On Fri, Aug 23, 2013 at 5:12 PM, janI <[email protected]> wrote: > On 23 August 2013 21:11, Rob Weir <[email protected]> wrote: > >> On Fri, Aug 23, 2013 at 12:21 PM, janI <[email protected]> wrote: >> > On 23 August 2013 17:58, Rob Weir <[email protected]> wrote: >> > >> >> (Responses to [email protected], please) >> >> >> >> Obviously our website is quite large. Google reports 21207 pages >> >> indexed in the www subdomain, and a further 48075 pages in the wiki >> >> subdomain. But for purpose of this post, when I talk about the "home >> >> page" I'm talking about the contents of our main index.html and the >> >> most commonly visited pages directly linked to it, e.g., the >> >> why/download/product/get-involved, etc. pages. >> >> >> >> This core homepage content amounts to around 25 pages. >> >> >> >> Today this content is scattered around the content tree. Some of it >> >> is in the root. Some of it in /why and /download directories. Some >> >> of it is template-related and is in /templates rather than in >> >> /content. >> >> >> >> As a test I tried to create my own NL page, in the fictitious "xx" >> >> locale. You can see it here: http://www.openoffice.org/xx/ >> >> >> >> It is not working correctly, but it already required a lot of >> >> non-trivial hacking: >> >> >> >> 1) I had to hunt around and guess which files to copy. Do I copy >> >> scripts, images and CSS, or just content pages? Some of the >> >> directories had out-dated content that was not linked to my anyone. >> >> It was hard to figure out what the minimum amount of content needed >> >> was, and where it was located. >> >> >> >> 2) The main index.html file had to be edited to refer to CSS in the >> >> root, rather than current directory >> >> >> >> 3) Download page is missed up, missing CSS and/or scripts. >> >> Presumably I need to copy something into the xx/download dir, or edit >> >> scripts to make them refer /download off the root. >> >> >> >> 4) The /xx/why pages are not showing the right side navigation now. I >> >> must have missed something there as well. >> >> >> >> Of course, I could figure the above out eventually. It just requires >> >> some time and effort and trial and error. But none of this is >> >> documented, and even if it were this is a fragile approach and >> >> probably beyond th web development skills of a typical translator. >> >> >> >> But we do know this has been done for some languages. They got it to >> >> work. The German page is a good example: >> >> >> >> http://www.openoffice.org/de/ >> >> >> >> Now this looks good, but it is still a messy thing from a maintenance >> >> perspective. If we make structural changes to the main English page, >> >> then those changes need to be manually merged into to every NL page. >> >> >> >> What can we do to improve this? >> >> >> >> Here's my idea: >> >> >> >> 1) What if we refactored the home page so it was all self-contained >> >> into these directories: /scripts, /styles, /images and /en/? >> >> >> >> 2) Make the /en directory be pure content. Only the stuff that needs >> >> to be translated. It loads everything else, scripts, images, etc., >> >> via URLs relative to the root, e.g.., in /scripts, /styles, etc. >> >> >> >> 3) Reduce or eliminate any embedded Javascript within pages. For >> >> example, refactor the code in download/index.html so it is external >> >> and depends on JSON resource files for translated strings. Aim so >> >> translators never need to touch script. >> >> >> >> 4) Ultimate goal is for someone to be able to jump start a new NL home >> >> page by simply requesting an svn copy of the /en directory, and then >> >> editing the resulting files. No one should ever need to do what I'm >> >> doing with the "xx" pages. >> >> >> >> 5) Maintenance is far easier. Most things like changing the scripts, >> >> is done in one place only. But even changes to the HTML are easier. >> >> Since we then have a common branch point via the svn copy, when >> >> structural changes are added to the main /en HTML, these can be merged >> >> in more elegantly to the translated versions, using Subversion. >> >> >> >> 6) Via Apache redirects we can ensure that the default call to >> >> www.openoffice.org/ goes to /en/. Conceivably we could also do locale >> >> detection and send requests automatically to the appropriate NL home >> >> page. >> >> >> >> A variation on the above would be to use Pootle, rather than svn >> >> copy/merge to maintain the translations. But that would require the >> >> same refactoring work to enable it. And it would require further >> >> investigation to identify a way of extracting and merging translation >> >> strings in MDText files as well as (X)HTML files. >> >> >> >> This is obviously more than a one-person task. So I'd be interested >> >> in hearing what you think in general about this approach, whether >> >> there is a simpler alternative I've missed, and whether this is >> >> something you'd be interested in helping with. >> >> >> > >> > I like a lot of your ideas, let me add my own experience. >> > >> >> Thanks. >> >> > If the our pages do not contain text, but that is totally outsourced in >> one >> > or more json objects, then translation becomes easy, and the pages >> themself >> > stay simple. when the url is called without arguments the "en-json" is >> > used, and if called with lang="xx" "xx-json" is used. >> > >> >> I like the idea of content/code separation. We certainly do this is >> the code, for example. But two challenges to taking this approach all >> the way with the website. >> >> 1) If we do JSON everywhere then we have a Javscript dependency >> everywhere. This has an impact for visibility of the pages to search >> engines, but there are workarounds. But it may be a bigger issue for >> users who block Javascript. >> >> No we would do that solely on the server side, it would not be a good idea > to have JS retrieve the json objects. > > We could eg. use php, that retrieved to correct json object, and > transmitted a finished page. >
OK. We're on the same page. > >> 2) There may be cases where a translation requires direct access to >> the HTML or CSS. For example, I think the Tamil translation needed >> access to specify a specific font. And for some languages they might >> need to set text direction to RTL. These kinds of things make almost >> any approach more complicated. >> > > Look at e.g. our mwiki that handles those details all on server side. > > And just as a suggestion, if we were to use wordpress, things like fonts > would be solved. WP also have a possibility (not json) for multi language, > which I could easy adopt in genLang (for translation). > > >> >> So the question we need to answer is how far we take this? I think >> we have some examples where the code is so intertwined with the text >> that translation becomes very hard and risky. For example, the >> generation of the "boxes" on the download page. But then we have >> some other pages, especially the MDText pages, where I would be >> comfortable handing it directly to a translator and expect they could >> edit it without breaking anything. >> > We can always find examples where it becomes hard, but typically you can > reformulate the problem so it fits in a standard (boxes are no real > problem). The only problem I see is with JS, where are ask and get answers > e.g. YN. > > >> >> The Javascript dependency might be broken if we make this be a CMS >> build-time text replacement rather than a runtime/Javascript >> replacement. So the CMS would detect when the Pootle files change and >> automatically generate new HTML pages from them. But even then we >> still would need some runtime integration of strings, specifically on >> the download page where language and OS are determined at runtime >> based on browser request headers. >> > I would consider not to use cms, because we basically dont need it. > So within the realm of server-side software, what is possible? My impress was that Infra generally cautions against runtime server side execution due to the greater opportunity for security problems. That's why we're using build-time page generation. This approach also performs well, since it is static HTML pages at runtime, and is very stable. But in any case, I think the refactoring work is approximately the same thing regardless of how the pages are generated. -Rob > rgds > jan I. > >> >> -Rob >> >> > If we use json objects, then pootle becomes an elegant tool for >> > translation, because it knows how to handle xml, and if we want to stay >> > with po files its about 1 day work in genLang. >> > >> > A number of top companies (incl. the one I used to work for) do it like >> > this, they of course then hire a translator to translate the json >> objects. >> > >> > Splitting functionality and text is the key, when thats done the rest is >> > trvial work. >> > >> > This will of course make cms a bit top kill, but I can live with that :-) >> > >> >> >> >> >> > rgds >> > jan I. >> > >> > >> >> >> >> Regards, >> >> >> >> -Rob >> >> >> >> --------------------------------------------------------------------- >> >> To unsubscribe, e-mail: [email protected] >> >> For additional commands, e-mail: [email protected] >> >> >> >> >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: [email protected] >> For additional commands, e-mail: [email protected] >> >> --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
