#3299 - Add URI redirect for missing translations

#3299 - Add URI redirect for missing translations

This change is related to the previous commit for #3421.

After this change, If someone has manually chosen Mexican (es-MX), and they
surf to a page that does not have a Mexican, but a Spanish (es) version
available, they will be flipped to Spanish and not return when later surfing to
a page available in Mexican.

The existing behavior was to just show 404s in these cases, which would be
terrible experience with all those half-complete translation sets we have
currently.

We could check for an available translation on every request, and flip them back
to Mexican when possible, but that will be counter-intuitive to those who have
manually chosen a different language in the top-right selector on the page.

The optimal solution would be to keep their last chosen language stored in
session and say "You're currently surfing the Mexican site but this page only
has Spanish translation available", but that would mean revisiting privacy
policy and server mechanics, which is out of scope for this commit.

[f.nicolaisen, 2017-09-06 23:00:19]

https://codereview.adblockplus.org/29537689/diff/29537690/modules/web/templat...
File modules/web/templates/site.conf.erb (right):

https://codereview.adblockplus.org/29537689/diff/29537690/modules/web/templat...
modules/web/templates/site.conf.erb:66: location ~
^/([a-z][a-z]\_[A-Z][A-Z])(/.+) {
It would be cool if we could join both these locations into a single one, or
perhaps treat it under the / location above. Will look into that tomorrow.

https://codereview.adblockplus.org/29537689/diff/29537690/modules/web/templat...
modules/web/templates/site.conf.erb:69: rewrite ^/([a-z][a-z]\_[A-Z][A-Z])(/.+) 
$2 redirect;
Lots of duplication here :(

[f.nicolaisen, 2017-09-08 13:51:11]

Note that the PS2 patch was hotfixed to production yesterday.

In PS3, I haven't changed functionality, merely a bit more thought to the
comments.

I couldn't find a good way to reduce the regexp duplication. I also didn't want
to be consistent with the existing \w\w pattern because we always use upper case
letters for region (e.g. '/es_MX/'). Anyhow, we're far from doing anything like
the RFC instructs (https://www.w3.org/Protocols/rfc2616/rfc2616-sec3.html). 

There is also a silly case where if you go to the translated URI for a page that
does not exist anywhere, you end up with /firefox because of how the redirects
work in combination. Bit of a mess, really.

[mathias, 2017-09-08 19:20:21]

(Wrong reviewer addresses here again, fixed for now.)

https://codereview.adblockplus.org/29537689/diff/29539748/modules/web/templat...
File modules/web/templates/site.conf.erb (right):

https://codereview.adblockplus.org/29537689/diff/29539748/modules/web/templat...
modules/web/templates/site.conf.erb:60: location ~
^/([a-z][a-z]\_[A-Z][A-Z])(/.+) {
There's a newline missing in front of the bracket.

https://codereview.adblockplus.org/29537689/diff/29539748/modules/web/templat...
modules/web/templates/site.conf.erb:65: rewrite
^/([a-z][a-z])\_([A-Z][A-Z])(/.+) /$1$3 redirect;
This seems a bit quirky, it would move the user from `/aa-AA/` to `/aa/` and
thus remove "session" information. But then it seems like that quirky stuff is
actually how it's meant to be? The Hungarian ABP pages even link directly to the
`/en/` ones where a translation is missing.. https://adblockplus.org/hu/chrome

https://codereview.adblockplus.org/29537689/diff/29539748/modules/web/templat...
modules/web/templates/site.conf.erb:68: 
There's a newline missing in front of the bracket.

[f.nicolaisen, 2017-09-11 07:33:14]

https://codereview.adblockplus.org/29537689/diff/29539748/modules/web/templat...
File modules/web/templates/site.conf.erb (right):

https://codereview.adblockplus.org/29537689/diff/29539748/modules/web/templat...
modules/web/templates/site.conf.erb:60: location ~
^/([a-z][a-z]\_[A-Z][A-Z])(/.+) {
On 2017/09/08 19:20:20, mathias wrote:
> There's a newline missing in front of the bracket.

Acknowledged.

https://codereview.adblockplus.org/29537689/diff/29539748/modules/web/templat...
modules/web/templates/site.conf.erb:65: rewrite
^/([a-z][a-z])\_([A-Z][A-Z])(/.+) /$1$3 redirect;
On 2017/09/08 19:20:20, mathias wrote:
> This seems a bit quirky, it would move the user from `/aa-AA/` to `/aa/` and
> thus remove "session" information. But then it seems like that quirky stuff is
> actually how it's meant to be? The Hungarian ABP pages even link directly to
the
> `/en/` ones where a translation is missing.. https://adblockplus.org/hu/chrome

Yes, this is quirky - if someone has manually chosen Mexican (es-MX), and they
surf to a page that does not have a Mexican, but a Spanish (es) version
available, they will be flipped to Spanish and not return when later surfing to
a page available in Mexican. 

The existing behavior was to just show 404s in these cases, which would be
terrible experience with all those half-complete translation sets we have
currently.

We could check for an available translation on every request, and flip them back
to Mexican when possible, but that will be counter-intuitive to those who have
manually chosen a different language in the top-right selector on the page. 

The optimal solution would be to keep their last chosen language stored in
session and say "You're currently surfing the Mexican site but this page only
has Spanish translation available", but that would mean revisiting privacy
policy and server mechanics, which is out of scope for this commit.

Regarding the related CMS-generated links, I have no idea, but I *hope* they
have the same logic - I only found some hints of documented behavior in
https://github.com/adblockplus/cms/blob/master/README.md#configuration-settin...

https://codereview.adblockplus.org/29537689/diff/29539748/modules/web/templat...
modules/web/templates/site.conf.erb:68: 
On 2017/09/08 19:20:20, mathias wrote:
> There's a newline missing in front of the bracket.

Acknowledged.

[mathias, 2017-09-11 12:52:52]

LGTM. Please improve on the commit message so it becomes a sentence, i.e. "#0000
- Add redirect ..." or similar.