Imported GNU Classpath 0.19 + gcj-import-20051115.

* sources.am: Regenerated.
       * Makefile.in: Likewise.
       * scripts/makemake.tcl: Use glob -nocomplain.

From-SVN: r107049
This commit is contained in:
Mark Wielaard 2005-11-15 23:20:01 +00:00
parent 02e549bfaa
commit 8f523f3a10
1241 changed files with 97711 additions and 25284 deletions

View file

@ -147,6 +147,24 @@ public final class StringBuffer implements Serializable, CharSequence
str.getChars(0, count, value, 0);
}
/**
* Create a new <code>StringBuffer</code> with the characters from the
* specified <code>CharSequence</code>. Initial capacity will be the
* size of the CharSequence plus 16.
*
* @param sequence the <code>String</code> to convert
* @throws NullPointerException if str is null
*
* @since 1.5
*/
public StringBuffer(CharSequence sequence)
{
count = Math.max(0, sequence.length());
value = new char[count + DEFAULT_CAPACITY];
for (int i = 0; i < count; ++i)
value[i] = sequence.charAt(i);
}
/**
* Get the length of the <code>String</code> this <code>StringBuffer</code>
* would create. Not to be confused with the <em>capacity</em> of the
@ -234,7 +252,6 @@ public final class StringBuffer implements Serializable, CharSequence
* @param index the index of the character to get, starting at 0
* @return the character at the specified index
* @throws IndexOutOfBoundsException if index is negative or &gt;= length()
* (while unspecified, this is a StringIndexOutOfBoundsException)
*/
public synchronized char charAt(int index)
{
@ -243,6 +260,39 @@ public final class StringBuffer implements Serializable, CharSequence
return value[index];
}
/**
* Get the code point at the specified index. This is like #charAt(int),
* but if the character is the start of a surrogate pair, and the
* following character completes the pair, then the corresponding
* supplementary code point is returned.
* @param index the index of the codepoint to get, starting at 0
* @return the codepoint at the specified index
* @throws IndexOutOfBoundsException if index is negative or &gt;= length()
* @since 1.5
*/
public synchronized int codePointAt(int index)
{
return Character.codePointAt(value, index, count);
}
/**
* Get the code point before the specified index. This is like
* #codePointAt(int), but checks the characters at <code>index-1</code> and
* <code>index-2</code> to see if they form a supplementary code point.
* @param index the index just past the codepoint to get, starting at 0
* @return the codepoint at the specified index
* @throws IndexOutOfBoundsException if index is negative or &gt;= length()
* @since 1.5
*/
public synchronized int codePointBefore(int index)
{
// Character.codePointBefore() doesn't perform this check. We
// could use the CharSequence overload, but this is just as easy.
if (index >= count)
throw new IndexOutOfBoundsException();
return Character.codePointBefore(value, index, 1);
}
/**
* Get the specified array of characters. <code>srcOffset - srcEnd</code>
* characters will be copied into the array you pass in.
@ -340,6 +390,46 @@ public final class StringBuffer implements Serializable, CharSequence
return this;
}
/**
* Append the <code>CharSequence</code> value of the argument to this
* <code>StringBuffer</code>.
*
* @param sequence the <code>CharSequence</code> to append
* @return this <code>StringBuffer</code>
* @see #append(Object)
* @since 1.5
*/
public synchronized StringBuffer append(CharSequence sequence)
{
if (sequence == null)
sequence = "null";
return append(sequence, 0, sequence.length());
}
/**
* Append the specified subsequence of the <code>CharSequence</code>
* argument to this <code>StringBuffer</code>.
*
* @param sequence the <code>CharSequence</code> to append
* @param start the starting index
* @param end one past the ending index
* @return this <code>StringBuffer</code>
* @see #append(Object)
* @since 1.5
*/
public synchronized StringBuffer append(CharSequence sequence,
int start, int end)
{
if (sequence == null)
sequence = "null";
if (start < 0 || end < 0 || start > end || end > sequence.length())
throw new IndexOutOfBoundsException();
ensureCapacity_unsynchronized(this.count + end - start);
for (int i = start; i < end; ++i)
value[count++] = sequence.charAt(i);
return this;
}
/**
* Append the <code>char</code> array to this <code>StringBuffer</code>.
* This is similar (but more efficient) than
@ -406,6 +496,25 @@ public final class StringBuffer implements Serializable, CharSequence
return this;
}
/**
* Append the code point to this <code>StringBuffer</code>.
* This is like #append(char), but will append two characters
* if a supplementary code point is given.
*
* @param code the code point to append
* @return this <code>StringBuffer</code>
* @see Character#toChars(int, char[], int)
* @since 1.5
*/
public synchronized StringBuffer appendCodePoint(int code)
{
int len = Character.charCount(code);
ensureCapacity_unsynchronized(count + len);
Character.toChars(code, value, count);
count += len;
return this;
}
/**
* Append the <code>String</code> value of the argument to this
* <code>StringBuffer</code>. Uses <code>String.valueOf()</code> to convert
@ -659,6 +768,54 @@ public final class StringBuffer implements Serializable, CharSequence
return this;
}
/**
* Insert the <code>CharSequence</code> argument into this
* <code>StringBuffer</code>. If the sequence is null, the String
* "null" is used instead.
*
* @param offset the place to insert in this buffer
* @param sequence the <code>CharSequence</code> to insert
* @return this <code>StringBuffer</code>
* @throws IndexOutOfBoundsException if offset is out of bounds
* @since 1.5
*/
public synchronized StringBuffer insert(int offset, CharSequence sequence)
{
if (sequence == null)
sequence = "null";
return insert(offset, sequence, 0, sequence.length());
}
/**
* Insert a subsequence of the <code>CharSequence</code> argument into this
* <code>StringBuffer</code>. If the sequence is null, the String
* "null" is used instead.
*
* @param offset the place to insert in this buffer
* @param sequence the <code>CharSequence</code> to insert
* @param start the starting index of the subsequence
* @param end one past the ending index of the subsequence
* @return this <code>StringBuffer</code>
* @throws IndexOutOfBoundsException if offset, start,
* or end are out of bounds
* @since 1.5
*/
public synchronized StringBuffer insert(int offset, CharSequence sequence,
int start, int end)
{
if (sequence == null)
sequence = "null";
if (start < 0 || end < 0 || start > end || end > sequence.length())
throw new IndexOutOfBoundsException();
int len = end - start;
ensureCapacity_unsynchronized(count + len);
VMSystem.arraycopy(value, offset, value, offset + len, count - offset);
for (int i = start; i < end; ++i)
value[offset++] = sequence.charAt(i);
count += len;
return this;
}
/**
* Insert the <code>char[]</code> argument into this
* <code>StringBuffer</code>.
@ -879,6 +1036,106 @@ public final class StringBuffer implements Serializable, CharSequence
return new String(this);
}
/**
* This may reduce the amount of memory used by the StringBuffer,
* by resizing the internal array to remove unused space. However,
* this method is not required to resize, so this behavior cannot
* be relied upon.
* @since 1.5
*/
public synchronized void trimToSize()
{
int wouldSave = value.length - count;
// Some random heuristics: if we save less than 20 characters, who
// cares.
if (wouldSave < 20)
return;
// If we save more than 200 characters, shrink.
// If we save more than 1/4 of the buffer, shrink.
if (wouldSave > 200 || wouldSave * 4 > value.length)
{
char[] newValue = new char[count];
VMSystem.arraycopy(value, 0, newValue, 0, count);
value = newValue;
}
}
/**
* Return the number of code points between two indices in the
* <code>StringBuffer</code>. An unpaired surrogate counts as a
* code point for this purpose. Characters outside the indicated
* range are not examined, even if the range ends in the middle of a
* surrogate pair.
*
* @param start the starting index
* @param end one past the ending index
* @return the number of code points
* @since 1.5
*/
public synchronized int codePointCount(int start, int end)
{
if (start < 0 || end >= count || start > end)
throw new StringIndexOutOfBoundsException();
int count = 0;
while (start < end)
{
char base = value[start];
if (base < Character.MIN_HIGH_SURROGATE
|| base > Character.MAX_HIGH_SURROGATE
|| start == end
|| start == count
|| value[start + 1] < Character.MIN_LOW_SURROGATE
|| value[start + 1] > Character.MAX_LOW_SURROGATE)
{
// Nothing.
}
else
{
// Surrogate pair.
++start;
}
++start;
++count;
}
return count;
}
/**
* Starting at the given index, this counts forward by the indicated
* number of code points, and then returns the resulting index. An
* unpaired surrogate counts as a single code point for this
* purpose.
*
* @param start the starting index
* @param codePoints the number of code points
* @return the resulting index
* @since 1.5
*/
public synchronized int offsetByCodePoints(int start, int codePoints)
{
while (codePoints > 0)
{
char base = value[start];
if (base < Character.MIN_HIGH_SURROGATE
|| base > Character.MAX_HIGH_SURROGATE
|| start == count
|| value[start + 1] < Character.MIN_LOW_SURROGATE
|| value[start + 1] > Character.MAX_LOW_SURROGATE)
{
// Nothing.
}
else
{
// Surrogate pair.
++start;
}
++start;
--codePoints;
}
return start;
}
/**
* An unsynchronized version of ensureCapacity, used internally to avoid
* the cost of a second lock on the same object. This also has the side