If outputting text with ANSI controls, the Junit formatter will create invalid XML. For instance:

@test "my-test" {
  printf "\e[1mhello\e[0m"
  return 1
}

Produces this output (with properties redacted)

<?xml version="1.0" encoding="UTF-8"?>
<testsuite name="my_test.bats" tests="1" failures="1" errors="0" skipped="0" time="0" timestamp="2020-05-12T15:01:25">
    <testcase classname="my_test.bats" name="my-test" time="0">
        <failure type="failure"><![CDATA[
(in test file test/my_test.bats, line 106)
  `printf "\e[1mhello\e[0m"' failed
�[1mhello�[0m]]></failure>
    </testcase>
  <system-out></system-out>
  <system-err></system-err>
</testsuite>

The CDATA section includes the control character literal � (Unicode: 0x1b), which are invalid XML characters. This is causing some XML parsers to fail parsing the resulting file.

I think maybe there's two ways to handle this:

1. Encode the escape character (as \e or \033 or something)
2. Strip all control characters from the captured output