CRLF to LF line ending conversion is applied depending on whether body is ASCII-only or not.

[manueljacob]

I was debugging an error in a test suite that does checks on mails sent by ActionMailer. Mail::Body#to_s returned lines with LF or CRLF line endings depending on whether the body was ASCII-only or not. I tracked it down to the following:

Mail::Body.new("test a\r\n").to_s # => "test a\n"
Mail::Body.new("test ä\r\n").to_s # => "test ä\r\n"

The first line uses Mail::Encodings::SevenBit, which converts line endings, while the second line uses Mail::Encodings::EightBit, which does not. Commit f8b57f6 changed Mail::Encodings::EightBit to stop converting line endings. Although #1325 reports similar symptoms, it appears to be a different issue, as that one was introduced between versions 2.7.0 and 2.7.1.

Is this the intended behavior?

For my use case, it would be beneficial if there was a way to get the e-mail body with consistent line endings (probably LF) regardless of whether the body happens to contain a non-ASCII character or not. It seems to me like a better heuristic whether to convert line endings would be whether the content type is “text”.

In my test suite, there are instances of Mail::Message (returned by ActionMailer::Base.deliveries.last), on which I call .body.to_s. That’s probably anyway not quite what I want, as it doesn’t take into account the charset of the body. There is Mail::Message#decode_body_as_text, which is probably what I should use instead. It seems like that method would be a good place to apply the CRLF to LF conversion.