Re: [PATCH][oc-csl] Improve reference parsing

[András Simonyi]

Dear All,

On Wed, 2 Nov 2022 at 07:28, Ihor Radchenko <yantar92@posteo.net> wrote:

> I do not think that CSL limitations are really limiting us.
>
> - Allowing macros will be handled by ox.el itself automatically
> - Export snippets can also be processed without much issue (consider
>   direct LaTeX code)
> - inline-babel-call and inline src blocks may be useful with :exports
>   results when some auto-generation of text is needed. They will also be
>   handled automatically by ob-exp.
> - latex-fragments are either equivalent to direct LaTeX or to inserting
>   an image
> - timestamps could be exported as text, although I do not see any
>   obvious utility of timestamps inside references.

I'm not really familiar with the internals of the Org exporter but,
looking at the ox.el code, macros and babel calls are processed and
resolved before processing citations, so they seemingly have no
bearing on the org-cite-csl--parse-reference function my patch is
concerned with.

> However, oc-csl should not ignore the export processor to support all
> the above. I am not sure why you need a dedicated export processor
> instead of passing the string to current processor (or derivative)
> instead.

> If you really need to mark certain constructs specially for CSL, you can
> create a derived export backend for the current backend and replace the
> transcoders for the object types that must be treated specially.

Other than macros and babel calls, e.g., timestamps, LaTeX  fragments
etc. the problem is that citeproc-el expects and needs the affixes and
locator to be passed in the very limited html-like markup supported by
CSL (see https://www.zotero.org/support/kb/rich_text_bibliography for
a rudimentary description), and, crucially, the assumption is that
everything else is plain text, which, if necessary, will be escaped
according to the target format, i.e., '$' signs are escaped by
citeproc-el's own LaTeX formatter. The reason for this limitation is
that the affixes and especially the locator have to be parsed into
citeproc-el's internal rich-text representation for further processing
according to the used CSL style. (Affixes are only concatenated to
other elements but locators can be the subject of any type of
formatting.)  As a consequence, I think the only real alternatives are
using a custom backend as I do in the current patch or a backend
derived from the plain text Org exporter -- I don't have a strong
preference as to which solution we choose, just went with the
seemingly more minimalist option. (The proper way of dealing with
LaTeX fragments in this context, in particular with LaTeX math
fragments, would be to support those in citeproc-el's internal
representation and markup, which is planned but not implemented yet.)

> > +(defconst org-cite-csl--export-backend
> > +  (org-export-create-backend
> > +   :transcoders
> > +   '((bold . (lambda (_bold contents _info) (format "<b>%s</b>" contents)))
> > +     (code . org-cite-csl--element-value)
> > +     (entity . (lambda (entity _contents _info)
> > +                 (format "\\%s" (org-element-property :name entity))))
>
> Why :name, but not :html?

Good point, thinking about it a bit more, :utf-8 would probably be a
slightly better solution (in keeping with citeproc-el's 'plain text'
requirement), I'will change this when we will have sorted out the
other details.

best wishes,
András