be different each time the common data is loaded and we may add
865 * multiple copies of the data.
866 * In this case, use a mutex to prevent the race.
867 * Use a specific mutex to avoid nested locks of the global mutex.
868 */
869 #if MAP_IMPLEMENTATION==MAP_STDIO
870 static UMutex extendICUDataMutex;
871 umtx_lock(&extendICUDataMutex);
872 #endif
873 if(!umtx_loadAcquire(gHaveTriedToLoadCommonData)) {
874 /* See if we can explicitly open a .dat file for the ICUData. */
875 pData = openCommonData(
876 U_ICUDATA_NAME, /* "icudt20l" , for example. */
877 -1, /* Pretend we're not opening ICUData */
878 pErr);
879
880 /* How about if there is no pData, eh... */
881
882 UDataMemory_init(©PData);
883 if(pData != NULL) {
884 UDatamemory_assign(©PData, pData);
885 copyPData.map = 0; /* The mapping for this data is owned by the hash table */
886 copyPData.mapAddr = 0; /* which will unmap it when ICU is shut down. */
887 /* CommonICUData is also unmapped when ICU is shut down.*/
888 /* To avoid unmapping the data twice, zero out the map */
889 /* fields in the UDataMemory that we're assigning */
890 /* to CommonICUData. */
891
892 didUpdate = /* no longer using this result */
893 setCommonICUData(©PData,/* The new common data. */
894 FALSE, /* No warnings if write didn't happen */
895 pErr); /* setCommonICUData honors errors; NOP if error set */
896 }
897
898 umtx_storeRelease(gHaveTriedToLoadCommonData, 1);
899 }
900
901 didUpdate = findCommonICUDataByName(U_ICUDATA_NAME, *pErr); /* Return 'true' when a racing writes out the extended */
902 /* data after another thread has failed to see it (in openCommonData), so */
903 /* extended data can be examined. */
904 /* Also handles a race through here before gHaveTriedToLoadCommonData is set. */
905
906 #if MAP_IMPLEMENTATION==MAP_STDIO
907 umtx_unlock(&extendICUDataMutex);
908 #endif
909 return didUpdate; /* Return true if ICUData pointer was updated. */
910 /* (Could potentially have been done by another thread racing */
911 /* us through here, but that's fine, we still return true */
912 /* so that current thread will also examine extended data. */
913 }
914
915 /*----------------------------------------------------------------------*
916 * *
917 * udata_setCommonData *
918 * *
919 *----------------------------------------------------------------------*/
920 U_CAPI void U_EXPORT2
udata_setCommonData(const void * data,UErrorCode * pErrorCode)921 udata_setCommonData(const void *data, UErrorCode *pErrorCode) {
922 UDataMemory dataMemory;
923
924 if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
925 return;
926 }
927
928 if(data==NULL) {
929 *pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
930 return;
931 }
932
933 /* set the data pointer and test for validity */
934 UDataMemory_init(&dataMemory);
935 UDataMemory_setData(&dataMemory, data);
936 udata_checkCommonData(&dataMemory, pErrorCode);
937 if (U_FAILURE(*pErrorCode)) {return;}
938
939 /* we have good data */
940 /* Set it up as the ICU Common Data. */
941 setCommonICUData(&dataMemory, TRUE, pErrorCode);
942 }
943
944 /*---------------------------------------------------------------------------
945 *
946 * udata_setAppData
947 *
948 *---------------------------------------------------------------------------- */
949 U_CAPI void U_EXPORT2
udata_setAppData(const char * path,const void * data,UErrorCode * err)950 udata_setAppData(const char *path, const void *data, UErrorCode *err)
951 {
952 UDataMemory udm;
953
954 if(err==NULL || U_FAILURE(*err)) {
955 return;
956 }
957 if(data==NULL) {
958 *err=U_ILLEGAL_ARGUMENT_ERROR;
959 return;
960 }
961
962 UDataMemory_init(&udm);
963 UDataMemory_setData(&udm, data);
964 udata_checkCommonData(&udm, err);
965 udata_cacheDataItem(path, &udm, err);
966 }
967
968 /*----------------------------------------------------------------------------*
969 * *
970 * checkDataItem Given a freshly located/loaded data item, either *
971 * an entry in a common file or a separately loaded file, *
972 * sanity check its header, and see if the data is *
973 * acceptable to the app. *
974 * If the data is good, create and return a UDataMemory *
975 * object that can be returned to the application. *
976 * Return NULL on any sort of failure. *
977 * *
978 *----------------------------------------------------------------------------*/
979 static UDataMemory *
checkDataItem(const DataHeader * pHeader,UDataMemoryIsAcceptable * isAcceptable,void * context,const char * type,const char * name,UErrorCode * nonFatalErr,UErrorCode * fatalErr)980 checkDataItem
981 (
982 const DataHeader *pHeader, /* The data item to be checked. */
983 UDataMemoryIsAcceptable *isAcceptable, /* App's call-back function */
984 void *context, /* pass-thru param for above. */
985 const char *type, /* pass-thru param for above. */
986 const char *name, /* pass-thru param for above. */
987 UErrorCode *nonFatalErr, /* Error code if this data was not acceptable */
988 /* but openChoice should continue with */
989 /* trying to get data from fallback path. */
990 UErrorCode *fatalErr /* Bad error, caller should return immediately */
991 )
992 {
993 UDataMemory *rDataMem = NULL; /* the new UDataMemory, to be returned. */
994
995 if (U_FAILURE(*fatalErr)) {
996 return NULL;
997 }
998
999 if(pHeader->dataHeader.magic1==0xda &&
1000 pHeader->dataHeader.magic2==0x27 &&
1001 (isAcceptable==NULL || isAcceptable(context, type, name, &pHeader->info))
1002 ) {
1003 rDataMem=UDataMemory_createNewInstance(fatalErr);
1004 if (U_FAILURE(*fatalErr)) {
1005 return NULL;
1006 }
1007 rDataMem->pHeader = pHeader;
1008 } else {
1009 /* the data is not acceptable, look further */
1010 /* If we eventually find something good, this errorcode will be */
1011 /* cleared out. */
1012 *nonFatalErr=U_INVALID_FORMAT_ERROR;
1013 }
1014 return rDataMem;
1015 }
1016
1017 /**
1018 * @return 0 if not loaded, 1 if loaded or err
1019 */
doLoadFromIndividualFiles(const char * pkgName,const char * dataPath,const char * tocEntryPathSuffix,const char * path,const char * type,const char * name,UDataMemoryIsAcceptable * isAcceptable,void * context,UErrorCode * subErrorCode,UErrorCode * pErrorCode)1020 static UDataMemory *doLoadFromIndividualFiles(const char *pkgName,
1021 const char *dataPath, const char *tocEntryPathSuffix,
1022 /* following arguments are the same as doOpenChoice itself */
1023 const char *path, const char *type, const char *name,
1024 UDataMemoryIsAcceptable *isAcceptable, void *context,
1025 UErrorCode *subErrorCode,
1026 UErrorCode *pErrorCode)
1027 {
1028 const char *pathBuffer;
1029 UDataMemory dataMemory;
1030 UDataMemory *pEntryData;
1031
1032 /* look in ind. files: package\nam.typ ========================= */
1033 /* init path iterator for individual files */
1034 UDataPathIterator iter(dataPath, pkgName, path, tocEntryPathSuffix, FALSE, pErrorCode);
1035
1036 while ((pathBuffer = iter.next(pErrorCode)) != NULL)
1037 {
1038 #ifdef UDATA_DEBUG
1039 fprintf(stderr, "UDATA: trying individual file %s\n", pathBuffer);
1040 #endif
1041 if (uprv_mapFile(&dataMemory, pathBuffer, pErrorCode))
1042 {
1043 pEntryData = checkDataItem(dataMemory.pHeader, isAcceptable, context, type, name, subErrorCode, pErrorCode);
1044 if (pEntryData != NULL) {
1045 /* Data is good.
1046 * Hand off ownership of the backing memory to the user's UDataMemory.
1047 * and return it. */
1048 pEntryData->mapAddr = dataMemory.mapAddr;
1049 pEntryData->map = dataMemory.map;
1050
1051 #ifdef UDATA_DEBUG
1052 fprintf(stderr, "** Mapped file: %s\n", pathBuffer);
1053 #endif
1054 return pEntryData;
1055 }
1056
1057 /* the data is not acceptable, or some error occurred. Either way, unmap the memory */
1058 udata_close(&dataMemory);
1059
1060 /* If we had a nasty error, bail out completely. */
1061 if (U_FAILURE(*pErrorCode)) {
1062 return NULL;
1063 }
1064
1065 /* Otherwise remember that we found data but didn't like it for some reason */
1066 *subErrorCode=U_INVALID_FORMAT_ERROR;
1067 }
1068 #ifdef UDATA_DEBUG
1069 fprintf(stderr, "%s\n", UDataMemory_isLoaded(&dataMemory)?"LOADED":"not loaded");
1070 #endif
1071 }
1072 return NULL;
1073 }
1074
1075 /**
1076 * @return 0 if not loaded, 1 if loaded or err
1077 */
doLoadFromCommonData(UBool isICUData,const char *,const char *,const char *,const char * tocEntryName,const char * path,const char * type,const char * name,UDataMemoryIsAcceptable * isAcceptable,void * context,UErrorCode * subErrorCode,UErrorCode * pErrorCode)1078 static UDataMemory *doLoadFromCommonData(UBool isICUData, const char * /*pkgName*/,
1079 const char * /*dataPath*/, const char * /*tocEntryPathSuffix*/, const char *tocEntryName,
1080 /* following arguments are the same as doOpenChoice itself */
1081 const char *path, const char *type, const char *name,
1082 UDataMemoryIsAcceptable *isAcceptable, void *context,
1083 UErrorCode *subErrorCode,
1084 UErrorCode *pErrorCode)
1085 {
1086 UDataMemory *pEntryData;
1087 const DataHeader *pHeader;
1088 UDataMemory *pCommonData;
1089 int32_t commonDataIndex;
1090 UBool checkedExtendedICUData = FALSE;
1091 /* try to get common data. The loop is for platforms such as the 390 that do
1092 * not initially load the full set of ICU data. If the lookup of an ICU data item
1093 * fails, the full (but slower to load) set is loaded, the and the loop repeats,
1094 * trying the lookup again. Once the full set of ICU data is loaded, the loop wont
1095 * repeat because the full set will be checked the first time through.
1096 *
1097 * The loop also handles the fallback to a .dat file if the application linked
1098 * to the stub data library rather than a real library.
1099 */
1100 for (commonDataIndex = isICUData ? 0 : -1;;) {
1101 pCommonData=openCommonData(path, commonDataIndex, subErrorCode); /** search for pkg **/
1102
1103 if(U_SUCCESS(*subErrorCode) && pCommonData!=NULL) {
1104 int32_t length;
1105
1106 /* look up the data piece in the common data */
1107 pHeader=pCommonData->vFuncs->Lookup(pCommonData, tocEntryName, &length, subErrorCode);
1108 #ifdef UDATA_DEBUG
1109 fprintf(stderr, "%s: pHeader=%p - %s\n", tocEntryName, (void*) pHeader, u_errorName(*subErrorCode));
1110 #endif
1111
1112 if(pHeader!=NULL) {
1113 pEntryData = checkDataItem(pHeader, isAcceptable, context, type, name, subErrorCode, pErrorCode);
1114 #ifdef UDATA_DEBUG
1115 fprintf(stderr, "pEntryData=%p\n", (void*) pEntryData);
1116 #endif
1117 if (U_FAILURE(*pErrorCode)) {
1118 return NULL;
1119 }
1120 if (pEntryData != NULL) {
1121 pEntryData->length = length;
1122 return pEntryData;
1123 }
1124 }
1125 }
1126 // If we failed due to being out-of-memory, then stop early and report the error.
1127 if (*subErrorCode == U_MEMORY_ALLOCATION_ERROR) {
1128 *pErrorCode = *subErrorCode;
1129 return NULL;
1130 }
1131 /* Data wasn't found. If we were looking for an ICUData item and there is
1132 * more data available, load it and try again,
1133 * otherwise break out of this loop. */
1134 if (!isICUData) {
1135 return NULL;
1136 } else if (pCommonData != NULL) {
1137 ++commonDataIndex; /* try the next data package */
1138 } else if ((!checkedExtendedICUData) && extendICUData(subErrorCode)) {
1139 checkedExtendedICUData = TRUE;
1140 /* try this data package slot again: it changed from NULL to non-NULL */
1141 } else {
1142 return NULL;
1143 }
1144 }
1145 }
1146
1147 /*
1148 * Identify the Time Zone resources that are subject to special override data loading.
1149 */
isTimeZoneFile(const char * name,const char * type)1150 static UBool isTimeZoneFile(const char *name, const char *type) {
1151 return ((uprv_strcmp(type, "res") == 0) &&
1152 (uprv_strcmp(name, "zoneinfo64") == 0 ||
1153 uprv_strcmp(name, "timezoneTypes") == 0 ||
1154 uprv_strcmp(name, "windowsZones") == 0 ||
1155 uprv_strcmp(name, "metaZones") == 0));
1156 }
1157
1158 /*
1159 * A note on the ownership of Mapped Memory
1160 *
1161 * For common format files, ownership resides with the UDataMemory object
1162 * that lives in the cache of opened common data. These UDataMemorys are private
1163 * to the udata implementation, and are never seen directly by users.
1164 *
1165 * The UDataMemory objects returned to users will have the address of some desired
1166 * data within the mapped region, but they wont have the mapping info itself, and thus
1167 * won't cause anything to be removed from memory when they are closed.
1168 *
1169 * For individual data files, the UDataMemory returned to the user holds the
1170 * information necessary to unmap the data on close. If the user independently
1171 * opens the same data file twice, two completely independent mappings will be made.
1172 * (There is no cache of opened data items from individual files, only a cache of
1173 * opened Common Data files, that is, files containing a collection of data items.)
1174 *
1175 * For common data passed in from the user via udata_setAppData() or
1176 * udata_setCommonData(), ownership remains with the user.
1177 *
1178 * UDataMemory objects themselves, as opposed to the memory they describe,
1179 * can be anywhere - heap, stack/local or global.
1180 * They have a flag to indicate when they're heap allocated and thus
1181 * must be deleted when closed.
1182 */
1183
1184
1185 /*----------------------------------------------------------------------------*
1186 * *
1187 * main data loading functions *
1188 * *
1189 *----------------------------------------------------------------------------*/
1190 static UDataMemory *
doOpenChoice(const char * path,const char * type,const char * name,UDataMemoryIsAcceptable * isAcceptable,void * context,UErrorCode * pErrorCode)1191 doOpenChoice(const char *path, const char *type, const char *name,
1192 UDataMemoryIsAcceptable *isAcceptable, void *context,
1193 UErrorCode *pErrorCode)
1194 {
1195 UDataMemory *retVal = NULL;
1196
1197 const char *dataPath;
1198
1199 int32_t tocEntrySuffixIndex;
1200 const char *tocEntryPathSuffix;
1201 UErrorCode subErrorCode=U_ZERO_ERROR;
1202 const char *treeChar;
1203
1204 UBool isICUData = FALSE;
1205
1206
1207 FileTracer::traceOpen(path, type, name);
1208
1209
1210 /* Is this path ICU data? */
1211 if(path == NULL ||
1212 !strcmp(path, U_ICUDATA_ALIAS) || /* "ICUDATA" */
1213 !uprv_strncmp(path, U_ICUDATA_NAME U_TREE_SEPARATOR_STRING, /* "icudt26e-" */
1214 uprv_strlen(U_ICUDATA_NAME U_TREE_SEPARATOR_STRING)) ||
1215 !uprv_strncmp(path, U_ICUDATA_ALIAS U_TREE_SEPARATOR_STRING, /* "ICUDATA-" */
1216 uprv_strlen(U_ICUDATA_ALIAS U_TREE_SEPARATOR_STRING))) {
1217 isICUData = TRUE;
1218 }
1219
1220 #if (U_FILE_SEP_CHAR != U_FILE_ALT_SEP_CHAR) /* Windows: try "foo\bar" and "foo/bar" */
1221 /* remap from alternate path char to the main one */
1222 CharString altSepPath;
1223 if(path) {
1224 if(uprv_strchr(path,U_FILE_ALT_SEP_CHAR) != NULL) {
1225 altSepPath.append(path, *pErrorCode);
1226 char *p;
1227 while ((p = uprv_strchr(altSepPath.data(), U_FILE_ALT_SEP_CHAR)) != NULL) {
1228 *p = U_FILE_SEP_CHAR;
1229 }
1230 #if defined (UDATA_DEBUG)
1231 fprintf(stderr, "Changed path from [%s] to [%s]\n", path, altSepPath.s);
1232 #endif
1233 path = altSepPath.data();
1234 }
1235 }
1236 #endif
1237
1238 CharString tocEntryName; /* entry name in tree format. ex: 'icudt28b/coll/ar.res' */
1239 CharString tocEntryPath; /* entry name in path format. ex: 'icudt28b\\coll\\ar.res' */
1240
1241 CharString pkgName;
1242 CharString treeName;
1243
1244 /* ======= Set up strings */
1245 if(path==NULL) {
1246 pkgName.append(U_ICUDATA_NAME, *pErrorCode);
1247 } else {
1248 const char *pkg;
1249 const char *first;
1250 pkg = uprv_strrchr(path, U_FILE_SEP_CHAR);
1251 first = uprv_strchr(path, U_FILE_SEP_CHAR);
1252 if(uprv_pathIsAbsolute(path) || (pkg != first)) { /* more than one slash in the path- not a tree name */
1253 /* see if this is an /absolute/path/to/package path */
1254 if(pkg) {
1255 pkgName.append(pkg+1, *pErrorCode);
1256 } else {
1257 pkgName.append(path, *pErrorCode);
1258 }
1259 } else {
1260 treeChar = uprv_strchr(path, U_TREE_SEPARATOR);
1261 if(treeChar) {
1262 treeName.append(treeChar+1, *pErrorCode); /* following '-' */
1263 if(isICUData) {
1264 pkgName.append(U_ICUDATA_NAME, *pErrorCode);
1265 } else {
1266 pkgName.append(path, (int32_t)(treeChar-path), *pErrorCode);
1267 if (first == NULL) {
1268 /*
1269 This user data has no path, but there is a tree name.
1270 Look up the correct path from the data cache later.
1271 */
1272 path = pkgName.data();
1273 }
1274 }
1275 } else {
1276 if(isICUData) {
1277 pkgName.append(U_ICUDATA_NAME, *pErrorCode);
1278 } else {
1279 pkgName.append(path, *pErrorCode);
1280 }
1281 }
1282 }
1283 }
1284
1285 #ifdef UDATA_DEBUG
1286 fprintf(stderr, " P=%s T=%s\n", pkgName.data(), treeName.data());
1287 #endif
1288
1289 /* setting up the entry name and file name
1290 * Make up a full name by appending the type to the supplied
1291 * name, assuming that a type was supplied.
1292 */
1293
1294 /* prepend the package */
1295 tocEntryName.append(pkgName, *pErrorCode);
1296 tocEntryPath.append(pkgName, *pErrorCode);
1297 tocEntrySuffixIndex = tocEntryName.length();
1298
1299 if(!treeName.isEmpty()) {
1300 tocEntryName.append(U_TREE_ENTRY_SEP_CHAR, *pErrorCode).append(treeName, *pErrorCode);
1301 tocEntryPath.append(U_FILE_SEP_CHAR, *pErrorCode).append(treeName, *pErrorCode);
1302 }
1303
1304 tocEntryName.append(U_TREE_ENTRY_SEP_CHAR, *pErrorCode).append(name, *pErrorCode);
1305 tocEntryPath.append(U_FILE_SEP_CHAR, *pErrorCode).append(name, *pErrorCode);
1306 if(type!=NULL && *type!=0) {
1307 tocEntryName.append(".", *pErrorCode).append(type, *pErrorCode);
1308 tocEntryPath.append(".", *pErrorCode).append(type, *pErrorCode);
1309 }
1310 // The +1 is for the U_FILE_SEP_CHAR that is always appended above.
1311 tocEntryPathSuffix = tocEntryPath.data() + tocEntrySuffixIndex + 1; /* suffix starts here */
1312
1313 #ifdef UDATA_DEBUG
1314 fprintf(stderr, " tocEntryName = %s\n", tocEntryName.data());
1315 fprintf(stderr, " tocEntryPath = %s\n", tocEntryName.data());
1316 #endif
1317
1318 #if !defined(ICU_DATA_DIR_WINDOWS)
1319 if(path == NULL) {
1320 path = COMMON_DATA_NAME; /* "icudt26e" */
1321 }
1322 #else
1323 // When using the Windows system data, we expects only a single data file.
1324 path = COMMON_DATA_NAME; /* "icudt26e" */
1325 #endif
1326
1327 /************************ Begin loop looking for ind. files ***************/
1328 #ifdef UDATA_DEBUG
1329 fprintf(stderr, "IND: inBasename = %s, pkg=%s\n", "(n/a)", packageNameFromPath(path));
1330 #endif
1331
1332 /* End of dealing with a null basename */
1333 dataPath = u_getDataDirectory();
1334
1335 /**** Time zone individual files override */
1336 if (isICUData && isTimeZoneFile(name, type)) {
1337 const char *tzFilesDir = u_getTimeZoneFilesDirectory(pErrorCode);
1338 if (tzFilesDir[0] != 0) {
1339 #ifdef UDATA_DEBUG
1340 fprintf(stderr, "Trying Time Zone Files directory = %s\n", tzFilesDir);
1341 #endif
1342 retVal = doLoadFromIndividualFiles(/* pkgName.data() */ "", tzFilesDir, tocEntryPathSuffix,
1343 /* path */ "", type, name, isAcceptable, context, &subErrorCode, pErrorCode);
1344 if((retVal != NULL) || U_FAILURE(*pErrorCode)) {
1345 return retVal;
1346 }
1347 }
1348 }
1349
1350 /**** COMMON PACKAGE - only if packages are first. */
1351 if(gDataFileAccess == UDATA_PACKAGES_FIRST) {
1352 #ifdef UDATA_DEBUG
1353 fprintf(stderr, "Trying packages (UDATA_PACKAGES_FIRST)\n");
1354 #endif
1355 /* #2 */
1356 retVal = doLoadFromCommonData(isICUData,
1357 pkgName.data(), dataPath, tocEntryPathSuffix, tocEntryName.data(),
1358 path, type, name, isAcceptable, context, &subErrorCode, pErrorCode);
1359 if((retVal != NULL) || U_FAILURE(*pErrorCode)) {
1360 return retVal;
1361 }
1362 }
1363
1364 /**** INDIVIDUAL FILES */
1365 if((gDataFileAccess==UDATA_PACKAGES_FIRST) ||
1366 (gDataFileAccess==UDATA_FILES_FIRST)) {
1367 #ifdef UDATA_DEBUG
1368 fprintf(stderr, "Trying individual files\n");
1369 #endif
1370 /* Check to make sure that there is a dataPath to iterate over */
1371 if ((dataPath && *dataPath) || !isICUData) {
1372 retVal = doLoadFromIndividualFiles(pkgName.data(), dataPath, tocEntryPathSuffix,
1373 path, type, name, isAcceptable, context, &subErrorCode, pErrorCode);
1374 if((retVal != NULL) || U_FAILURE(*pErrorCode)) {
1375 return retVal;
1376 }
1377 }
1378 }
1379
1380 /**** COMMON PACKAGE */
1381 if((gDataFileAccess==UDATA_ONLY_PACKAGES) ||
1382 (gDataFileAccess==UDATA_FILES_FIRST)) {
1383 #ifdef UDATA_DEBUG
1384 fprintf(stderr, "Trying packages (UDATA_ONLY_PACKAGES || UDATA_FILES_FIRST)\n");
1385 #endif
1386 retVal = doLoadFromCommonData(isICUData,
1387 pkgName.data(), dataPath, tocEntryPathSuffix, tocEntryName.data(),
1388 path, type, name, isAcceptable, context, &subErrorCode, pErrorCode);
1389 if((retVal != NULL) || U_FAILURE(*pErrorCode)) {
1390 return retVal;
1391 }
1392 }
1393
1394 /* Load from DLL. If we haven't attempted package load, we also haven't had any chance to
1395 try a DLL (static or setCommonData/etc) load.
1396 If we ever have a "UDATA_ONLY_FILES", add it to the or list here. */
1397 if(gDataFileAccess==UDATA_NO_FILES) {
1398 #ifdef UDATA_DEBUG
1399 fprintf(stderr, "Trying common data (UDATA_NO_FILES)\n");
1400 #endif
1401 retVal = doLoadFromCommonData(isICUData,
1402 pkgName.data(), "", tocEntryPathSuffix, tocEntryName.data(),
1403 path, type, name, isAcceptable, context, &subErrorCode, pErrorCode);
1404 if((retVal != NULL) || U_FAILURE(*pErrorCode)) {
1405 return retVal;
1406 }
1407 }
1408
1409 /* data not found */
1410 if(U_SUCCESS(*pErrorCode)) {
1411 if(U_SUCCESS(subErrorCode)) {
1412 /* file not found */
1413 *pErrorCode=U_FILE_ACCESS_ERROR;
1414 } else {
1415 /* entry point not found or rejected */
1416 *pErrorCode=subErrorCode;
1417 }
1418 }
1419 return retVal;
1420 }
1421
1422
1423
1424 /* API ---------------------------------------------------------------------- */
1425
1426 U_CAPI UDataMemory * U_EXPORT2
udata_open(const char * path,const char * type,const char * name,UErrorCode * pErrorCode)1427 udata_open(const char *path, const char *type, const char *name,
1428 UErrorCode *pErrorCode) {
1429 #ifdef UDATA_DEBUG
1430 fprintf(stderr, "udata_open(): Opening: %s : %s . %s\n", (path?path:"NULL"), name, type);
1431 fflush(stderr);
1432 #endif
1433
1434 if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
1435 return NULL;
1436 } else if(name==NULL || *name==0) {
1437 *pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
1438 return NULL;
1439 } else {
1440 return doOpenChoice(path, type, name, NULL, NULL, pErrorCode);
1441 }
1442 }
1443
1444
1445
1446 U_CAPI UDataMemory * U_EXPORT2
udata_openChoice(const char * path,const char * type,const char * name,UDataMemoryIsAcceptable * isAcceptable,void * context,UErrorCode * pErrorCode)1447 udata_openChoice(const char *path, const char *type, const char *name,
1448 UDataMemoryIsAcceptable *isAcceptable, void *context,
1449 UErrorCode *pErrorCode) {
1450 #ifdef UDATA_DEBUG
1451 fprintf(stderr, "udata_openChoice(): Opening: %s : %s . %s\n", (path?path:"NULL"), name, type);
1452 #endif
1453
1454 if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
1455 return NULL;
1456 } else if(name==NULL || *name==0 || isAcceptable==NULL) {
1457 *pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
1458 return NULL;
1459 } else {
1460 return doOpenChoice(path, type, name, isAcceptable, context, pErrorCode);
1461 }
1462 }
1463
1464
1465
1466 U_CAPI void U_EXPORT2
udata_getInfo(UDataMemory * pData,UDataInfo * pInfo)1467 udata_getInfo(UDataMemory *pData, UDataInfo *pInfo) {
1468 if(pInfo!=NULL) {
1469 if(pData!=NULL && pData->pHeader!=NULL) {
1470 const UDataInfo *info=&pData->pHeader->info;
1471 uint16_t dataInfoSize=udata_getInfoSize(info);
1472 if(pInfo->size>dataInfoSize) {
1473 pInfo->size=dataInfoSize;
1474 }
1475 uprv_memcpy((uint16_t *)pInfo+1, (const uint16_t *)info+1, pInfo->size-2);
1476 if(info->isBigEndian!=U_IS_BIG_ENDIAN) {
1477 /* opposite endianness */
1478 uint16_t x=info->reservedWord;
1479 pInfo->reservedWord=(uint16_t)((x<<8)|(x>>8));
1480 }
1481 } else {
1482 pInfo->size=0;
1483 }
1484 }
1485 }
1486
1487
udata_setFileAccess(UDataFileAccess access,UErrorCode *)1488 U_CAPI void U_EXPORT2 udata_setFileAccess(UDataFileAccess access, UErrorCode * /*status*/)
1489 {
1490 // Note: this function is documented as not thread safe.
1491 gDataFileAccess = access;
1492 }
1493