Re: Timestamps in ...-autoloads.el files

[Ludovic Courtès]

Alex Kost <address@hidden> skribis:

> Ludovic Courtès (2015-10-20 02:09 +0300) wrote:
>
>> If we diff as explained in the doc below, we see that the Git
>> discrepancies are due to timestamp in Perl’s POD files as well as
>> sorted-by-inode-number ‘tclIndex’ files.  For the Emacs modes, the
>> problem is the autogenerated autoloads files, which include some sort of
>> a timestamp as well.
>>
>> You’re welcome to help fix these issues!
>
> Thanks for the info!
>
> Those timestamps are inserted by `autoload-insert-section-header' and
> we can avoid them by advising this function, for example, like this:
>
> (defun guix-autoload-no-timestamp (fun outbuf autoloads load-name file time)
>   (funcall fun outbuf autoloads load-name file 0))
>
> (advice-add 'autoload-insert-section-header
>             :around 'guix-autoload-no-timestamp)
>
> So after putting this code into ‘emacs-generate-autoloads’ procedure
> from (guix build emacs-utils) module, there will be zeros instead of
> non-deterministic timestamps.

Great.

> However this will fix only those packages, that use
> ‘emacs-generate-autoloads’ directly or via ‘emacs-build-system’.  But
> there are also packages that generate autoloads on their own (for
> example, 'emacs-w3m' or 'guix' itself).  What to do for these ones?
>
> Perhaps we can make a special 'emacs-build' package (that will advise
> ‘autoload-insert-section-header’ function somehow) and use it as an
> input for emacs-packages, or are there other ways?

What about patching Emacs directly?

An upstreamable patch would be one that honors the ‘SOURCE_DATE_EPOCH’
variable¹.

Alternately, a patch that simply changes
‘autoload-insert-section-header’ to always use zero as the timestamp
would work as well, unless this would somehow break functionality.

Thoughts?

Thanks for looking into it,
Ludo’.

¹ https://reproducible-builds.org/specs/source-date-epoch/