Re: Wrapping an Emacs dynamic module in a package

[Stefan Monnier]

> 1) I wrote an emacs module _mymod.so, as well as wrappers mymod.el (with
> some local setup) and mymod-pkg.el (which only has a call to
> define-package).

FWIW, I doubt you wrote _mymod.so.  Most likely you wrote something like
_mymod.c and then you asked a C compiler and a linker turn that into
a _mycmd.so.

Yes, it's a nitpick, but when you send your bug-report, please try to
use precise wording, because it's often surprisingly easy to introduce
major confusion otherwise.

> In my messages buffer, I see that Emacs tries to generate autoloads for the
> .so, and fails with the error indicated in the original email ("_mymod.so:0:0:
> error: scan-error: (Containing expression ends prematurely 238020
> 238021))").

In your bug-report, please show the corresponding backtrace.

> 1) Should Emacs be able to generate autoloads for an .so?

No.  It shouldn't scan those files for autoloads.

> 2) Does anybody have any experience with embedding an emacs module in an
> emacs package? Any best practices I should be aware of?

You might very well be the first to try it out.

Note, tho, that the ELPA format is a tarball containing .el files,
i.e. *source* files.  Similarly, it should hold _mycmd.c rather than
_mycmd.so.  Then again, package.el doesn't know how to turn that
_mycmd.c into _mycmd.so.   You can probably work around that problem by
adding something like

    (eval-when-compile
      (call-process "cc" nil nil nil <args>))

in your mymod.el (so that when the .el file gets byte-compiled, it
triggers compilation and linking of the C module).

But yes, including the .so *should* happen work (barring bugs) as long
as you're lucky enough to use the package on the right OS with the right
Emacs version, on the right architecture.


        Stefan