Stefano Mazzocchi wrote:
David Crossley wrote:
Stefano Mazzocchi wrote:
Bertrand Delacretaz wrote:
Reinhard Poetz a ?crit :
At http://wiki.apache.org/cocoon/CocoonDocumentationSystem I propose the architecture of a new extensible documentation system...
Thanks Reinhard - to me this looks really good, and the fact that you're backing your proposal with actual work makes all the difference ;-)
Touche'.
I've tried with Doco but I never put it up with the expectations :-(
What can I say: go for it!
Well, some of the ideas and discussion from Doco would evolve into this system. Thanks for that much, it all helps.
Glad to hear that.
Yep, your proposal and the discussions helped a lot :-)
... and not to forget over the last 10 days, Upayavira helped me with many talks.
I have only one problem with Reinhard's approach: open editing.
And not because I fear defacing but I do fear link spamming bots.
Please, do not leave the comments open for non-humans or it's going to be hell over again (at the very least let's use captchas)
good point.
I also want that every comment (not only doc changes) has to be approved by a committer. So we have a double-barrier (hope that's understandable English) for spamming bots.
Ok, captchas + human moderation is clearly too high of a barrier for spammers and even for defacers. Even infra@ would not have a problem with that.
Two question to Stefano (and everybody else): I proposed numbers as document IDs? What do you think about this?
I used to be a fanatic of 'readable URL'... but I think they present more problems than they solve.
First of all, the encoding is a pain. It's fine for english, but until we ave IRI (internationalized resource identifiers, think "unicode meets URI") support forget chinese, japanese, cyrillic, hebrew, korean and so on.
One normaly solution is to have an english title even for non-english pages. I dislike that, it's very anglo-centric.
Second, people like Nielsen argue that readable URLs are easier to use and to remember. I think it's bullshit. Not even my bookmarks satisfy me anymore in terms of link management (del.icio.us + google killed my browser bookmarks), do you really think I would type in or remember any URL today? nonsense.
There are a few values of a readable URL. The first is actionable breadcrumbs.
So, if you find yourself in
http://site.com/a/b/c/d/e/f/g
you can automatically infer something like
site.com > a > b > c > d > e > f > g
and, for shitty web sites, that is a *tremendous* navigation help. For URLs like
http://site.com/page/39884984
that's it, there is no hierarchical context that you can infer from it.
Now, we will not have a shitty web site, so this argument doesn't apply and Amazon (which is the most used e-commerce site in the world *and* has the worst URL space ever imagined!) shows that URL-space design does not impact usability, if the pages don't require so.
Actually, since geeks are used to hack into URLs but normal people do not, having a flat or bad URL space forces usability people to think about navigation in the page and not outside.
Another argument, and probably more important, is that a flat URL structure gives a sense of 'wikiness' that people have come to dislike.
Now, again, this is a false impression (inspired by a plethora of bad practices rather than effectual technological limitations) but a strong one nevertheless (I do feel the same about it at times).
But *exactly* because of that, I think we should be brave and show the world that a flat URL space *does not* automatically yield 'wiki-like' flat spaces that are extremely painful to navigate.
Flat numeric URL spaces have also extremely interesting advantages:
- pages can have their titles adjusted without impacting persistance (links are more solid over time)
- pages can be rearranged/repurposed/re-aggregated/re-used without impacting persistance
Another question is the structure of URLs - the new efforts of Sylvain who wants to provide some docs in French needs some thinking where to put them. I propose
http://c.a.o/ ............... editable global docs (own repository)
http://c.a.o/fr/ ............. editable global docs in French (own repository)
http://c.a.o/2.2/ ............ editable docs of 2.2 (own repository)
http://c.a.o/2.2/fr/ ......... editable docs of 2.2 in French (own repository)
http://c.a.o/2.2.1/ .......... "frozen" docs of the 2.2.1 release
http://c.a.o/2.2.1/en/ ....... "frozen" French docs of the 2.2.1 release
I don't think we should have frozen docs at any time, they are included in the distributions anyway and those distributions will be persisted for the longest time.
Sun did this with the Java API did this and created a mess, people linked to java/1.4.2/ and then 1.4.3 was created and all links broke down.
If a document shipped in 2.1.3 has a bug and was fixed in 2.1.4, why would anybody want to see it? and if 2.1.4 removed something useful for 2.1.3, that's a bug and we should fix it in the doc, rather than make everything available on the web.
So I'm -1 on this.
As for french docs, I *strongly* think that we should do this thru content-negotiation rather than URL design. A person accessing the page with a french browser will get the page in french, that's all they have to know (and the page will have a series of flags that will trigger an overload in locale, but that's going to be a parameter of the URL, not part of it).
The language a page is written, just like the data-type of the page, should not belong in the URL.
This makes the URL space way more "solid" overtime: I can link to
http://cocoon.apache.org/2.2/3984948
and *be sure* that it will be there a few years from now and, by then, maybe a translation in my native language would have poped up!
let's be brave!
-- Stefano.
