This logic here:

does not work with all recover_line_* methods. Some methods use the parameter line, others use self.current_line. The overall logic is fine, but there is a major bug:

process collects all fails (self.__failed is appended with line) in the first loop (for line in self.parse(report))
In the second loop (for exc, line in self.__failed), recover_line is called with line.
If recover_line accesses self.current_line, the data is wrong, as self.current_line is then the last line of the report, not the actual one.

Unfortunately, simply fixing some recover_line_* functions is not enough, the process in self.process needs to be thought through and eventually adapted as well.

• self.current_line should be deleted after the parsing end to prohibit this error in the future.
• self.recover_line behaviour should be harmonized, making it applicable for use in self.parse_line and in self.process
• self.process should be investigated

In the future we also need better tests, but that's a bigger task and I'm afraid we can't stem that on a short term. And unfortunately the issue is important, as it leads to bogus (wrong/duplicated) data in the dumps and therefore loss of data.