OracleLinux 8.9中sed执行出现Invalid RangeEnd错误求助

问题概述 #

同一sed脚本在OracleLinux 7.9上可正常执行，但在OracleLinux 8.9运行时提示"invalid range end"错误，输出文件为空。

执行命令 #

sed -r -f ./sed/transform-diff.sed  < 15-diffToUpdate.xml > 15-diffToUpdate.xml.tmp

错误信息 #

sed: file ./sed/transform-diff.sed line 11: Invalid range end

涉事sed脚本内容 #

# sed script to transform generated Liquibase diff file

# Problem to solve: If a column is dropped, a RDBM drops as well every index, that uses these columns. In the
# Liquibase diff script, columns are dropped first, indexes later. In the previously described case, an index
# might already be deleted and the update might fail.
# Solution: Put every <dropIndex> in a separate <changeSet>, containing a <indexExists> preCondition that
# only executes the <dropIndex> if it still exists.
/<changeSet/ {
    N
    s%([ \t]*)<changeSet author="([a-zA-Z \(\)]*)" id="([a-zA-Z0-9-]*)">[ \t\n]*(<dropIndex indexName="([a-zA-Z0-9_]*)"[a-zA-Z_-=" ]*>)%\1<changeSet author="\2" id="\3a">\n\1\1<preConditions onFail="CONTINUE">\n\1\1\1<indexExists indexName="\5" />\n\1\1</preConditions>\n\1\1\4\n\1</changeSet>\n\1<changeSet author="\2" id="\3b">%g
}

# Problem to solve: Liquibase diff reads from type FLOAT not the precision but the the required bytes adds
# this information to its change log. A FLOAT (with default precision of 126) is transformed to a FLOAT(22),
# which is too short. By removing all precision information in the generated diff file, we avoid this
# problem.
/type="FLOAT\(/ {
    s%type="FLOAT\([0-9]*\)"%type="FLOAT"%g
}

测试环境验证 #

• Oracle Linux 7.9：脚本运行正常，成功生成输出文件
• Oracle Linux 8.9：触发错误，输出文件为空

错误原因与修复方案 #

错误根源 #

报错行（第11行）的正则表达式中，字符组[a-zA-Z_-=" ]*存在问题：在GNU sed的新版本（OracleLinux 8.9搭载的版本更高）中，字符组内的-若处于两个字符之间，会被解析为字符范围定义符。此处-_试图定义从_（ASCII码95）到=（ASCII码61）的字符范围，这是一个无效的反向范围（起始字符ASCII码大于结束字符），因此触发"Invalid range end"错误。而OracleLinux 7.9的sed版本对这类无效范围检测更宽松，未触发报错。

修复方法 #

修改第11行正则中的字符组，通过以下两种方式避免-被解析为范围符：

1. 将-移至字符组的开头或结尾：

   s%([ \t]*)<changeSet author="([a-zA-Z \(\)]*)" id="([a-zA-Z0-9-]*)">[ \t\n]*(<dropIndex indexName="([a-zA-Z0-9_]*)"[a-zA-Z_=" -]*>)%\1<changeSet author="\2" id="\3a">\n\1\1<preConditions onFail="CONTINUE">\n\1\1\1<indexExists indexName="\5" />\n\1\1</preConditions>\n\1\1\4\n\1</changeSet>\n\1<changeSet author="\2" id="\3b">%g
   

2. 用反斜杠\转义-：

   s%([ \t]*)<changeSet author="([a-zA-Z \(\)]*)" id="([a-zA-Z0-9-]*)">[ \t\n]*(<dropIndex indexName="([a-zA-Z0-9_]*)"[a-zA-Z_\-=" ]*>)%\1<changeSet author="\2" id="\3a">\n\1\1<preConditions onFail="CONTINUE">\n\1\1\1<indexExists indexName="\5" />\n\1\1</preConditions>\n\1\1\4\n\1</changeSet>\n\1<changeSet author="\2" id="\3b">%g
   

修改后重新执行脚本，即可在OracleLinux 8.9上正常运行。

内容的提问来源于stack exchange，提问作者Uwe A