Skip to content

Commit

Permalink
Experiment: Make char[] stack-allocated by JVM
Browse files Browse the repository at this point in the history
  • Loading branch information
@trustin
trustin committed Dec 19, 2019
1 parent c399ff9 commit 8383ef9
Show file tree
Hide file tree
Showing 2 changed files with 44 additions and 56 deletions.
5 changes: 2 additions & 3 deletions core/src/main/java/com/linecorp/armeria/common/QueryParamGetters.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -334,8 +334,7 @@ default Stream<String> valueStream(String name) {
default String toQueryString() {
final TemporaryThreadLocals tempThreadLocals = TemporaryThreadLocals.get();
final StringBuilder buf = tempThreadLocals.stringBuilder();
QueryStringEncoder.encodeParams(tempThreadLocals, buf, this);
return buf.toString();
return appendQueryString(buf).toString();
}

/**
Expand All @@ -347,7 +346,7 @@ default String toQueryString() {
*/
default StringBuilder appendQueryString(StringBuilder buf) {
requireNonNull(buf, "buf");
QueryStringEncoder.encodeParams(TemporaryThreadLocals.get(), buf, this);
QueryStringEncoder.encodeParams(buf, this);
return buf;
}
}
95 changes: 42 additions & 53 deletions core/src/main/java/com/linecorp/armeria/common/QueryStringEncoder.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -58,18 +58,16 @@ private static byte[] createSafeOctets(String safeChars) {
return octets;
}

static void encodeParams(TemporaryThreadLocals tempThreadLocals,
StringBuilder buf, QueryParamGetters params) {
static void encodeParams(StringBuilder buf, QueryParamGetters params) {
for (Entry<String, String> e : params) {
encodeComponent(tempThreadLocals, buf, e.getKey()).append('=');
encodeComponent(tempThreadLocals, buf, e.getValue()).append('&');
encodeComponent(buf, e.getKey()).append('=');
encodeComponent(buf, e.getValue()).append('&');
}

buf.setLength(buf.length() - 1);
}

private static StringBuilder encodeComponent(TemporaryThreadLocals tempThreadLocals,
StringBuilder buf, String s) {
private static StringBuilder encodeComponent(StringBuilder buf, String s) {
final int firstUnsafeOctetIdx = indexOfUnsafeOctet(s, 0);
if (firstUnsafeOctetIdx < 0) {
buf.append(s);
Expand All @@ -78,37 +76,21 @@ private static StringBuilder encodeComponent(TemporaryThreadLocals tempThreadLoc
buf.append(s, 0, firstUnsafeOctetIdx);
}

encodeUtf8Component(tempThreadLocals, buf, s, firstUnsafeOctetIdx);
encodeUtf8Component(buf, s, firstUnsafeOctetIdx);
}
return buf;
}

private static int indexOfUnsafeOctet(String s, int start) {
final int len = s.length();
for (int i = start; i < len; i++) {
if (!isSafeOctet(s.charAt(i))) {
return i;
}
}

return -1;
}

private static boolean isSafeOctet(char c) {
return c < SAFE_OCTETS.length && SAFE_OCTETS[c] != 0;
}

/**
* Encodes a query component (name or value). The octet at {@code start} must be an unsafe octet.
*/
private static void encodeUtf8Component(TemporaryThreadLocals tempThreadLocals,
StringBuilder buf, String s, int start) {
private static void encodeUtf8Component(StringBuilder buf, String s, int start) {
final int end = s.length();
if (start == end) {
return;
}

final char[] tmp = tempThreadLocals.charArray(12);
final char[] tmp = new char[12];
tmp[0] = tmp[3] = tmp[6] = tmp[9] = '%'; // Pre-set '%' so we don't have to set anymore.

int i = start;
Expand Down Expand Up @@ -147,8 +129,34 @@ private static void encodeUtf8Component(TemporaryThreadLocals tempThreadLocals,
buf.append(UTF_UNKNOWN);
break;
} else {
// Extra method to allow inlining the rest of writeUtf8 which is the most likely code path.
writeUtf8Surrogate(buf, tmp, c, s.charAt(i++));
final char c2 = s.charAt(i++);
if (!Character.isLowSurrogate(c2)) {
buf.append(UTF_UNKNOWN);
if (Character.isHighSurrogate(c2)) {
buf.append(UTF_UNKNOWN);
} else {
tmp[2] = UPPER_HEX_DIGITS[c2 & 0xF];
tmp[1] = UPPER_HEX_DIGITS[c2 >>> 4];
buf.append(tmp, 0, 3);
}
} else {
int codePoint = Character.toCodePoint(c, c2);
// See http://www.unicode.org/versions/Unicode7.0.0/ch03.pdf#G2630.
tmp[11] = UPPER_HEX_DIGITS[codePoint & 0xF];
codePoint >>>= 4;
tmp[10] = UPPER_HEX_DIGITS[0x8 | (codePoint & 0x3)];
codePoint >>>= 2;
tmp[8] = UPPER_HEX_DIGITS[codePoint & 0xF];
codePoint >>>= 4;
tmp[7] = UPPER_HEX_DIGITS[0x8 | (codePoint & 0x3)];
codePoint >>>= 2;
tmp[5] = UPPER_HEX_DIGITS[codePoint & 0xF];
codePoint >>>= 4;
tmp[4] = UPPER_HEX_DIGITS[0x8 | (codePoint & 0x3)];
tmp[2] = UPPER_HEX_DIGITS[(codePoint >>> 2) & 0xF];
tmp[1] = 'F';
buf.append(tmp, 0, 12);
}
}

// Find and append the safe region as-is.
Expand All @@ -167,35 +175,16 @@ private static void encodeUtf8Component(TemporaryThreadLocals tempThreadLocals,
}
}

private static void writeUtf8Surrogate(StringBuilder buf, char[] tmp, char c, char c2) {
if (!Character.isLowSurrogate(c2)) {
buf.append(UTF_UNKNOWN);
if (Character.isHighSurrogate(c2)) {
buf.append(UTF_UNKNOWN);
} else {
tmp[2] = UPPER_HEX_DIGITS[c2 & 0xF];
tmp[1] = UPPER_HEX_DIGITS[c2 >>> 4];
buf.append(tmp, 0, 3);
private static int indexOfUnsafeOctet(String s, int start) {
final int len = s.length();
for (int i = start; i < len; i++) {
final char c = s.charAt(i);
if (c >= SAFE_OCTETS.length || SAFE_OCTETS[c] == 0) {
return i;
}
return;
}

int codePoint = Character.toCodePoint(c, c2);
// See http://www.unicode.org/versions/Unicode7.0.0/ch03.pdf#G2630.
tmp[11] = UPPER_HEX_DIGITS[codePoint & 0xF];
codePoint >>>= 4;
tmp[10] = UPPER_HEX_DIGITS[0x8 | (codePoint & 0x3)];
codePoint >>>= 2;
tmp[8] = UPPER_HEX_DIGITS[codePoint & 0xF];
codePoint >>>= 4;
tmp[7] = UPPER_HEX_DIGITS[0x8 | (codePoint & 0x3)];
codePoint >>>= 2;
tmp[5] = UPPER_HEX_DIGITS[codePoint & 0xF];
codePoint >>>= 4;
tmp[4] = UPPER_HEX_DIGITS[0x8 | (codePoint & 0x3)];
tmp[2] = UPPER_HEX_DIGITS[(codePoint >>> 2) & 0xF];
tmp[1] = 'F';
buf.append(tmp, 0, 12);
return -1;
}

private QueryStringEncoder() {}
Expand Down

0 comments on commit 8383ef9

Please sign in to comment.