2 ******************************************************************************
4 * Copyright (C) 1999-2013, International Business Machines
5 * Corporation and others. All Rights Reserved.
7 ******************************************************************************
10 * tab size: 8 (not used)
13 * created on: 1999oct25
14 * created by: Markus W. Scherer
17 #include "unicode/utypes.h" /* U_PLATFORM etc. */
21 #define ATTRIBUTE_WEAK __attribute__ ((weak))
22 might have to #include some other header
26 #include "unicode/putil.h"
27 #include "unicode/udata.h"
28 #include "unicode/uversion.h"
41 /***********************************************************************
43 * Notes on the organization of the ICU data implementation
45 * All of the public API is defined in udata.h
47 * The implementation is split into several files...
49 * - udata.c (this file) contains higher level code that knows about
50 * the search paths for locating data, caching opened data, etc.
52 * - umapfile.c contains the low level platform-specific code for actually loading
53 * (memory mapping, file reading, whatever) data into memory.
55 * - ucmndata.c deals with the tables of contents of ICU data items within
56 * an ICU common format data file. The implementation includes
57 * an abstract interface and support for multiple TOC formats.
58 * All knowledge of any specific TOC format is encapsulated here.
60 * - udatamem.c has code for managing UDataMemory structs. These are little
61 * descriptor objects for blocks of memory holding ICU data of
65 /* configuration ---------------------------------------------------------- */
67 /* If you are excruciatingly bored turn this on .. */
68 /* #define UDATA_DEBUG 1 */
70 #if defined(UDATA_DEBUG)
74 #define LENGTHOF(array) (int32_t)(sizeof(array)/sizeof((array)[0]))
79 * Forward declarations
81 static UDataMemory *udata_findCachedData(const char *path);
83 /***********************************************************************
85 * static (Global) data
87 ************************************************************************/
90 * Pointers to the common ICU data.
92 * We store multiple pointers to ICU data packages and iterate through them
93 * when looking for a data item.
95 * It is possible to combine this with dependency inversion:
96 * One or more data package libraries may export
97 * functions that each return a pointer to their piece of the ICU data,
98 * and this file would import them as weak functions, without a
99 * strong linker dependency from the common library on the data library.
101 * Then we can have applications depend on only that part of ICU's data
102 * that they really need, reducing the size of binaries that take advantage
105 static UDataMemory *gCommonICUDataArray[10] = { NULL };
107 static UBool gHaveTriedToLoadCommonData = FALSE; /* See extendICUData(). */
109 static UHashtable *gCommonDataCache = NULL; /* Global hash table of opened ICU data files. */
110 static icu::UInitOnce gCommonDataCacheInitOnce = U_INITONCE_INITIALIZER;
112 static UDataFileAccess gDataFileAccess = UDATA_DEFAULT_ACCESS;
114 static UBool U_CALLCONV
119 if (gCommonDataCache) { /* Delete the cache of user data mappings. */
120 uhash_close(gCommonDataCache); /* Table owns the contents, and will delete them. */
121 gCommonDataCache = NULL; /* Cleanup is not thread safe. */
123 gCommonDataCacheInitOnce.reset();
125 for (i = 0; i < LENGTHOF(gCommonICUDataArray) && gCommonICUDataArray[i] != NULL; ++i) {
126 udata_close(gCommonICUDataArray[i]);
127 gCommonICUDataArray[i] = NULL;
129 gHaveTriedToLoadCommonData = FALSE;
131 return TRUE; /* Everything was cleaned up */
134 static UBool U_CALLCONV
135 findCommonICUDataByName(const char *inBasename)
140 UDataMemory *pData = udata_findCachedData(inBasename);
144 for (i = 0; i < LENGTHOF(gCommonICUDataArray); ++i) {
145 if ((gCommonICUDataArray[i] != NULL) && (gCommonICUDataArray[i]->pHeader == pData->pHeader)) {
146 /* The data pointer is already in the array. */
157 * setCommonICUData. Set a UDataMemory to be the global ICU Data
160 setCommonICUData(UDataMemory *pData, /* The new common data. Belongs to caller, we copy it. */
161 UBool warn, /* If true, set USING_DEFAULT warning if ICUData was */
162 /* changed by another thread before we got to it. */
165 UDataMemory *newCommonData = UDataMemory_createNewInstance(pErr);
167 UBool didUpdate = FALSE;
168 if (U_FAILURE(*pErr)) {
172 /* For the assignment, other threads must cleanly see either the old */
173 /* or the new, not some partially initialized new. The old can not be */
174 /* deleted - someone may still have a pointer to it lying around in */
176 UDatamemory_assign(newCommonData, pData);
178 for (i = 0; i < LENGTHOF(gCommonICUDataArray); ++i) {
179 if (gCommonICUDataArray[i] == NULL) {
180 gCommonICUDataArray[i] = newCommonData;
181 ucln_common_registerCleanup(UCLN_COMMON_UDATA, udata_cleanup);
184 } else if (gCommonICUDataArray[i]->pHeader == pData->pHeader) {
185 /* The same data pointer is already in the array. */
191 if (i == LENGTHOF(gCommonICUDataArray) && warn) {
192 *pErr = U_USING_DEFAULT_WARNING;
195 uprv_free(newCommonData);
201 setCommonICUDataPointer(const void *pData, UBool /*warn*/, UErrorCode *pErrorCode) {
203 UDataMemory_init(&tData);
204 UDataMemory_setData(&tData, pData);
205 udata_checkCommonData(&tData, pErrorCode);
206 return setCommonICUData(&tData, FALSE, pErrorCode);
210 findBasename(const char *path) {
211 const char *basename=uprv_strrchr(path, U_FILE_SEP_CHAR);
221 packageNameFromPath(const char *path)
223 if((path == NULL) || (*path == 0)) {
224 return U_ICUDATA_NAME;
227 path = findBasename(path);
229 if((path == NULL) || (*path == 0)) {
230 return U_ICUDATA_NAME;
237 /*----------------------------------------------------------------------*
239 * Cache for common data *
240 * Functions for looking up or adding entries to a cache of *
241 * data that has been previously opened. Avoids a potentially *
242 * expensive operation of re-opening the data for subsequent *
245 * Data remains cached for the duration of the process. *
247 *----------------------------------------------------------------------*/
249 typedef struct DataCacheElement {
257 * Deleter function for DataCacheElements.
258 * udata cleanup function closes the hash table; hash table in turn calls back to
259 * here for each entry.
261 static void U_CALLCONV DataCacheElement_deleter(void *pDCEl) {
262 DataCacheElement *p = (DataCacheElement *)pDCEl;
263 udata_close(p->item); /* unmaps storage */
264 uprv_free(p->name); /* delete the hash key string. */
265 uprv_free(pDCEl); /* delete 'this' */
268 static void udata_initHashTable() {
269 UErrorCode err = U_ZERO_ERROR;
270 U_ASSERT(gCommonDataCache == NULL);
271 gCommonDataCache = uhash_open(uhash_hashChars, uhash_compareChars, NULL, &err);
272 if (U_FAILURE(err)) {
273 // TODO: handle errors better.
274 gCommonDataCache = NULL;
276 if (gCommonDataCache != NULL) {
277 uhash_setValueDeleter(gCommonDataCache, DataCacheElement_deleter);
278 ucln_common_registerCleanup(UCLN_COMMON_UDATA, udata_cleanup);
282 /* udata_getCacheHashTable()
283 * Get the hash table used to store the data cache entries.
284 * Lazy create it if it doesn't yet exist.
286 static UHashtable *udata_getHashTable() {
287 umtx_initOnce(gCommonDataCacheInitOnce, &udata_initHashTable);
288 return gCommonDataCache;
293 static UDataMemory *udata_findCachedData(const char *path)
296 UDataMemory *retVal = NULL;
297 DataCacheElement *el;
298 const char *baseName;
300 baseName = findBasename(path); /* Cache remembers only the base name, not the full path. */
301 htable = udata_getHashTable();
303 el = (DataCacheElement *)uhash_get(htable, baseName);
309 fprintf(stderr, "Cache: [%s] -> %p\n", baseName, retVal);
315 static UDataMemory *udata_cacheDataItem(const char *path, UDataMemory *item, UErrorCode *pErr) {
316 DataCacheElement *newElement;
317 const char *baseName;
320 DataCacheElement *oldValue = NULL;
321 UErrorCode subErr = U_ZERO_ERROR;
323 if (U_FAILURE(*pErr)) {
327 /* Create a new DataCacheElement - the thingy we store in the hash table -
328 * and copy the supplied path and UDataMemoryItems into it.
330 newElement = (DataCacheElement *)uprv_malloc(sizeof(DataCacheElement));
331 if (newElement == NULL) {
332 *pErr = U_MEMORY_ALLOCATION_ERROR;
335 newElement->item = UDataMemory_createNewInstance(pErr);
336 if (U_FAILURE(*pErr)) {
337 uprv_free(newElement);
340 UDatamemory_assign(newElement->item, item);
342 baseName = findBasename(path);
343 nameLen = (int32_t)uprv_strlen(baseName);
344 newElement->name = (char *)uprv_malloc(nameLen+1);
345 if (newElement->name == NULL) {
346 *pErr = U_MEMORY_ALLOCATION_ERROR;
347 uprv_free(newElement->item);
348 uprv_free(newElement);
351 uprv_strcpy(newElement->name, baseName);
353 /* Stick the new DataCacheElement into the hash table.
355 htable = udata_getHashTable();
357 oldValue = (DataCacheElement *)uhash_get(htable, path);
358 if (oldValue != NULL) {
359 subErr = U_USING_DEFAULT_WARNING;
364 newElement->name, /* Key */
365 newElement, /* Value */
371 fprintf(stderr, "Cache: [%s] <<< %p : %s. vFunc=%p\n", newElement->name,
372 newElement->item, u_errorName(subErr), newElement->item->vFuncs);
375 if (subErr == U_USING_DEFAULT_WARNING || U_FAILURE(subErr)) {
376 *pErr = subErr; /* copy sub err unto fillin ONLY if something happens. */
377 uprv_free(newElement->name);
378 uprv_free(newElement->item);
379 uprv_free(newElement);
380 return oldValue ? oldValue->item : NULL;
383 return newElement->item;
386 /*----------------------------------------------------------------------*==============
388 * Path management. Could be shared with other tools/etc if need be *
391 *----------------------------------------------------------------------*/
393 #define U_DATA_PATHITER_BUFSIZ 128 /* Size of local buffer for paths */
394 /* Overflow causes malloc of larger buf */
398 class UDataPathIterator
401 UDataPathIterator(const char *path, const char *pkg,
402 const char *item, const char *suffix, UBool doCheckLastFour,
403 UErrorCode *pErrorCode);
404 const char *next(UErrorCode *pErrorCode);
407 const char *path; /* working path (u_icudata_Dir) */
408 const char *nextPath; /* path following this one */
409 const char *basename; /* item's basename (icudt22e_mt.res)*/
410 const char *suffix; /* item suffix (can be null) */
412 uint32_t basenameLen; /* length of basename */
414 CharString itemPath; /* path passed in with item name */
415 CharString pathBuffer; /* output path for this it'ion */
416 CharString packageStub; /* example: "/icudt28b". Will ignore that leaf in set paths. */
418 UBool checkLastFour; /* if TRUE then allow paths such as '/foo/myapp.dat'
419 * to match, checks last 4 chars of suffix with
420 * last 4 of path, then previous chars. */
424 * @param iter The iterator to be initialized. Its current state does not matter.
425 * @param path The full pathname to be iterated over. If NULL, defaults to U_ICUDATA_NAME
426 * @param pkg Package which is being searched for, ex "icudt28l". Will ignore leave directories such as /icudt28l
427 * @param item Item to be searched for. Can include full path, such as /a/b/foo.dat
428 * @param suffix Optional item suffix, if not-null (ex. ".dat") then 'path' can contain 'item' explicitly.
429 * Ex: 'stuff.dat' would be found in '/a/foo:/tmp/stuff.dat:/bar/baz' as item #2.
430 * '/blarg/stuff.dat' would also be found.
432 UDataPathIterator::UDataPathIterator(const char *inPath, const char *pkg,
433 const char *item, const char *inSuffix, UBool doCheckLastFour,
434 UErrorCode *pErrorCode)
437 fprintf(stderr, "SUFFIX1=%s PATH=%s\n", inSuffix, inPath);
441 path = u_getDataDirectory();
448 packageStub.append(U_FILE_SEP_CHAR, *pErrorCode).append(pkg, *pErrorCode);
450 fprintf(stderr, "STUB=%s [%d]\n", packageStub.data(), packageStub.length());
455 basename = findBasename(item);
456 basenameLen = (int32_t)uprv_strlen(basename);
459 if(basename == item) {
462 itemPath.append(item, (int32_t)(basename-item), *pErrorCode);
463 nextPath = itemPath.data();
466 fprintf(stderr, "SUFFIX=%s [%p]\n", inSuffix, inSuffix);
470 if(inSuffix != NULL) {
476 checkLastFour = doCheckLastFour;
478 /* pathBuffer will hold the output path strings returned by this iterator */
481 fprintf(stderr, "%p: init %s -> [path=%s], [base=%s], [suff=%s], [itempath=%s], [nextpath=%s], [checklast4=%s]\n",
489 checkLastFour?"TRUE":"false");
494 * Get the next path on the list.
496 * @param iter The Iter to be used
497 * @param len If set, pointer to the length of the returned path, for convenience.
498 * @return Pointer to the next path segment, or NULL if there are no more.
500 const char *UDataPathIterator::next(UErrorCode *pErrorCode)
502 if(U_FAILURE(*pErrorCode)) {
506 const char *currentPath = NULL;
508 const char *pathBasename;
512 if( nextPath == NULL ) {
515 currentPath = nextPath;
517 if(nextPath == itemPath.data()) { /* we were processing item's path. */
518 nextPath = path; /* start with regular path next tm. */
519 pathLen = (int32_t)uprv_strlen(currentPath);
521 /* fix up next for next time */
522 nextPath = uprv_strchr(currentPath, U_PATH_SEP_CHAR);
523 if(nextPath == NULL) {
524 /* segment: entire path */
525 pathLen = (int32_t)uprv_strlen(currentPath);
527 /* segment: until next segment */
528 pathLen = (int32_t)(nextPath - currentPath);
539 fprintf(stderr, "rest of path (IDD) = %s\n", currentPath);
540 fprintf(stderr, " ");
543 for(qqq=0;qqq<pathLen;qqq++)
545 fprintf(stderr, " ");
548 fprintf(stderr, "^\n");
551 pathBuffer.clear().append(currentPath, pathLen, *pErrorCode);
553 /* check for .dat files */
554 pathBasename = findBasename(pathBuffer.data());
556 if(checkLastFour == TRUE &&
558 uprv_strncmp(pathBuffer.data() +(pathLen-4), suffix, 4)==0 && /* suffix matches */
559 uprv_strncmp(findBasename(pathBuffer.data()), basename, basenameLen)==0 && /* base matches */
560 uprv_strlen(pathBasename)==(basenameLen+4)) { /* base+suffix = full len */
563 fprintf(stderr, "Have %s file on the path: %s\n", suffix, pathBuffer.data());
568 { /* regular dir path */
569 if(pathBuffer[pathLen-1] != U_FILE_SEP_CHAR) {
571 uprv_strncmp(pathBuffer.data()+(pathLen-4), ".dat", 4) == 0)
574 fprintf(stderr, "skipping non-directory .dat file %s\n", pathBuffer.data());
579 /* Check if it is a directory with the same name as our package */
580 if(!packageStub.isEmpty() &&
581 (pathLen > packageStub.length()) &&
582 !uprv_strcmp(pathBuffer.data() + pathLen - packageStub.length(), packageStub.data())) {
584 fprintf(stderr, "Found stub %s (will add package %s of len %d)\n", packageStub.data(), basename, basenameLen);
586 pathBuffer.truncate(pathLen - packageStub.length());
588 pathBuffer.append(U_FILE_SEP_CHAR, *pErrorCode);
592 pathBuffer.append(packageStub.data()+1, packageStub.length()-1, *pErrorCode);
594 if(*suffix) /* tack on suffix */
596 pathBuffer.append(suffix, *pErrorCode);
601 fprintf(stderr, " --> %s\n", pathBuffer.data());
604 return pathBuffer.data();
608 /* fell way off the end */
614 /* ==================================================================================*/
617 /*----------------------------------------------------------------------*
619 * Add a static reference to the common data library *
620 * Unless overridden by an explicit udata_setCommonData, this will be *
623 *----------------------------------------------------------------------*/
624 extern "C" const ICU_Data_Header U_DATA_API U_ICUDATA_ENTRY_POINT;
627 * This would be a good place for weak-linkage declarations of
628 * partial-data-library access functions where each returns a pointer
629 * to its data package, if it is linked in.
632 extern const void *uprv_getICUData_collation(void) ATTRIBUTE_WEAK;
633 extern const void *uprv_getICUData_conversion(void) ATTRIBUTE_WEAK;
636 /*----------------------------------------------------------------------*
638 * openCommonData Attempt to open a common format (.dat) file *
639 * Map it into memory (if it's not there already) *
640 * and return a UDataMemory object for it. *
642 * If the requested data is already open and cached *
643 * just return the cached UDataMem object. *
645 *----------------------------------------------------------------------*/
647 openCommonData(const char *path, /* Path from OpenChoice? */
648 int32_t commonDataIndex, /* ICU Data (index >= 0) if path == NULL */
649 UErrorCode *pErrorCode)
652 const char *pathBuffer;
653 const char *inBasename;
655 if (U_FAILURE(*pErrorCode)) {
659 UDataMemory_init(&tData);
661 /* ??????? TODO revisit this */
662 if (commonDataIndex >= 0) {
663 /* "mini-cache" for common ICU data */
664 if(commonDataIndex >= LENGTHOF(gCommonICUDataArray)) {
667 if(gCommonICUDataArray[commonDataIndex] == NULL) {
669 for(i = 0; i < commonDataIndex; ++i) {
670 if(gCommonICUDataArray[i]->pHeader == &U_ICUDATA_ENTRY_POINT.hdr) {
671 /* The linked-in data is already in the list. */
676 /* Add the linked-in data to the list. */
678 * This is where we would check and call weakly linked partial-data-library
682 if (uprv_getICUData_collation) {
683 setCommonICUDataPointer(uprv_getICUData_collation(), FALSE, pErrorCode);
685 if (uprv_getICUData_conversion) {
686 setCommonICUDataPointer(uprv_getICUData_conversion(), FALSE, pErrorCode);
689 setCommonICUDataPointer(&U_ICUDATA_ENTRY_POINT.hdr, FALSE, pErrorCode);
691 return gCommonICUDataArray[commonDataIndex];
695 /* request is NOT for ICU Data. */
697 /* Find the base name portion of the supplied path. */
698 /* inBasename will be left pointing somewhere within the original path string. */
699 inBasename = findBasename(path);
701 fprintf(stderr, "inBasename = %s\n", inBasename);
705 /* no basename. This will happen if the original path was a directory name, */
706 /* like "a/b/c/". (Fallback to separate files will still work.) */
708 fprintf(stderr, "ocd: no basename in %s, bailing.\n", path);
710 *pErrorCode=U_FILE_ACCESS_ERROR;
714 /* Is the requested common data file already open and cached? */
715 /* Note that the cache is keyed by the base name only. The rest of the path, */
716 /* if any, is not considered. */
718 UDataMemory *dataToReturn = udata_findCachedData(inBasename);
719 if (dataToReturn != NULL) {
724 /* Requested item is not in the cache.
725 * Hunt it down, trying all the path locations
728 UDataPathIterator iter(u_getDataDirectory(), inBasename, path, ".dat", TRUE, pErrorCode);
730 while((UDataMemory_isLoaded(&tData)==FALSE) && (pathBuffer = iter.next(pErrorCode)) != NULL)
733 fprintf(stderr, "ocd: trying path %s - ", pathBuffer);
735 uprv_mapFile(&tData, pathBuffer);
737 fprintf(stderr, "%s\n", UDataMemory_isLoaded(&tData)?"LOADED":"not loaded");
741 #if defined(OS390_STUBDATA) && defined(OS390BATCH)
742 if (!UDataMemory_isLoaded(&tData)) {
743 char ourPathBuffer[1024];
744 /* One more chance, for extendCommonData() */
745 uprv_strncpy(ourPathBuffer, path, 1019);
746 ourPathBuffer[1019]=0;
747 uprv_strcat(ourPathBuffer, ".dat");
748 uprv_mapFile(&tData, ourPathBuffer);
752 if (!UDataMemory_isLoaded(&tData)) {
754 *pErrorCode=U_FILE_ACCESS_ERROR;
758 /* we have mapped a file, check its header */
759 udata_checkCommonData(&tData, pErrorCode);
762 /* Cache the UDataMemory struct for this .dat file,
763 * so we won't need to hunt it down and map it again next time
764 * something is needed from it. */
765 return udata_cacheDataItem(inBasename, &tData, pErrorCode);
769 /*----------------------------------------------------------------------*
771 * extendICUData If the full set of ICU data was not loaded at *
772 * program startup, load it now. This function will *
773 * be called when the lookup of an ICU data item in *
774 * the common ICU data fails. *
776 * return true if new data is loaded, false otherwise.*
778 *----------------------------------------------------------------------*/
779 static UBool extendICUData(UErrorCode *pErr)
782 UDataMemory copyPData;
783 UBool didUpdate = FALSE;
786 * There is a chance for a race condition here.
787 * Normally, ICU data is loaded from a DLL or via mmap() and
788 * setCommonICUData() will detect if the same address is set twice.
789 * If ICU is built with data loading via fread() then the address will
790 * be different each time the common data is loaded and we may add
791 * multiple copies of the data.
792 * In this case, use a mutex to prevent the race.
793 * Use a specific mutex to avoid nested locks of the global mutex.
795 #if MAP_IMPLEMENTATION==MAP_STDIO
796 static UMutex extendICUDataMutex = U_MUTEX_INITIALIZER;
797 umtx_lock(&extendICUDataMutex);
799 if(!gHaveTriedToLoadCommonData) {
800 /* See if we can explicitly open a .dat file for the ICUData. */
801 pData = openCommonData(
802 U_ICUDATA_NAME, /* "icudt20l" , for example. */
803 -1, /* Pretend we're not opening ICUData */
806 /* How about if there is no pData, eh... */
808 UDataMemory_init(©PData);
810 UDatamemory_assign(©PData, pData);
811 copyPData.map = 0; /* The mapping for this data is owned by the hash table */
812 copyPData.mapAddr = 0; /* which will unmap it when ICU is shut down. */
813 /* CommonICUData is also unmapped when ICU is shut down.*/
814 /* To avoid unmapping the data twice, zero out the map */
815 /* fields in the UDataMemory that we're assigning */
816 /* to CommonICUData. */
818 didUpdate = /* no longer using this result */
819 setCommonICUData(©PData,/* The new common data. */
820 FALSE, /* No warnings if write didn't happen */
821 pErr); /* setCommonICUData honors errors; NOP if error set */
824 gHaveTriedToLoadCommonData = TRUE;
827 didUpdate = findCommonICUDataByName(U_ICUDATA_NAME); /* Return 'true' when a racing writes out the extended */
828 /* data after another thread has failed to see it (in openCommonData), so */
829 /* extended data can be examined. */
830 /* Also handles a race through here before gHaveTriedToLoadCommonData is set. */
832 #if MAP_IMPLEMENTATION==MAP_STDIO
833 umtx_unlock(&extendICUDataMutex);
835 return didUpdate; /* Return true if ICUData pointer was updated. */
836 /* (Could potentialy have been done by another thread racing */
837 /* us through here, but that's fine, we still return true */
838 /* so that current thread will also examine extended data. */
841 /*----------------------------------------------------------------------*
843 * udata_setCommonData *
845 *----------------------------------------------------------------------*/
846 U_CAPI void U_EXPORT2
847 udata_setCommonData(const void *data, UErrorCode *pErrorCode) {
848 UDataMemory dataMemory;
850 if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
855 *pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
859 /* set the data pointer and test for validity */
860 UDataMemory_init(&dataMemory);
861 UDataMemory_setData(&dataMemory, data);
862 udata_checkCommonData(&dataMemory, pErrorCode);
863 if (U_FAILURE(*pErrorCode)) {return;}
865 /* we have good data */
866 /* Set it up as the ICU Common Data. */
867 setCommonICUData(&dataMemory, TRUE, pErrorCode);
870 /*---------------------------------------------------------------------------
874 *---------------------------------------------------------------------------- */
875 U_CAPI void U_EXPORT2
876 udata_setAppData(const char *path, const void *data, UErrorCode *err)
880 if(err==NULL || U_FAILURE(*err)) {
884 *err=U_ILLEGAL_ARGUMENT_ERROR;
888 UDataMemory_init(&udm);
889 UDataMemory_setData(&udm, data);
890 udata_checkCommonData(&udm, err);
891 udata_cacheDataItem(path, &udm, err);
894 /*----------------------------------------------------------------------------*
896 * checkDataItem Given a freshly located/loaded data item, either *
897 * an entry in a common file or a separately loaded file, *
898 * sanity check its header, and see if the data is *
899 * acceptable to the app. *
900 * If the data is good, create and return a UDataMemory *
901 * object that can be returned to the application. *
902 * Return NULL on any sort of failure. *
904 *----------------------------------------------------------------------------*/
908 const DataHeader *pHeader, /* The data item to be checked. */
909 UDataMemoryIsAcceptable *isAcceptable, /* App's call-back function */
910 void *context, /* pass-thru param for above. */
911 const char *type, /* pass-thru param for above. */
912 const char *name, /* pass-thru param for above. */
913 UErrorCode *nonFatalErr, /* Error code if this data was not acceptable */
914 /* but openChoice should continue with */
915 /* trying to get data from fallback path. */
916 UErrorCode *fatalErr /* Bad error, caller should return immediately */
919 UDataMemory *rDataMem = NULL; /* the new UDataMemory, to be returned. */
921 if (U_FAILURE(*fatalErr)) {
925 if(pHeader->dataHeader.magic1==0xda &&
926 pHeader->dataHeader.magic2==0x27 &&
927 (isAcceptable==NULL || isAcceptable(context, type, name, &pHeader->info))
929 rDataMem=UDataMemory_createNewInstance(fatalErr);
930 if (U_FAILURE(*fatalErr)) {
933 rDataMem->pHeader = pHeader;
935 /* the data is not acceptable, look further */
936 /* If we eventually find something good, this errorcode will be */
938 *nonFatalErr=U_INVALID_FORMAT_ERROR;
944 * @return 0 if not loaded, 1 if loaded or err
946 static UDataMemory *doLoadFromIndividualFiles(const char *pkgName,
947 const char *dataPath, const char *tocEntryPathSuffix,
948 /* following arguments are the same as doOpenChoice itself */
949 const char *path, const char *type, const char *name,
950 UDataMemoryIsAcceptable *isAcceptable, void *context,
951 UErrorCode *subErrorCode,
952 UErrorCode *pErrorCode)
954 const char *pathBuffer;
955 UDataMemory dataMemory;
956 UDataMemory *pEntryData;
958 /* look in ind. files: package\nam.typ ========================= */
959 /* init path iterator for individual files */
960 UDataPathIterator iter(dataPath, pkgName, path, tocEntryPathSuffix, FALSE, pErrorCode);
962 while((pathBuffer = iter.next(pErrorCode)))
965 fprintf(stderr, "UDATA: trying individual file %s\n", pathBuffer);
967 if(uprv_mapFile(&dataMemory, pathBuffer))
969 pEntryData = checkDataItem(dataMemory.pHeader, isAcceptable, context, type, name, subErrorCode, pErrorCode);
970 if (pEntryData != NULL) {
972 * Hand off ownership of the backing memory to the user's UDataMemory.
974 pEntryData->mapAddr = dataMemory.mapAddr;
975 pEntryData->map = dataMemory.map;
978 fprintf(stderr, "** Mapped file: %s\n", pathBuffer);
983 /* the data is not acceptable, or some error occured. Either way, unmap the memory */
984 udata_close(&dataMemory);
986 /* If we had a nasty error, bail out completely. */
987 if (U_FAILURE(*pErrorCode)) {
991 /* Otherwise remember that we found data but didn't like it for some reason */
992 *subErrorCode=U_INVALID_FORMAT_ERROR;
995 fprintf(stderr, "%s\n", UDataMemory_isLoaded(&dataMemory)?"LOADED":"not loaded");
1002 * @return 0 if not loaded, 1 if loaded or err
1004 static UDataMemory *doLoadFromCommonData(UBool isICUData, const char * /*pkgName*/,
1005 const char * /*dataPath*/, const char * /*tocEntryPathSuffix*/, const char *tocEntryName,
1006 /* following arguments are the same as doOpenChoice itself */
1007 const char *path, const char *type, const char *name,
1008 UDataMemoryIsAcceptable *isAcceptable, void *context,
1009 UErrorCode *subErrorCode,
1010 UErrorCode *pErrorCode)
1012 UDataMemory *pEntryData;
1013 const DataHeader *pHeader;
1014 UDataMemory *pCommonData;
1015 int32_t commonDataIndex;
1016 UBool checkedExtendedICUData = FALSE;
1017 /* try to get common data. The loop is for platforms such as the 390 that do
1018 * not initially load the full set of ICU data. If the lookup of an ICU data item
1019 * fails, the full (but slower to load) set is loaded, the and the loop repeats,
1020 * trying the lookup again. Once the full set of ICU data is loaded, the loop wont
1021 * repeat because the full set will be checked the first time through.
1023 * The loop also handles the fallback to a .dat file if the application linked
1024 * to the stub data library rather than a real library.
1026 for (commonDataIndex = isICUData ? 0 : -1;;) {
1027 pCommonData=openCommonData(path, commonDataIndex, subErrorCode); /** search for pkg **/
1029 if(U_SUCCESS(*subErrorCode) && pCommonData!=NULL) {
1032 /* look up the data piece in the common data */
1033 pHeader=pCommonData->vFuncs->Lookup(pCommonData, tocEntryName, &length, subErrorCode);
1035 fprintf(stderr, "%s: pHeader=%p - %s\n", tocEntryName, pHeader, u_errorName(*subErrorCode));
1039 pEntryData = checkDataItem(pHeader, isAcceptable, context, type, name, subErrorCode, pErrorCode);
1041 fprintf(stderr, "pEntryData=%p\n", pEntryData);
1043 if (U_FAILURE(*pErrorCode)) {
1046 if (pEntryData != NULL) {
1047 pEntryData->length = length;
1052 /* Data wasn't found. If we were looking for an ICUData item and there is
1053 * more data available, load it and try again,
1054 * otherwise break out of this loop. */
1057 } else if (pCommonData != NULL) {
1058 ++commonDataIndex; /* try the next data package */
1059 } else if ((!checkedExtendedICUData) && extendICUData(subErrorCode)) {
1060 checkedExtendedICUData = TRUE;
1061 /* try this data package slot again: it changed from NULL to non-NULL */
1069 * A note on the ownership of Mapped Memory
1071 * For common format files, ownership resides with the UDataMemory object
1072 * that lives in the cache of opened common data. These UDataMemorys are private
1073 * to the udata implementation, and are never seen directly by users.
1075 * The UDataMemory objects returned to users will have the address of some desired
1076 * data within the mapped region, but they wont have the mapping info itself, and thus
1077 * won't cause anything to be removed from memory when they are closed.
1079 * For individual data files, the UDataMemory returned to the user holds the
1080 * information necessary to unmap the data on close. If the user independently
1081 * opens the same data file twice, two completely independent mappings will be made.
1082 * (There is no cache of opened data items from individual files, only a cache of
1083 * opened Common Data files, that is, files containing a collection of data items.)
1085 * For common data passed in from the user via udata_setAppData() or
1086 * udata_setCommonData(), ownership remains with the user.
1088 * UDataMemory objects themselves, as opposed to the memory they describe,
1089 * can be anywhere - heap, stack/local or global.
1090 * They have a flag to indicate when they're heap allocated and thus
1091 * must be deleted when closed.
1095 /*----------------------------------------------------------------------------*
1097 * main data loading functions *
1099 *----------------------------------------------------------------------------*/
1100 static UDataMemory *
1101 doOpenChoice(const char *path, const char *type, const char *name,
1102 UDataMemoryIsAcceptable *isAcceptable, void *context,
1103 UErrorCode *pErrorCode)
1105 UDataMemory *retVal = NULL;
1107 const char *dataPath;
1109 int32_t tocEntrySuffixIndex;
1110 const char *tocEntryPathSuffix;
1111 UErrorCode subErrorCode=U_ZERO_ERROR;
1112 const char *treeChar;
1114 UBool isICUData = FALSE;
1117 /* Is this path ICU data? */
1119 !strcmp(path, U_ICUDATA_ALIAS) || /* "ICUDATA" */
1120 !uprv_strncmp(path, U_ICUDATA_NAME U_TREE_SEPARATOR_STRING, /* "icudt26e-" */
1121 uprv_strlen(U_ICUDATA_NAME U_TREE_SEPARATOR_STRING)) ||
1122 !uprv_strncmp(path, U_ICUDATA_ALIAS U_TREE_SEPARATOR_STRING, /* "ICUDATA-" */
1123 uprv_strlen(U_ICUDATA_ALIAS U_TREE_SEPARATOR_STRING))) {
1127 #if (U_FILE_SEP_CHAR != U_FILE_ALT_SEP_CHAR) /* Windows: try "foo\bar" and "foo/bar" */
1128 /* remap from alternate path char to the main one */
1129 CharString altSepPath;
1131 if(uprv_strchr(path,U_FILE_ALT_SEP_CHAR) != NULL) {
1132 altSepPath.append(path, *pErrorCode);
1134 while((p=uprv_strchr(altSepPath.data(), U_FILE_ALT_SEP_CHAR))) {
1135 *p = U_FILE_SEP_CHAR;
1137 #if defined (UDATA_DEBUG)
1138 fprintf(stderr, "Changed path from [%s] to [%s]\n", path, altSepPath.s);
1140 path = altSepPath.data();
1145 CharString tocEntryName; /* entry name in tree format. ex: 'icudt28b/coll/ar.res' */
1146 CharString tocEntryPath; /* entry name in path format. ex: 'icudt28b\\coll\\ar.res' */
1149 CharString treeName;
1151 /* ======= Set up strings */
1153 pkgName.append(U_ICUDATA_NAME, *pErrorCode);
1157 pkg = uprv_strrchr(path, U_FILE_SEP_CHAR);
1158 first = uprv_strchr(path, U_FILE_SEP_CHAR);
1159 if(uprv_pathIsAbsolute(path) || (pkg != first)) { /* more than one slash in the path- not a tree name */
1160 /* see if this is an /absolute/path/to/package path */
1162 pkgName.append(pkg+1, *pErrorCode);
1164 pkgName.append(path, *pErrorCode);
1167 treeChar = uprv_strchr(path, U_TREE_SEPARATOR);
1169 treeName.append(treeChar+1, *pErrorCode); /* following '-' */
1171 pkgName.append(U_ICUDATA_NAME, *pErrorCode);
1173 pkgName.append(path, (int32_t)(treeChar-path), *pErrorCode);
1174 if (first == NULL) {
1176 This user data has no path, but there is a tree name.
1177 Look up the correct path from the data cache later.
1179 path = pkgName.data();
1184 pkgName.append(U_ICUDATA_NAME, *pErrorCode);
1186 pkgName.append(path, *pErrorCode);
1193 fprintf(stderr, " P=%s T=%s\n", pkgName.data(), treeName.data());
1196 /* setting up the entry name and file name
1197 * Make up a full name by appending the type to the supplied
1198 * name, assuming that a type was supplied.
1201 /* prepend the package */
1202 tocEntryName.append(pkgName, *pErrorCode);
1203 tocEntryPath.append(pkgName, *pErrorCode);
1204 tocEntrySuffixIndex = tocEntryName.length();
1206 if(!treeName.isEmpty()) {
1207 tocEntryName.append(U_TREE_ENTRY_SEP_CHAR, *pErrorCode).append(treeName, *pErrorCode);
1208 tocEntryPath.append(U_FILE_SEP_CHAR, *pErrorCode).append(treeName, *pErrorCode);
1211 tocEntryName.append(U_TREE_ENTRY_SEP_CHAR, *pErrorCode).append(name, *pErrorCode);
1212 tocEntryPath.append(U_FILE_SEP_CHAR, *pErrorCode).append(name, *pErrorCode);
1213 if(type!=NULL && *type!=0) {
1214 tocEntryName.append(".", *pErrorCode).append(type, *pErrorCode);
1215 tocEntryPath.append(".", *pErrorCode).append(type, *pErrorCode);
1217 tocEntryPathSuffix = tocEntryPath.data()+tocEntrySuffixIndex; /* suffix starts here */
1220 fprintf(stderr, " tocEntryName = %s\n", tocEntryName.data());
1221 fprintf(stderr, " tocEntryPath = %s\n", tocEntryName.data());
1225 path = COMMON_DATA_NAME; /* "icudt26e" */
1228 /************************ Begin loop looking for ind. files ***************/
1230 fprintf(stderr, "IND: inBasename = %s, pkg=%s\n", "(n/a)", packageNameFromPath(path));
1233 /* End of dealing with a null basename */
1234 dataPath = u_getDataDirectory();
1236 /**** COMMON PACKAGE - only if packages are first. */
1237 if(gDataFileAccess == UDATA_PACKAGES_FIRST) {
1239 fprintf(stderr, "Trying packages (UDATA_PACKAGES_FIRST)\n");
1242 retVal = doLoadFromCommonData(isICUData,
1243 pkgName.data(), dataPath, tocEntryPathSuffix, tocEntryName.data(),
1244 path, type, name, isAcceptable, context, &subErrorCode, pErrorCode);
1245 if((retVal != NULL) || U_FAILURE(*pErrorCode)) {
1250 /**** INDIVIDUAL FILES */
1251 if((gDataFileAccess==UDATA_PACKAGES_FIRST) ||
1252 (gDataFileAccess==UDATA_FILES_FIRST)) {
1254 fprintf(stderr, "Trying individual files\n");
1256 /* Check to make sure that there is a dataPath to iterate over */
1257 if ((dataPath && *dataPath) || !isICUData) {
1258 retVal = doLoadFromIndividualFiles(pkgName.data(), dataPath, tocEntryPathSuffix,
1259 path, type, name, isAcceptable, context, &subErrorCode, pErrorCode);
1260 if((retVal != NULL) || U_FAILURE(*pErrorCode)) {
1266 /**** COMMON PACKAGE */
1267 if((gDataFileAccess==UDATA_ONLY_PACKAGES) ||
1268 (gDataFileAccess==UDATA_FILES_FIRST)) {
1270 fprintf(stderr, "Trying packages (UDATA_ONLY_PACKAGES || UDATA_FILES_FIRST)\n");
1272 retVal = doLoadFromCommonData(isICUData,
1273 pkgName.data(), dataPath, tocEntryPathSuffix, tocEntryName.data(),
1274 path, type, name, isAcceptable, context, &subErrorCode, pErrorCode);
1275 if((retVal != NULL) || U_FAILURE(*pErrorCode)) {
1280 /* Load from DLL. If we haven't attempted package load, we also haven't had any chance to
1281 try a DLL (static or setCommonData/etc) load.
1282 If we ever have a "UDATA_ONLY_FILES", add it to the or list here. */
1283 if(gDataFileAccess==UDATA_NO_FILES) {
1285 fprintf(stderr, "Trying common data (UDATA_NO_FILES)\n");
1287 retVal = doLoadFromCommonData(isICUData,
1288 pkgName.data(), "", tocEntryPathSuffix, tocEntryName.data(),
1289 path, type, name, isAcceptable, context, &subErrorCode, pErrorCode);
1290 if((retVal != NULL) || U_FAILURE(*pErrorCode)) {
1295 /* data not found */
1296 if(U_SUCCESS(*pErrorCode)) {
1297 if(U_SUCCESS(subErrorCode)) {
1298 /* file not found */
1299 *pErrorCode=U_FILE_ACCESS_ERROR;
1301 /* entry point not found or rejected */
1302 *pErrorCode=subErrorCode;
1310 /* API ---------------------------------------------------------------------- */
1312 U_CAPI UDataMemory * U_EXPORT2
1313 udata_open(const char *path, const char *type, const char *name,
1314 UErrorCode *pErrorCode) {
1316 fprintf(stderr, "udata_open(): Opening: %s : %s . %s\n", (path?path:"NULL"), name, type);
1320 if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
1322 } else if(name==NULL || *name==0) {
1323 *pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
1326 return doOpenChoice(path, type, name, NULL, NULL, pErrorCode);
1332 U_CAPI UDataMemory * U_EXPORT2
1333 udata_openChoice(const char *path, const char *type, const char *name,
1334 UDataMemoryIsAcceptable *isAcceptable, void *context,
1335 UErrorCode *pErrorCode) {
1337 fprintf(stderr, "udata_openChoice(): Opening: %s : %s . %s\n", (path?path:"NULL"), name, type);
1340 if(pErrorCode==NULL || U_FAILURE(*pErrorCode)) {
1342 } else if(name==NULL || *name==0 || isAcceptable==NULL) {
1343 *pErrorCode=U_ILLEGAL_ARGUMENT_ERROR;
1346 return doOpenChoice(path, type, name, isAcceptable, context, pErrorCode);
1352 U_CAPI void U_EXPORT2
1353 udata_getInfo(UDataMemory *pData, UDataInfo *pInfo) {
1355 if(pData!=NULL && pData->pHeader!=NULL) {
1356 const UDataInfo *info=&pData->pHeader->info;
1357 uint16_t dataInfoSize=udata_getInfoSize(info);
1358 if(pInfo->size>dataInfoSize) {
1359 pInfo->size=dataInfoSize;
1361 uprv_memcpy((uint16_t *)pInfo+1, (const uint16_t *)info+1, pInfo->size-2);
1362 if(info->isBigEndian!=U_IS_BIG_ENDIAN) {
1363 /* opposite endianness */
1364 uint16_t x=info->reservedWord;
1365 pInfo->reservedWord=(uint16_t)((x<<8)|(x>>8));
1374 U_CAPI void U_EXPORT2 udata_setFileAccess(UDataFileAccess access, UErrorCode * /*status*/)
1376 gDataFileAccess = access;