(this is actually quite fun for me - taking my mind off of mundane end of year firefighting and getting philosophical - thanks for this thread)
> First, with <pc> plus <sc>/<ec>, you're give them two
> concepts to implement instead of one.
In my mind this statement is too incomplete to tell the whole story. For me, it helps to add a bit more for completeness:
With <pc> plus <sc>/<ec>, you're give them two concepts to implement instead of one. While each could technically be used for all use cases, each is optimized for one case, and ugly for the other.
Case 1 (90 percent), an inline contained in a single segment.
<pc> is good. <sc>/<ec> is less good because it adds complexity (using empty elements to artificially emulate start and end tags is ugly - especially glaring when it is in a single segment).
Case 2 (10 percent), an inline starts in one segment and ends in a different segment
<sc>/<ec> is good. <pc> is less good because it adds complexity (extra attributes representing start and end are ugly).
So my thinking is to exclusively choose one or the other means there will always be an ugly solution being used in one of the use cases. By offering both, but stipulating (in the conformance clause) that (a) to use <pc> in Case 2 is a violation, and (b) to use <sc>/<ec> in Case 1 is a violation, we avoid ugliness. And we do not give a lazy/sloppy implementer (who want to pass the conformance test) a way to mess up.
And if we do need to choose just one, I would choose the one that is ugly only 10 percent of the time.
I completely agree with the rest of your argument about human nature.
> I guess I missed the XML philosophy decision while
> reviewing the past TC activity. Can you give me any pointers to it?
Sigh, I searched for about 20 minutes yesterday on the TC list. I tried all kinds of string searches ("XML Philosophy," "Operating Principle," "Principle ballot," I even tried misspelling (in case it was me who took notes that day) "Operating Principal"). While I found all kind of interesting results, I did not find the record of this vote. I considered leaving it out, but I remembered that Rodolfo recently referred to this vote as well. So if he can find the record of that vote (which I know I'm not dreaming), we'll have it to refer to. If he cannot, I will formally withdraw that part of my rant (and apologize for the extra bandwidth). Rodolfo, do you remember where the record of this vote can be found?
I have a bit of romantic admiration for Ted Nelson, though he's certainly been off the mark on many things. He once said, "the web is what we were trying to prevent"! Maybe a little out of touch with the real world. :-)
Let me restate my point about implementer sloppiness. First, with <pc> plus <sc>/<ec>, you're give them two concepts to implement instead of one. That is inherently more complicated. Second, the first concept is simpler to understand and covers 9x% of cases; the second requires more explanation and is much less common. Which do you think implementers will spend their time on and get right? So I think that in the real world <sc>/<ec> will be poorly understood and badly implemented by XLIFF processors, no matter how well you state the requirements or try to enforce conformance. It's just my gut, but based on my experience with localization standards.
I admit there is some appeal to <pc>. However, I would note that the inclusion of dispEnd in the <pc> start tag (as I understand is part of the <pc> proposal) looks ugly to me, because it separates the end markup from where it logically belongs. That's enough for me not to be attached to <pc> on aesthetic grounds.
I guess I missed the XML philosophy decision while reviewing the past TC activity. Can you give me any pointers to it? I found one message from you titled "Philosophical Goal for XLIFF: design for the 'lowest common denominator'". But I can't find the follow-up or the decision.
On Wed, Dec 14, 2011 at 4:30 PM, Schnabel, Bryan S <firstname.lastname@example.org> wrote:
I think it's great that you are hitting the ground running. Super cool that you bring your thoughts to the table - and extra credit to you for being so organized, concise, detailed and coherent (and citing Ted Nelson was a nice touch).
You make it a tough debate, but I have a few thoughts from the other point of view.
Redundancy is not always evil - and is sometimes needed. Painters only *really* need a tube of Blue, a tube of Red, and a tube of Yellow - and they can mix any color they want. But giving them Purple, Orange, and Green isn't so bad I think. Or more to the point, in HTML, since you have <b>, <i>, <br> and <font size="x">, you don't really need <h1>, <h2>, <h3> etc. But giving web designers these extra <hx> tags probably is a good thing.
I think characterizing <pc> as entirely redundant does not cover every use case. Yes, I can use <sc>/<ec> to model all cases. But processing <sc>/<ec> adds overhead in many cases. I might even say in most cases (as you say "Overlapping markup may not be the common case . . ."). I think the vast majority of inline cases could be handled by <pc> (though it's sometimes said that I live in an idyllic world where all XLIFF I deal has the beautiful luxury of being for well-formed XML - I wish!). In my mind <pc> has less overhead, and more application. Forcing all cases to use <sc>/<ec> is to force the ugly solution of the few onto the many.
I think we can mitigate the risk of the lazy implementer's sloppiness by stating a very clear conformance requirement. Use <pc> when you can, and <sc>/<ec> when you must. Actually, I'm not sure how that leads to sloppiness. I don't think it's at all difficult to know when to use one or the other.
I'm glad you brought up the XML-friendly matter (my turn to get philosophical). A while ago the TC came to a unanimous decision that we adopt an operating principle that we try to model features and processing in a way that is (to the extent that we can) compliant with XML philosophy. I don't think any of us were motivated by a love for XML, or an expectation that all, or even most of what XLIFF processes will be XML source. For me, it was because XML is a pretty simple (rational) common denominator. And I think some have even pondered the X in XLIFF. I think it is more than to say that the vocabulary of our Localization Interchange File Format is XML. I think nearly all Open Standards choose XML as their vocabulary. I personally have always look at the X to represent the methodology more than the vocabulary. So even though I did not coin the phrase, I tend to agree with the statement that we should not mandate a single method that is "XML hostile." In fact, XLIFF 1.2 was nearly not passed because of the way one of the OASIS voting members read our characterization of XML.
And to pick one final nit, I don't think we all necessarily agree on the specific necessity of <sc>/<ec>. I think, more accurately, we all agree that we need to accommodate spans that either cross segments or overlap. Plenty of examples have been shown that can support this requirement by adding attributes to either <pc> or <ph>. But I think most of us see that as kind of a bad idea. And speaking for myself, it seems as bad an idea as using <sc>/<ec> in cases where it simply isn't needed.
Just a different point of view . . .,
As a newcomer, I am not yet eligible to vote, but this is a chance to start
sharing my views. So here's my position on this ballot:
- It looks like most everyone agrees on the necessity of <sc>/<ec>. So if
<pc> were also included, it would be entirely redundant. Everything that
can be expressed with <pc> can also be expressed with <sc>/<ec>, and <pc>
becomes a "shorthand" for some forms of <sc>/<ec>. I think this is bad
design for several reasons:
- Consider how you would handle this as an implementor. When reading
XLIFF, you would almost certainly map both constructs to the same one
internally. When writing XLIFF, you have two choices: 1) always write
<sc>/<ec>; or 2) figure out when you can use <pc> and use write <pc> in
those cases, otherwise write <sc>/<ec>. Practically, why would anyone
do (2)? The result is everyone writes <sc>/<ec>. <pc> is useless
- Lazy implementors (and they're all lazy) are likely to implement <pc>
right, and <sc>/<ec> sloppily. This is a serious practical problem, and
if you think about the industry's experience with standards, I think
you'll agree that implementors routinely do the easy parts and skip the
hard parts. We're much more likely to get correct implementations if
there is only one way to do it. (And of course we offer good developer
- Imagine a future specification for "XLIFF normalization" (which will be
necessary someday). The obvious thing to do is normalize <pc> to
<sc>/<ec>. So <pc> is just extra work.
- I'm not entirely against shorthands. They are good for human-writable
documents, and for saving space. But I don't think either consideration
- Those who like <pc> for XML seem to agree that <sc>/<ec> are necessary
even for XML, in some cases. So nobody gets to live in a fairy land
without <sc>/<ec>. Why pretend? ;-) Maybe you can go a little way with
just <pc>, but then some day you have a use case that requires <sc>/<ec>.
You'll curse the day you started with <pc>!
- "XML friendliness" has little value for me. XLIFF is designed for
localizing any content, and while we should make sure it works well for
XML, I would be wary of any bias to working "best" with XML. Besides,
there's enough work between us and XLIFF 2.0 that giving special attention
to XML doesn't seem to be a good use of effort.
- (Ok, I can't resist veering into the philosophical.) I don't think
<sc>/<ec> is hackish at all. The right question to ask here is whether
overlapping markup has value in the real world. I think the answer is
obvious. Consider a word processor. They user selects words 1-3 and
presses "bold", then selects words 2-4 and presses "italics". How should
the word processor represent this? Overlapping markup is clearly a
reasonable option, and nested markup starts to look like a hack.
Overlapping markup may not be the common case, but it's not a freak
either. Don't treat it like one!
Some ideas borrowed from Ted Nelson:
On Tue, Dec 13, 2011 at 8:44 AM, Schnabel, Bryan S <email@example.com> wrote:
I wanted very much to attend this meeting to add my point of view on the "Type of codes Decision on what set of codes to use: a) ph/pc/sc/ec, or b) ph/sc/ec, or c) ph/pc, or d) ph."
But an unexpected work emergency kept me otherwise occupied (sounds strange that one could have a Web CMS emergency - but so it goes . . . ).
Though the attendees spoke in favor of b) - I just wanted to add that there was also a strong argument made against b) in the email thread (and in the interest of full disclosure, I am included in that camp). See
So I am grateful that the action is to have a ballot.
From: firstname.lastname@example.org [email@example.com] On Behalf Of Yves Savourel [firstname.lastname@example.org]
Sent: Tuesday, December 13, 2011 8:18 AM
Subject: [xliff-inline] XLIFF Inline Markup Subcommittee Teleconference - Dec-13-2011 - Summary
XLIFF Inline Markup Subcommittee Teleconference
Present: Fredrik, Yves, Andrew-P, Christian, Arle.
The summary of our main items is here:
Draft is under SVN, here:
(This will be replaced by DocBook version soon)
=== Type of codes
Decision on what set of codes to use: a) ph/pc/sc/ec, or b) ph/sc/ec, or c) ph/pc, or d) ph.
Fredrik: pc/sc/ec means you have to both types to handle -> more work for implementers.
And one cannot avoid sc/ec so that one is not optional.
Simpler and clearer.
Andrew-P: agree with Fredrik: redundant representation is more dangerous for various reasons.
Christian: no input.
ACTION ITEM: Yves will create a ballot.
=== Annotations representation
(req 1.5: Must allow to associate spans of content with metadata)
Current proposed representation:
<mrk [id='id'] type='type' [value='some value'] [ref='someId'] [translate='yes|no']>...</mrk>
--- Overlapping mrk:
-> split them and merge them as needed.
May require an attribute to make the distinction between lone consecutive annotations and split annotations.
ACTION ITEM: Yves to try implementation.
Fredrik: actual best representation may depend on whether there is pc or not.
--- annotation with only translate attribute.
What type value to use?
-> 'generic' (with only translate attribute allowed in that case?)
Fredrik: yes, understood that last time.
Christian: Is it wise to create dependencies between attributes and values etc?
This is tricky. Leads to schema question and also checking implementation for conformance.
Fredrik: Kind of agree with that.
Maybe wording can be changed. E.g. if type='generic' then ignore other attribute is possible.
Christian: working with MAY can be tricky.
Andrew-P: probably ok to have the dependencies
Implementing conformance could be done.
Fredrik: We could say MUST ignore.
Yves was to bring the question of schema version 1.1 vs 1.0 to the TC.
-> Done here:
Andrew-P: Agree with Rodolfo: need probably more than a schema to do proper validation.
Fredrik: Would prefer to validate as much as possible with schema.
Avoid to rely on a specific tool.
Christian: Maybe this is a general guidelines: Try first with 1.0, and if you can't: document it.
Being able to rely only on 1.0 would be very nice.
Fredrik: But then we would need to re-introduce things like <it>
Arle: Validation for TBX was hard with 1.0. so we used RNG
Fredrik: yes key of 1.1 is the better support for conditions, etc. But 1.1 is not final.
ACTION ITEM: Yves to bring that back to TC.
=== What about crc and assoc?
Those are not needed was last meeting consensus.
Yves had the action item to ask to TC if anyone is using crc or assoc and report on current status for those.
-> Done here:
Consensus would be to just drop then in 2.0.
ACTION ITEM: Yves double check presence of assoc.
=== Uniqueness of attribute names (added)
Andrew-P: Context based seem fine.
Fredrik: Agree too.
ACTION ITEM: Yves to escalate the question to TC.
Fredrik: Note on 'id' attribute. How to define uniqueness within a scope.
For example trans-unit id unique within <file>, etc.
We need to be very clear on those definitions.
Christian: what about xml:id?
Yves: xml:id needing to be unique at document level doesn't fit our id requirement.
Christian: Possibly have different namespaces for different type of ids. E.g. different prefix.
Would make check easier.
=== Representation of editing hints
Need to have notation for 'permissions'
Current permission we thought about:
1: a code can be deleted
2: a code can be replicated
3: a code can be moved out of its original order
4 (possibly): a code can be moved outside its original parent span
All were to think about representation for hints.
Arle: Like the letter based representation.
Fredrik: Skeptical about bit-field or letters.
Second option better (human readable)
But problem when extending it. For example with default value.
One attribute per info allows flexibility in extending things.
Discussion on how to represent defaults for new info.
Andrew-P: Using one attribute for each info is simpler/non-ambiguous
Yves: that list should be more or less done with for 2.0
Could also use new attributes for new info
Fredrik: possible addition: can be edited
Arle: hopefully this should be a close list.
=== Bi-direction markers
Several tools use Unicode characters in 1.2.
Goes against W3C recommendation but seems common practice.
-> need to know about other tools
-> ask for info to W3C
Sent (but not passed the W3C check yet).
Some reference material:
Arle: asked the same question to Felix and Mark, both were against Unicode chararacters.
Reason: bad because they are 'invisible'. Tags are visible.
Andrew-P: experience with implementing Unicode chars: it works well.
Fredrik: in some case xml 'looks' wrong. But with using proper editor things are ok.
I've heard argument about using tags. Basically it's for editing xml with non-rendering editor.
=== Codes and effect on segmentation
(req 1.16: Inline codes should have a way to store information about the effect on the segmentation)
=== Representation of added codes
Need to clarify use of duplicated code.
e.g. why not re-use same code?
=== Any other business
--- I assume Andrew-P is in Pacific time zone.
That would make now 3 members for 7am.
Do we need/is it possible/ to shift the time of the SC 1 hour later?
Fredrik: fine with me.
Arle: ULI meeting is at that time.
Yves: will have to stay at this time for now then.
To unsubscribe, e-mail: email@example.com
For additional commands, e-mail: firstname.lastname@example.org
To unsubscribe, e-mail: email@example.com
For additional commands, e-mail: firstname.lastname@example.org