> Making the change in mm-url.el sounds scary: that code was written
> many years ago, and who knows where it is used?

Yeah, that makes sense. I'll see what I can find about that, at least
within emacs itself.

> The reason for testing bolp there is not explained, but I'm guessing
> they didn't want to craete an empty line?

Not sure. It looks like it was introduced during a patch intended to fix
submitting binary data.  Would it make sense to CC the two authors of the
following commits?

I dug into the history a bit to try to find out where =(unless (bolp)=
came from

#+begin_src bash
git log -L '/(unless (bolp)/,+1:lisp/gnus/mm-url.el' > unless-bolp
#+end_src

It looks like it was introduced in this commit

#+begin_quote
commit fca2f70380dcb054497470aaf8eda6173063928e
Author: Kenjiro Nakayama <nakayamakenjiro@gmail.com>
Date:   Mon Nov 10 22:33:55 2014 +0100

    Allow uploading files from eww
#+end_quote

and the "\r\n" was included unconditionally before the boundary

#+begin_src elisp
   ;; use the boundary as a separator
   (concat "\r\n--" boundary "\r\n")
#+end_src

Then it this commit changed it

#+begin_quote
commit a6e0188dffc394698d9ffbef50401f14a31c8722
Author: Lars Ingebrigtsen <larsi@gnus.org>
Date:   Thu Oct 13 21:39:29 2016 +0200

    Fix problem with submitting binary data via HTTP forms
#+end_quote

to the following

#+begin_src elisp
 (unless (bolp)
      (insert "\r\n"))
#+end_src

plus a bunch of "\r\n"s scattered throughout the different conditions.
It's not clear to me how the behavior of the function changed.

It looks like maybe instead of the initial diff I proposed, something
like this would produce more accurate output for file uploads, and also
change the surrounding code as little as possible,

#+begin_quote
diff --git a/lisp/gnus/mm-url.el b/lisp/gnus/mm-url.el
index 11847a79f17..686dea20b6a 100644
--- a/lisp/gnus/mm-url.el
+++ b/lisp/gnus/mm-url.el
@@ -430,7 +430,8 @@ mm-url-encode-multipart-form-data
              (insert filedata))
             ;; How can this possibly be useful?
             ((integerp filedata)
-             (insert (number-to-string filedata))))))
+             (insert (number-to-string filedata)))))
+         (insert "\r\n"))
         ((equal name "submit")
          (insert
           "Content-Disposition: form-data;
name=\"submit\"\r\n\r\nSubmit\r\n"))
#+end_quote

But it seems like it would be better to do a larger rewrite that
consolidates the places where "\r\n" is added before the non-initial
boundary.

In order to figure that out, I will read the two versions more closely
to figure out what the differences are and write some ERT tests showing
the difference.  Hopefully that will clear things up.