A path through the thicket

[Daniel Strebe]

Vladimir:

I appreciate the effort you put into these responses. I apologize for this very lengthy response. Here is a synopsis for those who need to ration their reading time:

1. Rebuttal of the thesis that the situation with Web browsers argues for more imperatives in this specification.
2. Why we won't be able codify the useful possibilities for composite fonts.
3. Why the recipe creator's "intent" is not what is important, but rather the consuming author's intent that is important.
4. A composite font recipe is not content, it is a resource; therefore the requirement of fidelity to its instructions depends upon context.
5. Examples of composite font usage to clarify when fidelity is/isn't useful.
6. My interpretation of Vladimir's proposal, juxtaposed against my proposal. (You might find this instructive.)
7. More responses to specific arguments posed by Vladimir.
8. A list of objections to Vladimir's proposal.
9. Conclusions

1. Browsers as an analogy to composite fonts

I do not agree with Vladimir's analysis of the situation with Web browsers today or the failures of the W3C recommendations, and therefore I do not agree with the conclusions Vladimir draws. The World Wide Web was not conceived to deliver final-form presentation. We could go into great depth on this matter, but it is peripheral to the discussion, so I will leave it by averring that many Web designer vexations arise out of the natural tension between their need for design and the Web's need to deliver content on displays of unknown resolution and unknown color space, on devices of unknown computational capacity, in unknown window sizes, and using possibly unavailable resources-like fonts - a problem that is still not solved 15+ years later.

To a fair extent the reality of these unknowns forced specifications that allow choice in implementation along with soft failure. I do not claim that no mistakes were made; I do not know enough about those standards to know what mistakes were made and how grave. But my understanding of their purpose does not match the premise of Vladimir's argument well.

2. On codifying useful possibilities for composite fonts

To return to my elaborate example, Marina's hypothetical problems would not have been alleviated by "better" standards. It is pointless for Web browsers to implement typographical features when the designer's desired font generally isn't even available on the viewing computer. A standard can't do anything about copyright restrictions. If and when legal, commercial, and technical remedies arrive on the scene to deliver the fonts content designers want to Web pages, then surely we will see browsers supporting typography with ever greater sophistication.

Conversely, if standards had dictated such typographical sophistication from the outset, despite its uselessness, the barrier to entering the browser market would have been higher, and expensive effort would have been diverted from more useful functionality elsewhere. This is why I contrived such an elaborate scenario. Marina's quandary arises from conditions that exist now. They did not exist three years ago because there was no RIA framework for her to use. They probably will not exist three years from now because better typography probably will arrive in browsers and RIA frameworks once the type delivery service she subscribed to becomes commonplace.

Meanwhile, we, sitting here now comfortably at our computers contemplating conditions as they exist now, will find ourselves unable to enumerate all the uses people will find for composite fonts even today, let alone five years from now. Therefore I do not understand why we should fossilize our ignorance in a specification. What we do know, or can puzzle through, is what kinds of things people are likely to want to communicate with composite fonts. Therefore we should provide a language for this communication - but we should leave the actions of the communicators up to those who know what they are trying to achieve.

3. Why the recipe creator's "intent" may not be important, and what is

To address an implicit axiom in Vladimir's arguments, the intent of the composite font creator is not the only intent in play. What we have been calling a consumer - because it consumes a composite font recipe - may also be a content author. This "consuming author" has an intent we know nothing about, but by definition creative forces are in play. Whom are we helping if we stifle those creative forces by shoving them into the only channels we foresaw or that we felt were "important"? To the human authoring that content, our specification means nothing and the composite font recipe means opportunities. That human wants the software "she" is using to do what she expects it to do. How that software adapts an incoming composite font recipe to its purposes is not our business. It is not even the business of the recipe creator. It is the business of the creative mind operating the software because what comes out of that creative mind is original content. The font is just a tool in creating that content.

The consuming author has a reason for consuming composite fonts. Therefore the consuming author must be interested in what a composite font recipe has to say. That means the language the composite font speaks must be clear, but it does not mean the consuming author is interested in everything the recipe has to say. A font is not content. It is a tool or a resource. It may contain elements extraneous to the content for which it is being exploited. We must not confuse ourselves into lumping composite font recipes into the same category as MPEG videos or PDF, where fidelity of content must be preserved because fidelity of content is the purpose.

A font contains instructions. The purpose of the instructions is for the font to be used usefully. What is useful varies from circumstance to circumstance. The circumstances are innumerable.

4. On fidelity to the recipe, and when it is/isn't important

I do not discount the need for fidelity, but we need to be clear about what kind of fidelity we are talking about and why it is important. If Vladimir's argument is that consumers must retain the fidelity of the recipe creator's intent, then I completely disagree. For one thing, it's frankly none of the recipe creator's business how the tool "he" created gets used insofar as other interests are not harmed. For another, the creator will be unable to express all the ways he might be willing for the recipe to be used, both because we will not provide a language so complicated and because creators won't want to burden themselves with the infinite permutations of pedantic intent that consumers might find useful. Both the creator and the consumer would wonder what the point is, and the creator is probably even less likely to envision all the useful scenarios than we are. Just because a creator forgot to mention that he's fine with his composite font being used just for its glyph mappings is no reason someone shouldn't use it for that.

When does fidelity to the recipe creator's intent need to be preserved? When the recipe itself becomes part of the instructions for reproducing content. In other words, when the consuming author's content relies on the composite font recipe for whatever form of fidelity it needs. Since what is meaningful to preserve varies from content category to content category, and since neither we nor the recipe creator knows the consuming author's intent, it follows that we cannot be in the business of dictating what portions of a composite font recipe a consuming author's software is allowed to honor or discard.

5. Examples of when fidelity to a recipe is/isn't important

    Example A, page layout software: In page layout software, the composite font recipe will be relied upon to create consuming author's content. Moreover this reliance likely extends to the most minute details of glyph placement. Most content authors will expect a composite font to work the same regardless of what page layout software or word processor they use. Because page layout software's purpose is typographic fidelity, and because the composite font recipe's creator is likely to know more about the typography of the fonts in the recipe than the authors of the page layout software (who probably don't even know about the component fonts specifically), a normal implementation would be a complete one.

In practice, we are likely to see page layout software either not implement composite fonts or else implement the specification in its entirety. Partial implementations, at least as far as body text goes, would confuse content authors and reflect unfavorably on the page layout software publisher. They could result in imported content getting formatted differently from the original. Still, there may be interesting uses for partial implementations outside of body text.

    Example B, markup languages: In markup languages that include presentation semantics, some kinds of formatting are provided for while still conforming to environmental constraints. Normally this means line breaks are absent in the content, but some kinds of character and block formatting are present. Their presence ranges from fairly abstract ("header") to fairly concrete ("underline"), depending upon the particular attribute and and upon the particular markup language. Some provide for font categories ("Serif"); some provide for font names.

A composite font could provide for typographic and formatting control beyond the scope or intent of a renderer of the markup language. One could argue for nearly any range of composite font support in this context. If the renderer is a markup editor in "source" view, typographic needs might not exist. A composite font might get selected for some purely semantic reason, such as glyph complement, with no intent or need for typographic control because the editor's reason for choosing the composite font is only to be able to render any glyph expressed within literal strings.

If the markup editor provides a "Preview" mode to see the content formatted, how much typography the mode should enable depends on what expected target systems will be able to render. If the markup language does not even provide a mechanism to designate the specific font for rendering, then clearly the "preview" amounts to little more than a "mock-up", and the need to support various features of composite fonts becomes highly questionable. Transmission fidelity means nothing because the font is not part of the markup.

If, on the other hand, the specific font can be declared in the markup, then the markup editor ought to enable any typographic features that do not conflict with the markup itself. We are in no position to decide what that means in the general case. We may want the rendering engine to honor the small-caps instructions in the composite font recipe, for example, but the markup language or the style guide for the renderer may dictate that all-caps must be used. Or we may declare that a compliant renderer honor the glyph mappings of the composite font, but this markup includes mathematical expressions. The renderer is happy to use the composite font recipe for text, but it knows a lot more about mathematics than we do and than the generic markup editor did, so it always renders formulæ according to a specific font it knows behaves according to its own precise layout needs.

It should be clear from these examples, including the earlier elaborate example of Marina and her cartographic project, that we can contrive plausible subsets of functionality ad nauseum, and therefore that we will never anticipate all the diverse, legitimate uses of composite fonts.

6. What the contrasting proposals look like

What Vladimir's proposal looks like to me:

    Recipe creator states "his" intent:
        1. You must shift baseline of Font A 6 design units upward with respect to Font B.
        2. You must shear glyphs of Font C 4° rightward from base to top with respect to other fonts in the recipe.
        3. You must map Unicode range W-X to Font C for Latin script, X+1-Y to Font D for Chinese, Y+1-Z to Font E for mathematical symbols...

    We inject our intent here:
        1. You must implement.
        2. You may ignore if you declare yourself to be a partially compliant renderer according to Compliant Subset D.
        3. You may ignore if you are a Latin-script-only renderer according to Compliant Subset E.

    Consumer compares her intent against recipe creator's intent and our intent:
        1. I am in violation because I do not need to shift baselines because I do not mix component fonts on a single line.
        2. I need the design characteristics to match, so I shear glyphs.
        3. I am in violation because I always use Cambria Math for mathematical symbols because I can't lay out equations reliably using arbitrary fonts.

with the result that the recipe creator was unable to state his real intent because his real intent was merely to describe how the components of the composite font can work together to achieve the illusion of a single font, not to dictate what features of the composite font must be used, since he has no idea what the consumer needs anyway. The consumer, whose implementation suits her purposes precisely, does not comply with any recognized subset. Output downstream gives idiosyncratic results because the specification requires her to pass through the composite font recipe unmodified (in order to honor the recipe creator's intent and ignore her use of it).

What my proposal looks like to me:

    Recipe creator states his observations that:
        1. Shifting baseline of Font A 6 design units upward with respect to Font B gives uniform baselines.
        2. Shearing glyphs of Font C 4° rightward from base to top with respect to other fonts in the recipe gives matching oblique angles.
        3.     Mapping Unicode range W-X to Font C gives correct results for Latin scripts.
            Mapping X+1-Y to Font D for Chinese gives correct results for simplified Chinese.
            Mapping Y+1-Z to Font E gives correct results for mathematical symbols.
            ...

    We inject no intent.

    Consumer compares her intent against recipe creator's observations and our intent:
        1. I do not need to shift baselines because I never mix composite font components on a line.
        2. I need the design characteristics to match, so I shear glyphs.
        3. I always use Cambria Math for mathematical symbols because it is impossible to lay out equations reliably using arbitrary fonts.

with the result that this consumer got exactly what she wanted without violating the specification, without violating the creator's intent, and without harming anyone's interests. All output downstream from her deployment behaves correctly because she only passed through the portion of the composite font recipe that she actually deployed... or because her output is in final-form, not dependent upon anything external to the final-form document.

7. Responses to some specific arguments posed by Vladimir

I sympathize with Vladimir's argument that implementers will look to the specification for guidance even when they only implement part of it. However, I am unable to come up with realistic scenarios that require "interoperability" at partial levels. In scenarios I have managed to contrive, partial implementations are generally dead-end consumers or else their output would always be final-form. Neither scenario requires adherence to some canonized subset, nor does it seem like the situation would confuse implementers, since they have a clear idea of what they need. To ensure fidelity downstream when exporting its content, Partial Implementation A should simply excise those portions of a recipe that it did not use. (I would go so far as to say we should require this.) This way any consuming implementation that is a superset of Partial Implementation A (such as a full implementation) will deliver fidelity to the content created by Partial Implementation A. The odds that downstream consumer Partial Implementation B is both a partial implementation and carries the similar needs and intent as Partial Implementation A is remote unless the two already know about each other in some private or semi-private contract. Therefore our intervention adds little value, if any.

For example, Vladimir notes "...We can specify one subset that would be sufficient for authors creating content utilizing simple scripts only". But if we look at the sorts of functionality we propose for composite fonts, none of it really applies to simple versus complex scripts. Whether a given implementation will work with complex scripts depends on its general font and language handling machinery, not upon whether it supports some feature of composite fonts. Meanwhile "simple script" support means different things to different layout engines. We are not going to convince an established layout engine to improve or change its implementation of "simple scripts" just to gain "compliance" with whatever we decide is meaningful - particularly when our "requirements" actually turn out to be orthogonal to executing the instructions of a composite font recipe.

Vladimir states, "[Consumers] must be able to scale fonts and/or individual glyphs and their metrics to ensure correct layout... I believe that since all existing font engines today support scalable fonts and are capable to do this, the requirement does not present any additional burden on implementations." To which I respond that code must be written to interpret the semantics of the scaling, adjust positioning, and to instruct the font engine to perform the scaling - and even then coding is only a portion of the cost of deploying an implementation. Meanwhile, if such scaling is irrelevant to the content the consuming author creates, who benefits by requiring it?

8. Ten objections to Vladimir's proposal

To reiterate: I am happy for this committee to prepare a standard for a "complete" implementation. I think that will be of the most interest to the most people, and it covers the most demanding case: typographic fidelity. I am not at all happy to canonize some subset of partial implementations. My objections are:
    1. We will unable to enumerate all the useful possibilities.
    2. We do not have the resources to labor over those possibilities we would come up with.
    3. We do not have the wisdom to winnow down the possibilities into some manageable collection that suits (almost) all needs.
    4. Whatever list we might come up with would be fragile in the face of technology and market forces.
    5. Any given partial implementation will not have a large enough constituency for us to expend such effort on.
    6. Canonized subsets distract and dilute our ability to encourage and "enforce" the complete typographic case.
    7. We will invite confusion over the meaning of recipes by condoning a special subset of exceptions to their imperatives.
    8. By considering the recipe's instructions as imperatives at all, we exclude useful possibilities the recipe otherwise implies.
    9. We will stifle innovate uses which themselves do not harm other interests.
  10. We will be injecting our own intent into areas we have no business doing so.

9. In conclusion

We should allow partial implementations to support those aspects of the specification that they need. We should instruct that partial implementations must alter the recipes they consume to reflect their requirements for fidelity in the content they generate. We should instruct anyone intending for broad interoperability to implement the complete specification. We should not add to the specification such requirements for font feature support that are orthogonal to the business of executing the instructions in font recipes.

Regards,

- daan Strebe
Senior Computer Scientist
Adobe Systems Incorporated
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.aau.at/pipermail/mpeg-otspec/attachments/20090930/ddeadf83/attachment.html>