We were only returning the first line of the error message,
but that may not be enough to get the whole context of the error.
Since we have already parsed the response, just return the whole error
to the user.
Related to #694
Rely on a generic transport trait and allow passing in one. This
will enable use cases where we don't have a real Tokio TCP stream,
or have to bind a specific source address before establishing
the connection.
In contexts where FIPS certification is mandatory, having the
ability to use the certified boring TLS library is sometimes necessary.
Added initial support for it, only one TLS toolkit can be enabled at
one time.
Instead of injecting spaces to ensure that lines stay under 76 characters only wrap at whitespace characters. This avoids changing the headers.
A best-effort to keep lines under 76 characters is still done, however it is only done at whitespace. Notably there is no hard wrap enforced. This means that it is possible for headers to break the 1000 character line-length limit in the specification. It is just hoped that the receiver will allow long lines in this case.
Closes#688
Co-authored-by: Kevin Cox <kevincox@kevincox.ca>
This adds a `local_address: Option<IpAddr>` parameter to the synchronous, and tokio connect functions.
(As far as I can see there is no current way to support this in async-std, because the library doesn't provide any way to do an async connect for an existing socket)
The email_address crate is more strict in validating user and domain
parts of email addresses. For example, it verifies the total size
of the local part, which the current method does not, and this has
caused upstream servers to fail to accept an email.
The header needs to be properly formatted so that Unicode characters are encoded the same way they will be in the final message. Previously the logical header value was being encoded.
A notable example is that a `'` in the `To:` header needs to be encoded. This was being encoded incorrectly.
