This fixes#94 and basically reverts d0f0435. As James points out correctly in #94, we should not assume specific responses from the server. As long as the spec is followed and the server returns the correct SMTP code, we should not do our own magic.
I've also extended the `getTestConnection` method in client_test.go, so that we can specify more test environment options like `TEST_PORT` and `TEST_TLS_SKIP_VERIFY`. This was needed for testing with the ProtonMail Bridge which listens on a different port and has non-trusted certificates.
As proposed by @iwittkau the `SendError` type now has a `IsTemp()` method as well indicating to the user if the delivery error is retryable or not.
Since we want to use it in the error response from the Client functions like `Send` or `DialAndSend` we need to return the SendError type not only as part of the `*Msg` but also as return value for these methods. Hence, the changes made for #85 been overhauled to return the new error type instead. In the pre Go1.20 version of the `Send()` method we need to return an accumulated version of the SendError type, since we don't have `errors.Join()` and therefore, if more than one error occurred during the delivery we return an ambiguous error reason since we can't tell which of the captured errors is main error. For more details the user can always check the `*Msg.SendError`
Not that this particular part of the code is performance critical, but I figured that the `strconv.Atoi()` is actually useless in here.
Since all we want to know is if the error code from the SMTP server is a 4xx error, we can just check the first rune of the returned error. The `Atoi` provides us with no advantage over the simple rune compare (except of taking about 3ns longer to execute)
Did a complete overhaul of the senderror.go.
- the list of `errors.New()` has been replaced with constant itoa error reasons as `SendErrReason`. Instead, the `Error()` method now reports the corresponding error message based on the reason.
- The `SendError` received a `Is()` method so that we can use `errors.Is()` for very specific error checking. I.e. we can check for `&SendErrors{Reason: ErrSMTPMailFrom, isTemp: true}`. This provides much more flexibility in the error checking capabilities
- A `isTemp` field has been added to the `SendError` type, indicating whether the received error is temporary and can be retried or not. Accordingly, the `*Msg` now has a `SendErrorIsTemp()` method indicating the same. The decision is based on the first 3 characters returned from the SMTP server. If the error code is within the 4xx range, the error is seen as temporary
- A test for the SendError type has been added
Since the send error indicate an error during the mail delivery, in my opinion it should be reset when a re-send is initiated, so that the senderror field always represents the latest delivery error. The send error should be checked after mail delivery and before a retry is started.
This PR introduces the `SendError` type which implements the error interface.
A new `senderror` field has been added to the `Msg` as well, so introduce this type to it.
I've also added different error variables that indicate the different things that can go wrong during mail delivery. These variables can be checked for, for each `Msg` using the `errors.As` method
The `Error()` method of `SendError` will return a detailed error string on why the `Msg` could not be delivered.
Additionally, `HasSendError()` and `SendError()` methods have been added to `Msg`. While `HasSendError()` simply returns a bool in case a `Msg` failed during delivery, the `SendError()` will return the full `SendError` error interface.
This allows disabling the Noop command during the dial. This is useful for servers which delay potentially unwanted clients when they perform commands other than AUTH.
`Client.Send()` provides the possibility to send multiple `*Msg` in one go. If one of the `*Msg` caused an error with the sending mail server, we were returning completely, while not processing any `*Msg` that came after the failing message.
This PR fixes this behaviour by processing each message first and then return a accumulated error in case any of the `*Msg` processing failed
Additionally, this PR separates the `Client.Send()` method into two different versions. One that makes use of the new `errors.Join()` functionality that is introduced with Go 1.20 and one that handles it the old way for any supported version lower than Go 1.20
This PR introduces two major changes:
* SetHeader and SetHeaderPreformatted have been deprecated in favour of SetGenHeader and SetGenHeaderPreformatted
As pointed out in #80 the naming was pretty confusing, given that we already have SetAddrHeader. With the new naming convention it should be more clear. For compatibility reasons the old methods have been kept for now but in reality they are just aliases to the new methods
* GetAddrHeader and GetAddrHeaderString have been introduced
As requested in #80 analogous to GetGenHeader we also need a similar method for the address headers. Since address headers are *mail.Address pointer, we've also added a *String method that will extract the address string and return a string slice instead
Additionally we're introducing methods for the actual address headers: GetTo, GetFrom, GetCc and GetBcc (with a *String counterpart as well). This way the user has full flexibility. Either they use the more "low-level" GetAddrHeader method or the higher level methods for the corresponding address type
The error handling in the msgWriter.writeBody() method was not working properly. We basically overwrote the mw.err with nil if the function that followed after a failed write attempt was successful again
This patch fixes#81
With the SetHeaderPreformatted() method we have the ability to set headers that are already preformatted by the user and will not be altered in the mail message output
For middlewares to be able to access the fully written mail message, we need a way to execute WriteTo without the calling middleware to be handled, otherwise we end up in an infinite loop
Therefore, this PR introduces the MiddlewareType and the corresponding change of the Middleware interface. We now require to return the MiddlewareType when the Type() method on the interface is called
This way we can also introduce the WriteToSkipMiddleware method which takes a MiddlewareType as argument. This will allow us to use a WriteTo call with the initiating Middleware to be skipped
