Page MenuHomePhabricator

Strip navbox html from action mobileView
Closed, DuplicatePublic

Description

Mobileview transfers all navbox html, which is never shown. This increases page size - sometimes by 10's of k. (See enwiki Obama article)

Max, Yuvi said this commit I7841f74b04277ae250d9d703a80b751d634959d0 removed the generic html stripping code. We can has again?

Event Timeline

Mhurd created this task.Apr 22 2015, 3:12 AM
Mhurd raised the priority of this task from to Needs Triage.
Mhurd updated the task description. (Show Details)
Mhurd added a project: Mobile.
Mhurd added subscribers: Mhurd, yuvipanda, MaxSem.
Restricted Application added a subscriber: Aklapper. · View Herald TranscriptApr 22 2015, 3:12 AM
Mhurd set Security to None.
Mhurd moved this task from Needs Triage to Bug Backlog on the Wikipedia-iOS-App-Backlog board.
Mhurd moved this task from Bug Backlog to Tracking on the Wikipedia-iOS-App-Backlog board.

@MaxSem why was this removed? IIRC you switched to hiding via CSS in stead of DOM transforms, but why? Perf reasons?

Mhurd added a comment.Apr 22 2015, 3:19 AM

I'd argue that even if stripping this server-side is slightly less efficient, as long as we have a bit of horsepower to spare it's a good trade-off as it reduces transfer times - sometimes by a considerable amount.

22 апр. 2015 г. 6:19 пользователь "Mhurd" <
no-reply@phabricator.wikimedia.org> написал:

Mhurd added a comment.
I'd argue that even if stripping this server-side is slightly less

efficient, as long as we have a bit of horsepower to spare it's a good
trade-off as it reduces transfer times - sometimes by a considerable amount.

It's not about server resources that we have these days, but about latency.
I would be much more confident about a Cassandra-backed service that would
pre-render the mobile HTML on save instead of on-demand rendering.

@MaxSem but isn't action=mobileview cached and hence the extra latency amortized over several hits?

It is cached only for large pages and even then cache hit rate is not very good. What's the state of Bernd's API experiment?

Restricted Application added a subscriber: StudiesWorld. · View Herald TranscriptDec 6 2015, 5:56 AM
MaxSem removed a subscriber: MaxSem.Jan 6 2016, 9:53 PM
He7d3r added a subscriber: He7d3r.Jan 7 2016, 5:47 PM

It is cached only for large pages and even then cache hit rate is not very good. What's the state of Bernd's API experiment?

The "experiment" turned into a production services hosted in RESTBase. Currently, it is rolled out to the beta Android app. The production app will follow in a couple of weeks. By then we should already have pre-generation of the results, so getting a response should be very fast. It does strip out nav boxes and many other things the app doesn't display. It uses Parsoid for most pages, except for the main page. We currently don't use it for zhwiki, though, since language variants are not well handled in RESTBase/Parsoid.

MaxSem removed a subscriber: MaxSem.Jan 11 2016, 11:40 PM