Discussion List Archives

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Proposal to regulate markup in CIF files

There having been no further comment regarding the proposal below, I will move it forward by
(1) giving the CIF syntax Vol G authors the go-ahead to include the character encoding conventions as part of the CIF syntax chapter
(2) adding the 'Marked-up text' type (or alternative name) to the list of possible types in DDLm
(3) suggesting to IUCr journals that they develop a markup specification. Let me know if you would like to be part of that discussion.


On 21 September 2017 at 10:59, James Hester <jamesrhester@gmail.com> wrote:
Dear All,

I've turned the proposal into a discussion document, separating non-ASCII character markup and other markup as per John's comments. The best way forward for 'other markup' is not clear to me, so I've asked a few questions at the end of the document that I invite you to consider.

I've also inserted replies to John's comments inline below.


On 14 September 2017 at 01:08, Bollinger, John C <John.Bollinger@stjude.org> wrote:

Dear Colleagues,


I think the proposed approach of allowing markup convention to be specified in data files could be useful.  Moreover, the proposal would provide a mechanism for formalizing the specification of which items are subject to markup in the first place.  Supposedly, that is determined by items’ definitions, but in practice, few, if any, current dictionaries or definitions actually address the topic explicitly.




1. To the extent that the proposal envisions data files being enabled to self-specify a particular markup convention from among several choices, it seems to violate the principle that the meaning of an item should not depend on the value of a different item.

In my opinion,this principle needs to be clarified.  For example, any non-key data name 'depends' on the value of key data names in its category, or the meaning of a fractional coordinate 'depends' on the values of the cell parameters. We need a precise rephrasing of the principle, e.g. "no new key data names may be added to a pre-existing category" or "where possible, the meaning of data names should depend only on data names that are identifiers".  Once this is clarified we might be in a better position to judge when a proposal is suspect.  Does anyone know the underlying basis for this principle?


2. Since the codes for various supported markup conventions would be defined in domain dictionaries, it seems we might be setting ourselves up for future issues in the event that dictionary maintainers want to support new markup conventions, for then we will need to change existing definitions (which, granted, we have lately afforded ourselves some freedom to do).

The point being that software written with one set of markup conventions in mind would be caught unawares by data values written according to a new markup convention. However, the idea is that _publ.markup_convention is used to flag the use of a new convention and allow such software to act appropriately.  Adding new enumerated values is a normal way of developing dictionaries and I don't think is controversial.  I have removed this from the proposal for now.


3. Additionally, the proposal seems to imply that each domain dictionary would need to either specify its own analogue(s) of `_publ.markup_convention`, or else to explicitly depend on the Core item.  I am uncertain whether this is a good or bad thing.

Yes, this is true.  All of our dictionaries currently depend on the core dictionary in any case and this is unlikely to change. A completely different domain would need to define its own analogue.


4. The proposal does not preclude individual items from being defined to have whatever content type is desired for them, as long as the definitions are not flagged as carrying `Marked-up` data.  That’s good, but I can imagine hypothetical cases in which it would be confusing to dictionary authors.  For example, if an item were defined to contain HTML markup, it would be necessary for its definition to specify that its values were NOT `Marked-up`.

This is true.


5. Although we call the existing CIF conventions “markup” – and that’s fitting – for the most part they don’t serve the same purpose as Markdown, reStructured text, etc..  Almost all of our markup is aimed at encoding specific characters, whereas the other markup conventions mentioned are focused on document structure and styling.  These are largely orthogonal considerations, so perhaps we should approach them separately.

I have done this in the revised proposal.


6. The discussion accompanying the proposal distinguishes between items that are intended to be machine-actionable and those intended purely for human consumption, with the assertion that only the latter kind should be subject to markup.  I would accept that if it referred to structural markup only, but it is not so clear that such a rule is appropriate for markup that serves to encode characters.  Consider, for example, `_atom_site.label`.  As a key data name, it certainly has machine significance, but its values are also meant to identify atoms to humans.  Should CIFs, then, be forbidden from using `Cα` (i.e. `C\a`) as or in atom labels?  Forbidding use of markup would prevent literal `Cα` from being expressed in an atom label in a CIF 1.1 document, but not in a CIF 2.0 document, so that sets up a pathway wherein markup gets introduced into data values through format transliteration from CIF2 to CIF1.  If only the original document were considered valid in such cases, then that would constitute a rather nasty trap to set for ourselves.


Agreed, hopefully the rewritten proposal addresses this.


Initial analysis:


As presented, the proposal’s largest impact would probably be to provide for DDLm dictionaries to specify which items are primarily (or exclusively) intended for human consumption: those that are defined to have `Marked-up` content, regardless of whether any markup is actually present in their values.  Initially, at least, this would establish in which values to interpret the standard CIF markup conventions.  That would be worthwhile.


I am less certain about the prospects for or usefulness of enabling data files to select alternative markup conventions.  Perhaps that could be used to good effect to support revisions to the markup conventions.  Perhaps it would be more broadly applicable.  But perhaps we should avoid making any items’ meanings depend on other items’ values.

I have rewritten the proposal to remove the option of specifying particular types of markup, and indeed largely removed discussion of markup pending some feedback from this group.

Revised discussion paper for regulating markup of CIF text items


1. Data values containing backslash escapes for indicating non-ASCII
   characters are to be considered entirely equivalent to values
   obtained after all such substitutions and escapes have been applied.
2. Other mark-up (superscripts, subscripts, italic etc.) is given a
   new type, but is otherwise unspecified and needs to be discussed.


From the very first publication describing CIF, markup conventions
have been provided in order to extend the range of characters and font
effects representable in ASCII.  Which data values these conventions
might apply to, and whether or not this is more properly a CIF syntax
or dictionary (semantic) issue, has been left implicit.

Marked-up text according to the ad-hoc definitions described in Vol G
appears both in CIF data files and in dictionary definitions. While
COMCIFS has control over the conventions applying within dictionaries,
it has far less control over data values in data files, which are
produced both by dedicated software, such as publCIF, and hand-editing
or local ad-hoc solutions.  Marked-up text in data files plays an
important role in the publication workflow.

Vol G (First Edition) notes in section "It is hoped that in
future different types of such markup may be permitted so long as the
data values affected can be tagged with an indication of their content
type that allows the appropriate content handlers to be invoked". It is
not, however, clear that multiple alternative markups are desirable.

Moving forward

The markup in use can be divided into two classes: 'character encoding'
and 'font effects'.  Under this proposal, each class is treated differently.

1. Character encoding

   Character encoding markup represents non-ASCII letters using a
   backslash followed by one or more ASCII characters, for example,
   '\a' is 'greek letter alpha'.  This is a format-specific method of
   allowing access to the full characterset used by the DDL textual
   types. From the point of view of the (format-agnostic) dictionary
   data model, how a particular format wishes to encode characters is
   irrelevant. Therefore, the set of character escapes is most
   appropriately documented as part of the description of CIF syntax,
   not within a DDL dictionary.  In other words, CIF1/2 data values with
   backslash character escapes are semantically identical to CIF2 data
   values where those escapes have had their Unicode equivalents
2. Font effects

   Font effects differ from character encodings in not having a DDLm
   type that they are a concrete realisation of. By analogy, then, we
   could create a DDLm type 'Marked-up text', whose contents contain
   marked-up text.  Particular implementations and syntaxes might then
   specify what particular convention(s) 'Marked up text' should
   conform to.


1. An important function of the 'Marked-up text' type is to
   designate data values that are not intended to be machine-actionable.
   No DDLm functions or attributes are envisioned for manipulating the
   markup. The type could alternatively be something like 'Rich text'.
2. Enumerated values and identifiers must not be of type marked-up text.

3. The 'marked-up text' data value is obtained from a CIF syntax
   file after backslash character codes have been substituted.

Open questions

The above proposal does not specify a particular markup convention. Leaving
anything unspecified is dangerous for a standard, as it invites the appearance
of multiple, incompatible solutions.  We should as a matter of urgency answer
the following questions:

(1) Should alternative markup conventions be possible?
(2) If yes, should the markup convention in use be
    (i) per dataname?
    (ii) per datavalue (maybe via an embedded flag)?
    (iii) per data block?
    (iv) per dictionary?
    (v) per syntax? (e.g. CIF/CIF-JSON/HDF5 etc.)
(3) If no, the current convention is the only possible one for reasons of
    backward compatibility. Should it be:
    (i) a feature of CIF syntax?
    (ii) a feature of CIF syntax when combined with a DDLm dictionary?
    (iii) defined in DDLm?

My answers to these questions would be
(1) No alternatives should be possible, in order to simplify publishing
    workflows and maintain the publCIF investment
(3) (ii)

Some explanation regarding (ii), which possibly sounds a bit
abstruse. A CIF syntax file can be used (in theory) with an
alternative dictionary language and associated data model. Likewise,
DDLm dictionaries can be used to describe non-CIF files.  In each
case, the way in which syntactical data values are constructed to
match the dictionary types may differ (for example, numbers may be a
text string or binary).  Each combination of syntax and dictionary
must explicitly state how each dictionary type is represented in that
syntax. So I am suggesting that for the combination 'CIF + DDLm' that
we specify the current markup conventions to represent type 'Marked-up

T +61 (02) 9717 9907
F +61 (02) 9717 3145
M +61 (04) 0249 4148

T +61 (02) 9717 9907
F +61 (02) 9717 3145
M +61 (04) 0249 4148

Reply to: [list | sender only]