Different implementations of BigDecimal.stripTrailingZeros(), in
different versions of the JDK (and different versions of Android), have
differences in their handling of zero. To avoid this, ICU4J can return
BigDecimal.ZERO for any value that is equal to zero, instead of calling
BigDecimal.stripTrailingZeros() in this problematic case.
* ICU-13530 copy C/C++ files UTrie2 -> UTrie3
X-SVN-Rev: 40754
* ICU-13530 UTrie3 new files copied from UTrie2: rename types/functions/macros
X-SVN-Rev: 40755
* ICU-13530 debug-print building each UTrie2
X-SVN-Rev: 40756
* ICU-13530 remove two-byte-UTF-8 errorValue block; move highValue from end of data array into header; add errorValue to header
X-SVN-Rev: 40762
* ICU-13530 UTrie3 U16_NEXT/PREV: errorValue for unpaired surrogates
X-SVN-Rev: 40763
* ICU-13530 no more separate values for lead surrogate code units
X-SVN-Rev: 40764
* ICU-13530 change from 11:5 trie bits to 10:6 for simpler UTF-8 code
X-SVN-Rev: 40766
* ICU-13530 UTrie2 build UTrie3 as well, print sizes
X-SVN-Rev: 40767
* ICU-13530 debug-print countSame, sumOverlaps, countInitial
X-SVN-Rev: 40768
* ICU-13530 debug-print whether trie is for CanonIterData
X-SVN-Rev: 40769
* ICU-13530 no index-shift for BMP data, no separate index-2 for 2-byte UTF-8; builder changes incomplete
X-SVN-Rev: 40777
* ICU-13530 remove errorValue and highStart from UNewTrie3
X-SVN-Rev: 40778
* ICU-13530 rewrite UTrie3 builder code
X-SVN-Rev: 40783
* ICU-13530 UTrie3 bug fixes
X-SVN-Rev: 40788
* ICU-13530 fully re-inline _UTRIE3_U8_NEXT()
X-SVN-Rev: 40790
* ICU-13530 find most common all-same data block for dataNullBlock and initialValue
X-SVN-Rev: 40792
* ICU-13530 UTrie3 iterator functions take start and return the end of a range, rather than callback call for each range
X-SVN-Rev: 40800
* ICU-13530 mask off unused data value bits before building a UTrie3 with values less than 32 bits wide
X-SVN-Rev: 40803
* ICU-13530 split utrie3builder.h out of utrie3.h
X-SVN-Rev: 40804
* ICU-13530 separate types UTrie3 vs. UTrie3Builder, implement builder as wrapper over C++ class Trie3Builder in .cpp
X-SVN-Rev: 40809
* ICU-13530 function to make a UTrie3Builder from a UTrie3
X-SVN-Rev: 40810
* ICU-13530 debug-print some data; some cleanup
X-SVN-Rev: 40865
* ICU-13530 BMP 10:6 but supplementary 10:6:4
X-SVN-Rev: 40984
* ICU-13530 move errorValue & highValue to the end of the data table, minimal padding to 4 bytes
X-SVN-Rev: 41011
* ICU-13530 index-1 table gap of index-2 null blocks
X-SVN-Rev: 41018
* ICU-13530 test with more than 128k compacted data
X-SVN-Rev: 41034
* ICU-13530 supplementary bits 11:5:4 saves a little space
X-SVN-Rev: 41039
* ICU-13530 supplementary bits 6:5:5:4 instead of gap: about same size but simpler
X-SVN-Rev: 41050
* ICU-13530 remove unnecessary utrie3_clone(built trie)
X-SVN-Rev: 41058
* ICU-13530 remove unnecessary UTrie3StringIterator
X-SVN-Rev: 41059
* ICU-13530 back to UTRIE3_GET...() macros *returning* data values
X-SVN-Rev: 41060
* ICU-13530 fast vs. small
X-SVN-Rev: 41066
* ICU-13530 always load NFC data, add simple normalization performance test
X-SVN-Rev: 41110
* ICU-13530 change normalization main trie to UTrie3 with special values for lead surrogates; forbid non-inert surrogate code *points* because unable to store values different from code *units*; runtime code work around that for code point lookup and iteration; adjust UTS 46 for normalization no longer mapping unpaired surrogates to U+FFFD
X-SVN-Rev: 41122
* ICU-13530 simplenormperf bug fix and NFC base line
X-SVN-Rev: 41126
* ICU-13530 move normalization getRange skipping lead surrogates to API getRangeSkipLead()
X-SVN-Rev: 41182
* ICU-13530 switch CanonIterData and gennorm2 Norms to UTrie3
X-SVN-Rev: 41183
* ICU-13530 remove unused overwrite parameter from setRange()
X-SVN-Rev: 41184
* ICU-13530 getRange skip lead -> fixed surrogates
X-SVN-Rev: 41219
* ICU-13530 minor cleanup
X-SVN-Rev: 41221
* ICU-13530 UTS 46 code map unpaired surrogates to U+FFFD before normalization
X-SVN-Rev: 41224
* ICU-13530 minor internal-docs cleanup
X-SVN-Rev: 41225
* ICU-13530 rename UTrie3 to UCPTrie, and other name changes
X-SVN-Rev: 41226
* ICU-13530 add 8-bit data option; add type-any & valueBits-any for fromBinary(); macros consistently source type then data width
X-SVN-Rev: 41234
* ICU-13530 scrub the API docs for the proposal
X-SVN-Rev: 41319
* ICU-13530 tag internal definitions as such, or move them to an internal header
X-SVN-Rev: 41320
* ICU-13530 Java API skeleton
X-SVN-Rev: 41326
* ICU-13530 API feedback: ValueWidth, MutableCodePointTrie, base CodePointMap, ...
X-SVN-Rev: 41382
* ICU-13530 add UCPTrie valueWidth field and padding, and combine data pointers into a union
X-SVN-Rev: 41408
* ICU-13530 switch some macros to using dataAccess parameter: separate index vs. data lookups, no macro variant for each value width
X-SVN-Rev: 41409
* ICU-13530 StringIterator is no longer a java.util.Iterator (bad fit)
X-SVN-Rev: 41455
* ICU-13530 CodePointTrie.java code complete
X-SVN-Rev: 41518
* ICU-13530 finish Java port incl test; keep C++ parallel
* ICU-13530 adjust API for feedback: rename HandleValue to FilterValue, change getRange+getRangeFixedSurr(bool allSurr) to enum RangeOption+getRange(enum option); change remaining C macros to use dataAccess for 16/32/8-bit value widths; fix/clarify some API docs
* ICU-13530 add javadoc
* ICU-13530 document UCPTrie binary data format
* ICU-13530 update .nrm formatVersion 3->4, document change in surrogate handling with new trie
* ICU-13530 re-hardcode NFC data
* move trie swapper code into new file; add new files to Windows project files; turn off trie debugging
* ICU-13530 minor cleanup
* ICU-13530 test more range starts; fix a C test leak
* ICU-13530 regenerate Java data from scratch
* ICU-13530 review feedback changes: API docs typos, more @internal, C++11 field initializers, fix potential leak in MutableCodePointTrie::fromUCPTrie()
* ICU-13530 rename interface FilterValue to ValueFilter
Simplified redundant code and removed unnecessary code path for Java 7 and later runtime in various places. Also cleaned up stale comments.
There is one test code change in com.ibm.icu.dev.test.localespi.NumberFormatTest - number keyword in test case was changed to Arab to arab. This test case was skipped with Java 6 runtime. It looks this code was note tested on Java 7 and later. @number=Arab does not work because it's case sensitive and must be all lower case letters.
* ICU-13138 do not create symlink to Static ICU Data binary
For builds using mingw-w64 and MSVC with Static layout,
ICU installation folder contains '.dll' file, which is
a symbolic link to '.a' or '.lib' Data binary.
This have no sense anyway, since it can't be used as
binary with shared layout.
* ICU-13138 mingw: add lib prefix to Static Data binary name
For builds using mingw-w64 with Static layout, missing 'lib'
prefix in ICU Data binary name leads to Stubdata binary use
during ICU extras, samples and tests build. ICU Stubdata
binary also substitutes Data in installation directory.
The above leads to runtime errors of ICU tools, tests and
ICU-based software.
Thus need to add 'lib' prefix to ICU Data binary name during
Pkgdata run.
Task-number: ICU-13138
Task-number: ICU-13187
Previously, ULocale had to work on Java 6, therefore, Java 7 specific Locale methods needed to be invoked by reflection. For the next release, we can directly call Java 7 Locale APIs, so rather complicated reflection implementation were replaced with simple API calls.
Previously, some developers accidentally introduced Java 7 Objects class utility methods. At that time, we once added Java 6 compatible implementation in com.ibm.icu.impl.Utility class. Now, we use Java 7 as the minimum supported Java version, so we can use Java 7 Objects class and methods.
There are some extra changes dropping generics type from constructor. For example from `ArrayList<String> output = new ArrayList<String>();` to `ArrayList<String> output = new ArrayList<>();`. The updated syntax is allowed since Java 7. We have eclipse project configured to normalize such expression, therefore, files touched by this commit were automatically updated. These changes are not directly related to Java 7 Objects method replacement.
Includes all line break tailorings.
Corresponding updates to monkey test rules.
State table builder, fix missed table optimization, uncovered by new rule.
ICU very often ends up with double separators in the file paths used to open files. This change prevents extra slashes on data file paths.
- The feedback from Markus was that the suffix shouldn't really start with a separator.
- It turns out that the suffix can be used for either a file extension (like .dat or .res) or for an actual item name as well (like ibm-5348_P100-1997.cnv).
- If the suffix is an extension then we don't want to append the separator at all. We only want to append it if the suffix is an item name.
- Also use a StringPiece to save repeated implicit calls to uprv_strlen.
In the prior change for this ticket the logic was mixed up. For IntelliSense we actually want to turn *off* the renaming defines so that VS doesn't get confused when you press F12.
This will prevent littering the source tree with spurious .pyc files.
The potential faster execution when re-running a script that has an
up-to-date .pyc file is negligible.
At the moment, CLDR pseudolocale region codes don't have display names.
As long as that is so, ICU must not require them to exist in order to
not cause bogus test failures when building with pseudolocales.
ICU-20036 CurrencyPluralInfo class doesn't always check/handle OOM errors.
Changes include:
- Using LocalPointer instead of raw new/delete, in order to make the code cleaner.
- Using nullptr instead of NULL.
- Inspired by Andy's review feedback PluralRules changes, this change sets fPluralRules and fLocale to nullptr in the assignment operator in order to prevent possible double deletes in the failure case.
- More consistent about not checking for nullptr when calling delete.
- Using LocalUResourceBundlePointer in order to simply the code and not need manual deletes.
- Reduce memory usage by using the same LocalUResourceBundle with .getAlias() instead of allocating new ones.
Both the number of installed locales and the name of the last locale
change with updates to and customizations of CLDR/ICU data so test
expectations on these values will cause bogus test failures.
- There are a few locations in the RBNF class that don't check for out-of-memory (OOM) failures.
- Using LocalPointer to clean up the manual deletes.
- Change to use nullptr instead of NULL.
- A few minor typo fixes as well.
For builds using Cygwin and MSVC with Debug configuration, '.dll' and '.lib' binaries has 'd' suffix.
Doing the same for builds using MinGW and MinGW64, based on the code from MSVC toolchain.
Task-number: ICU-13138
ICU-20041 ICU4C NumberingSystem class doesn't handle out-of-memory (OOM) failures.
- Not all code paths in the NumberingSystem class check for OOM failures. This can lead to crashes in some cases as null pointers will be dereferenced without any checks.
- Change to use nullptr instead of NULL.
- Don't stomp on OOM errors when attempting to load resources. We should report back OOM to the caller.
- Use LocalPointer in order simplify the code and for automatic clean-up of memory.
- Use LocalUResourceBundlePointer as well to help simply things even more.
- PluralRules class doesn't handle out-of-memory (OOM) errors in some code paths.
- The clone and assignment operator (operator=) methods of construction don't take an error code parameter, meaning that if an OOM error occurs during the constructor, it will not reported back to the caller, and the caller has no way to know that the object is in a half-constructed state.
- Using an internal status variable for these above cases.
- Changes to the various PluralRules helper classes to check for OOM as well.
- Using nullptr instead NULL.
- Using LocalPointer instead of raw new/delete in some cases.
- Need to set mRules to nullptr, otherwise we can end up with double deletes in the failure case. (Thanks to Andy for the review).
- Using default member initializers for class members to avoid dangling pointers. Also allows for using default constructors too.
ICU-20034 ICU4C the Locale class's assignment operator should set the locale to "bogus" if an OOM error occurs when attempting to copy data over from the other locale.
Also need to check strdup, as that calls malloc and it can fail too.
