|
|
@ -16,15 +16,6 @@ class RuleAnalyzer(data: String, code: Boolean = false) { |
|
|
|
//设置平衡组函数,json或JavaScript时设置成chompCodeBalanced,否则为chompRuleBalanced |
|
|
|
//设置平衡组函数,json或JavaScript时设置成chompCodeBalanced,否则为chompRuleBalanced |
|
|
|
val chompBalanced = if (code) ::chompCodeBalanced else ::chompRuleBalanced |
|
|
|
val chompBalanced = if (code) ::chompCodeBalanced else ::chompRuleBalanced |
|
|
|
|
|
|
|
|
|
|
|
//当前平衡字段 |
|
|
|
|
|
|
|
fun currBalancedString( |
|
|
|
|
|
|
|
stepStart: Int = 1, |
|
|
|
|
|
|
|
stepEnd: Int = 1 |
|
|
|
|
|
|
|
): String { //stepStart平衡字符的起始分隔字串长度,stepEnd平衡字符的结束分隔字串长度 |
|
|
|
|
|
|
|
return queue.substring(start + stepStart, pos - stepEnd) //当前平衡字段 |
|
|
|
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
fun trim() { // 修剪当前规则之前的"@"或者空白符 |
|
|
|
fun trim() { // 修剪当前规则之前的"@"或者空白符 |
|
|
|
while (queue[pos] == '@' || queue[pos] < '!') pos++ |
|
|
|
while (queue[pos] == '@' || queue[pos] < '!') pos++ |
|
|
|
} |
|
|
|
} |
|
|
@ -38,7 +29,7 @@ class RuleAnalyzer(data: String, code: Boolean = false) { |
|
|
|
fun currString(): String { |
|
|
|
fun currString(): String { |
|
|
|
return queue.substring(start, pos) //当前拉取到的字段 |
|
|
|
return queue.substring(start, pos) //当前拉取到的字段 |
|
|
|
} |
|
|
|
} |
|
|
|
|
|
|
|
|
|
|
|
/** |
|
|
|
/** |
|
|
|
* 是否已无剩余字符? |
|
|
|
* 是否已无剩余字符? |
|
|
|
* @return 若剩余字串中已无字符则返回true |
|
|
|
* @return 若剩余字串中已无字符则返回true |
|
|
@ -431,7 +422,7 @@ class RuleAnalyzer(data: String, code: Boolean = false) { |
|
|
|
|
|
|
|
|
|
|
|
while (consumeTo(inner)) { //拉取成功返回true,ruleAnalyzes里的字符序列索引变量pos后移相应位置,否则返回false,且isEmpty为true |
|
|
|
while (consumeTo(inner)) { //拉取成功返回true,ruleAnalyzes里的字符序列索引变量pos后移相应位置,否则返回false,且isEmpty为true |
|
|
|
if (chompCodeBalanced('{', '}')) { |
|
|
|
if (chompCodeBalanced('{', '}')) { |
|
|
|
val frv = fr(currBalancedString(startStep, endStep)) |
|
|
|
val frv = fr(queue.substring(start + stepStart, pos - stepEnd)) |
|
|
|
if (frv != null) { |
|
|
|
if (frv != null) { |
|
|
|
st.append(queue.substring(startX, start) + frv) //压入内嵌规则前的内容,及内嵌规则解析得到的字符串 |
|
|
|
st.append(queue.substring(startX, start) + frv) //压入内嵌规则前的内容,及内嵌规则解析得到的字符串 |
|
|
|
startX = pos //记录下次规则起点 |
|
|
|
startX = pos //记录下次规则起点 |
|
|
|