yellow team mailing list archive

[Danilo Šegan <danilo@xxxxxxxxxxxxx>]

Hi all,

This is about the simpler of the two escalated translations bugs.

Bug 734765 is about adding some information about translations sharing
between templates to the +templates page (that one is registered on
different "series" views, such as DistroSeries and ProductSeries, so
that might need to be taken into consideration).

Basically, the goal is to show concise information from pages like
https://translations.launchpad.net/ubuntu/natty/+source/gnome-session/+sharing-details on the full listing https://translations.launchpad.net/ubuntu/natty/+templates

The view base itself (BaseSeriesTemplateView in
lib/lp/translations/browser/potemplate.py) is basically a thin (or not
so thin) wrapper around POTemplateSubset.  That is optimized to execute
one query to fetch all the templates, and the view does no batching
because one big list is much more useful for Ubuntu Translation
coordinators (and other than Ubuntu, nothing has more than say 150
templates, with 1 or 2 being most common).

Now, we get to the more interesting bits, which is sharing.  I'll
explain the entire concept in low level terms, even though most of it
doesn't apply to this particular bug.  Sharing is a way to ensure that
when translations happen in one translation template (POTemplate), they
automagically appear in another.  How those two templates relate is
critical to proper sharing.

Translation sharing happens along two axes: inside a single
product/source package, and between a product (series) and source
package (in distro series).  We can say that two POTemplates take part
in sharing if:

 1. they have the same name (potemplate1.name == potemplate2.name)

 2. one of the following holds:
   a. they are in the same product series 
      (potemplate1.productseries == potemplate2.productseries)
   b. they are in the same source package
      (potemplate1.sourcepackagename == potemplate2.sourcepackagename
       AND
       potemplate1.distroseries.distribution ==
potemplate2.distroseries.distribution)
   c. potemplate1.productseries is linked to potemplate2.sourcepackage
      OR
      potemplate1.sourcepackage is linked to potemplate2.productseries
      through the Packaging table

The bug is about showing links to potemplate2 for 2c.  How "sharing"
templates are calculated can be seen in getSharingSubset in
POTemplateSet.  Do note that since templates from different series in a
single product or sourcepackage already participate in sharing, that
means that they also participate in sharing with those templates linked
through packaging table (eg. product/1.0/template shares with
ubuntu/hardy/product-2.0/template), so 2c is not entirely correct, but I
believe it's sufficient for this bug.  I am asking for clarification.

However, it was a long-standing plan to do away with POTemplateSet which
kind of becomes meaningless/ugly in this particular circumstance (and
many others it is used in).  It would be so much nicer to switch the
view to use TranslationTemplatesCollection instead, and just add a
restriction similar to above using the logic like in joinOuterPOFile.
Of course, it'd be good to keep all the sharing conditions localised, so
sharing code between getSharingSubset and this new code would be very
nice.  Those're just my personal preference, and whoever picks it up can
do whatever they want. :)

Note that one of the important optimizations for +templates page was not
using fmt:url and instead constructing URLs by hand.  We might need to
be smart about that for the linked template as well (traversal
canonical_url is doing was too slow, taking roughly 30ms per link, and
considering there are over 1000 x 5 links to render, we were hitting
timeouts just in Python time).  Note that the links will include parts
of other context as well (eg. for ubuntu:+templates, a relevant linked
productseries which is different for each template), so we might need to
prefetch that too.

Sorry if I went into too much detail, thus ruining the exploratory path
for someone, but I just tried to explain the core concepts and some
gotchas with the page as is (and I did restrain myself from going deeper
into it, because I know I have very strong opinions about all the
translations code :).  If there is still some unclear things, I'd be
happy to chat about it.

Cheers,
Danilo