>>>>> Eric Abrahamsen writes:

    > On 04/29/24 21:39 PM, Daniel Semyonov wrote:
    >>>>>>> Mattias Engdegård writes:
    >> 
    >> > 29 apr. 2024 kl. 18.42 skrev Michael Albinus <michael.albinus@gmx.de>:
    >> >> Eric Abrahamsen <eric@ericabrahamsen.net> writes:
    >> >> 
    >> >>> This is still failing the tests, and I can't see why, I suspect the
    >> >>> reason is in the "... ..." below:
    >> 
    >> > No, you should ignore that and look at the text
    >> 
    >> >> The following options might have problems:
    >> >> variable: gnus-valid-select-methods
    >> >> value: (...)
    >> >> type: (...)
    >> 
    >> 
    >> > that precedes the ERT error. In this case, it appears that the
    >> > top-level list includes an element
    >> 
    >> >   ("nnatom" address)
    >> 
    >> > which doesn't match the declared type which requires that the
    >> > string be followed by one of {post, mail, none, post-mail}.
    >> 
    >> 
    >> This should be changed to `("nnatom" none address)', and the docstring
    >> should probably be changed to clarify this too.

    > That's what I've changed it to, and it was still failing, that's why I
    > was trying to dig more information out of the failure.

    > HOWEVER. Moments ago I thought to look at nnatom.el itself, and this bit
    > at the bottom:

    > (gnus-declare-backend (symbol-name nnatom-backend) 'address)

    > Also ends up setting gnus-valid-select-methods, and needs the equivalent
    > change:

    > (gnus-declare-backend (symbol-name nnatom-backend) 'none 'address)

    >> Honestly though, I don't understand why this is a user option in the
    >> first place, it only seems useful if you're implementing a new backend
    >> (in which case, you're probably using `gnus-declare-backend', which
    >> modifies `gnus-valid-select-methods' without checking the value anyway).

    > And here you mention that exact fact :)

    > I agree it doesn't make much sense, it doesn't do anything for the user.
    > I wonder if `gnus-declare-backend' was originally just meant to be used
    > by out-of-tree backend libraries.

Maybe the Gnus manual should indicate that `gnus-valid-select-methods'
should be modified directly for built-in backends (currently it only
describes using `gnus-declare-backend' in Hooking New Backends Into Gnus).

    > Anyway, it's redundant, but no great harm done, and the tests pass.

Thanks.
I also realized there is an example in nnfeed.el which is wrong in the
same way, I've attached a patch which fixes it.

Daniel