bug#21724: 25.0.50; (void-function (setf epg-context-armor))

bug#21724: 25.0.50; (void-function (setf epg-context-armor))

[Katsumi Yamaoka]

References: <http://thread.gmane.org/gmane.emacs.diffs/132036>

On Tue, 13 Oct 2015 09:36:28 +0900, Katsumi Yamaoka wrote:
> On Sun, 11 Oct 2015 21:03:07 +0000, Nicolas Petton wrote:
>>     * lisp/gnus/auth-source.el (auth-source-epa-make-gpg-token): Replace an
>>       usage of `epg-context-set-armor' with `setf'.
> [...]
>> -    (epg-context-set-armor context t)
>> +    (setf (epg-context-armor context) t)

> This doesn't seem to be expanded properly when building Emacs:

> In end of data:
> gnus/auth-source.el:2145:1:Warning: the function `(setf epg-context-armor)' is
>     not known to be defined.

> How do we fix it?

I got:

Debugger entered--Lisp error: (void-function \(setf\ epg-context-armor\))

Recipe:

$ emacs -Q
load: lisp/gnus/auth-source.elc
eval: (auth-source-epa-make-gpg-token "foo" "bar")

To fix it, epg.el(c) is required when compiling auth-source.el,
I think.  Though once I merged the change to the Gnus git master,
I reverted it thereafter, since old Emacsen that Gnus supports do
not have the setf-method for epg-context-armor.

Regards,

bug#21724: 25.0.50; (void-function (setf epg-context-armor))

[Katsumi Yamaoka]

On Wed, 21 Oct 2015 18:03:38 +0900, Katsumi Yamaoka wrote:
> References: <http://thread.gmane.org/gmane.emacs.diffs/132036>

>>> -    (epg-context-set-armor context t)
>>> +    (setf (epg-context-armor context) t)

> Debugger entered--Lisp error: (void-function \(setf\ epg-context-armor\))

> To fix it, epg.el(c) is required when compiling auth-source.el,
> I think.

First I tried adding: (eval-when-compile (require 'epg))
It fixed the setf compilation problem.  However, the byte compiler
got to issue a new warning as follows:

In end of data:
auth-source.el:2145:1:Warning: the following functions might not be defined at
    runtime: epg-context-set-passphrase-callback, epg-decrypt-string,
    epg-encrypt-string

So, I'm not quite sure why it helps, but I wrapped the existing
autoload settings for those functions with `eval-and-compile':

<http://article.gmane.org/gmane.emacs.diffs/132162>

bug#21724: 25.0.50; (void-function (setf epg-context-armor))

[Glenn Morris]

Katsumi Yamaoka wrote:

> So, I'm not quite sure why it helps, but I wrapped the existing
> autoload settings for those functions with `eval-and-compile':

IMO it is better to move the epg autoloads after the epg eval-when-compile.

bug#21724: 25.0.50; (void-function (setf epg-context-armor))

[Katsumi Yamaoka]

On Thu, 22 Oct 2015 20:27:16 -0400, Glenn Morris wrote:
> Katsumi Yamaoka wrote:
>> So, I'm not quite sure why it helps, but I wrapped the existing
>> autoload settings for those functions with `eval-and-compile':

> IMO it is better to move the epg autoloads after the epg eval-when-compile.

Thanks.  I see why adding (eval-when-compile (require 'epg))
caused a new warning.  It marked all the epg functions including
those autoloads maybe non-existent at the run-time, didn't it?
Committed.

bug#21724: 25.0.50; (void-function (setf epg-context-armor))

[Glenn Morris]

Katsumi Yamaoka wrote:

> Thanks.  I see why adding (eval-when-compile (require 'epg))
> caused a new warning.  It marked all the epg functions including
> those autoloads maybe non-existent at the run-time, didn't it?

Yes. I didn't investigate, but it's probably a compiler bug.
It doesn't properly track autoloaded-but-not-actually-defun'd definitions
before eval-when-compile, so whatever statement comes last wins.