Max Nikulin <manikulin@gmail.com> writes:

>> I think we can try to do the following:
>> 
>> 1. Use the load-dir test, warn, and try to re-load
>> 2. Use org-assert-version, but downgrade error to simple warning
>
> Ihor, final decision is up to you.

> I think, `org-reload' may be recommended in a warning message. I am 
> unsure if `org-reload' called from the assertion while an Org component 
> is loaded may not cause a more severe issue. My special concern is 
> reloading when new Org is being compiled: is it safe or reloading should 
> be suppressed completely?

I am thinking about the following:
1. Use `org-reload' instead of load-dir test. This will re-load all the
   currently loaded Org libraries, forcing them to be consistent. This
   is necessary because Org does not only rely upon explicit (require
   ...) statements - we have way too many `declare-function' expecting
   other libraries to be loaded by other means.
2. Run `org-reload' _after_ loading current file.
   Because (equal (org-release) ,(org-release)) will only ever trigger
   during compilation when we (require 'already-compiled-library),
   running `org-reload' after loading 'already-compiled-library will
   also ensure consistent Org state after (require
   'already-compiled-library) is finished.

> I believe, during creation of .elc files, assert macro should signal an 
> error since a warning is not enough to prevent loading mixed compilation 
> result to next Emacs session. I do not like that currently during 
> compiling with old Org loaded, `org-assert-version' causes just a 
> warning. Users have difficulties with clean recompiling.
>
> While loading, non-fatal error may be a better option when versions do 
> not differ too much.

I do not think that we should throw and error. `org-reload' approach
should hopefully solve the problem with mixed versions. If not, not
throwing an error is still better because mixed compilation actually
works relatively fine in many cases - much better than throwing a hard
error from user perspective. And we will still notify the user that
something is wrong and that random failures are expected.

See the attached tentative patch for bugfix.

P.S. I am not sure how to approach this patch in a view of the Emacs
release process. This is a critical bug, so we aim for bugfix Org
release, but we do not want to make life harder for Emacs devs as well.