byte-compile-warn in do-after-load-evaluation - a bug?

byte-compile-warn in do-after-load-evaluation - a bug?

[Alan Mackenzie]

Hello, Emacs,

in master (at least), in do-after-load-evaluation (subr.el), the code
uses byte-compile-warn to output an error message.  This seems wierd,
since there's no mention of the byte compiler anywhere near.

Might it be that somebody not having a good day just used that function
thinking it was a general purpose warning function?

The problem with it is that it uses an implicit (compilation) source
position in the message it outputs.  There is no compilation going on at
the moment.

Unless I've misunderstood something, I propose to replace this call of
byte-compile-warn with simply message.

-- 
Alan Mackenzie (Nuremberg, Germany).

Re: byte-compile-warn in do-after-load-evaluation - a bug?

[Stefan Monnier]

> Might it be that somebody not having a good day just used that function
> thinking it was a general purpose warning function?

No, and I think the code makes it pretty clear:

    [...]
    (cond
     ((bound-and-true-p byte-compile-current-file)
      ;; Don't warn about obsolete files using other obsolete files.
      (unless (and (stringp byte-compile-current-file)
                   (string-match-p "/obsolete/[^/]*\\'"
                                   (expand-file-name
                                    byte-compile-current-file
                                    byte-compile-root-dir)))
        (byte-compile-warn "%s" msg)))
    [...]

The intention is to use this code when the load was performed by the
byte-compiler (presumably because of a `require`).

> The problem with it is that it uses an implicit (compilation) source
> position in the message it outputs.  There is no compilation going on at
> the moment.

I suspect that what we should do here is to make the byte-compiler
provide more info.  E.g. instead of just testing
`byte-compile-current-file`, which is a kind of "accidental" info, we
should have `bytecomp.el` do a

    (let ((byte-compile-triggered-load-source <sourcedata>))
      ...)

so `do-after-load-evaluation` can check this specific var and use its
info to provide proper source information.

WDYT?


        Stefan

Re: byte-compile-warn in do-after-load-evaluation - a bug?

[Alan Mackenzie]

Hello, Stefan.

On Fri, Jan 21, 2022 at 13:39:26 -0500, Stefan Monnier wrote:
> > Might it be that somebody not having a good day just used that function
> > thinking it was a general purpose warning function?

> No, and I think the code makes it pretty clear:

Sorry, yes it does.  I think it's me that's having the bad day.

>     [...]
>     (cond
>      ((bound-and-true-p byte-compile-current-file)
>       ;; Don't warn about obsolete files using other obsolete files.
>       (unless (and (stringp byte-compile-current-file)
>                    (string-match-p "/obsolete/[^/]*\\'"
>                                    (expand-file-name
>                                     byte-compile-current-file
>                                     byte-compile-root-dir)))
>         (byte-compile-warn "%s" msg)))
>     [...]

> The intention is to use this code when the load was performed by the
> byte-compiler (presumably because of a `require`).

I think I've got it sorted out, now.  By pushing an element onto
byte-compile-form-stack when handling a `require', we can ensure
byte-compile-warn will have something to chew on.
(scratch/correct-warning-pos branch.)

> > The problem with it is that it uses an implicit (compilation) source
> > position in the message it outputs.  There is no compilation going on at
> > the moment.

> I suspect that what we should do here is to make the byte-compiler
> provide more info.  E.g. instead of just testing
> `byte-compile-current-file`, which is a kind of "accidental" info, we
> should have `bytecomp.el` do a

>     (let ((byte-compile-triggered-load-source <sourcedata>))
>       ...)

> so `do-after-load-evaluation` can check this specific var and use its
> info to provide proper source information.

> WDYT?

I'm not convinced.  I think byte-compile-form-stack already provides (or
can provide) sufficient source position information.

>         Stefan

-- 
Alan Mackenzie (Nuremberg, Germany).

Re: byte-compile-warn in do-after-load-evaluation - a bug?

[Stefan Monnier]

> I'm not convinced.  I think byte-compile-form-stack already provides (or
> can provide) sufficient source position information.

Fair enough,


        Stefan