After lots of experimenting with Wireshark and
current Chrome + Firefox on Ubuntu 13.10, this is my
current understanding of the caching when preloading images
with AJAX requests:
* on Chrome, the image request always comes from browser cache
* Firefox makes two separate requests by default
* Firefox with img.crossOrigin = 'anonymous' makes two separate
requests, but the second one is a 304 (does not load the
image twice)
* when the image has already been cached by the browser (but not in
this session), Chrome skips both requests; Firefox skips the AJAX
request, but sends the normal one, and it returns with 304.
"wish I knew this when I started" things:
* the Chrome DevTools has an option to disable cache. When this is
enabled, requests in the same document context still come from
cache (so if I load the page, fire an AJAX request, then without
reloading the page, fire an AJAX request to the same URL, then the
second request will be cached), but an AJAX request - image request
pair is an exception from this.
* when using Ctrl-F5 in Firefox, requests on that page will never hit
the cache (even AJAX request fired after user activity; even if
two identical requests follow each other). When using clear cache
+ normal reload, this is not the case.
* if the image does not have an Allow-Origin header and is loaded
with crossOrigin=true, Firefox will refuse to load it. Chrome will
log an error in the console saying it refused to load it, but will
actually load it.
* Wireshark rocks.
Pushed some tech debt (browser + domain whitelist) into other tickets:
https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/232https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/233
Reverted commits:
8a8d74f01d.
63021d0b0e.
Change-Id: I84ab2f3ac0a9706926adf7fe8726ecd9e9f843e0
Bug: 61542
Mingle: https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/207
Since there was no merge conflict and the changeset worked pre-merge, when jenkins merged the fullscreen changeset, it actually turned out to be incompatible with the new code and spews out a nasty JS error. I think from now on I'll always avoid +2ing a diff that hasn't been rebased to the latest. When looking at the diff not rebased to the latest, you just can't see the issues that might happen with the latest code merged and you certainly won't spot the potential breakage that comes with it.
In fact I think a lot of breakages we've encountered as of late were caused by this very issue.
Change-Id: I46b7dd93c55635f34c01bd8d3eee9785140b5f35
fileUsage tests were applying styles to the fixture element, which
apparently does not get cleaned between tests. This was because
fileUsage.$container had different semantics from element.$container
(widget's own container div vs. parent container div). FileUsage
now inherits from Element to make sure behavior is consistent.
Change-Id: I8fab8bcf084d8b7e480655114506d9848e9d9a49
We have had this method in all Repo subclasses for a while, but sadly
it never got used and we've unintentionally crippled foreign DB repos
for some time now.
{{fixed}}
Change-Id: I972eb739cdd56c666981d5fbc371fa53024ff359
Mingle: https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/217
* more robust method of obtaining URL
* decouple performance logging from providers (mostly)
* ignore fake XHR object which jQuery returns for JSONP requests
* guard for CORS requests - apparently Chrome refuses to return
certain information even with an Allow-Origin: * response header.
* Resource Timing is limited to 150 results, which causes fake
misses in debug mode. There is an API to increase the limit
but it is not implemented in Chrome. I am calling it nevertheless,
maybe IE understands it (it is present in the MSDN docs at least).
This seems to work for AJAX, CORS, JSONP, image AJAX; CORS requests
return 0 for a lot of values, per spec a Timing-Allow-Origin: *
header might help that.
Change-Id: I8353858022f51a7e70774e65513d0fa2554a5064
When the lightbox is opened, or prev/next pressed, preloads the
previous/next N images.
Technical debt introduced:
* initialization is a mess, with the viewer and the interface
randomly setting properties on each other in different phases of
execution. That got in the way and I shuffled things around
until they worked, which is obviously not the way to have a
robust system, but hopefully it will get scrapped soon anyway
in favor of a clean top-down dependency injection.
Change-Id: Idcb5c40de1ac0b3e482decd66e56c4de8ec71b6b
Mingle: https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/155
A simple task queue which can be processed or cancelled.
Will be used to handle preloading in a more robust way.
Change-Id: Ib33f9b2d814a35538f9d4f3691fce5ba5cdc82c1
* moves generic logic into ThumbnailSizeCalculator class
* moves UI-specific logic into interface class
* fixes bug where non-bucketed sizes were served on devices with
non-standard pixel density
* fixes bug where bucketed size was compared to css size instead of screen size
for resizing
Change-Id: I8ba3380b74fcc8fb0a6ecc3f3140627411851ad0
Mingle: https://wikimedia.mingle.thoughtworks.com/projects/multimedia/cards/196
* update paths
* make generate script return failure state on failure
* fix some issues so that it does not actually fail
Change-Id: Idd42e0d8e333c461091079aa1150b1b435e6360c
* userinfo api requests are cached now
* we use jsonp only if we have to (makes gender api calls measurable on WMF wikis)
* all API calls use providers now, provider.Api constructor can be used to
wrap mw.Api with metrics
Does not return a proper model, and gender API calls are not preloaded together
with the rest of the calls. Maybe next time.
Change-Id: I9b3ea73c65eef57e160ac8636d9e45d349150884
