Yikes. what a difference an ant build makes! I built a compressed dojo build with a cursory look at the firebug xml http requests and just put that in a custom profile and then overwrote the tapestry heads version of dojo and built that and boom: my site had no xmlhttp requests coming from dojo anymore (well, except, peculiarly, from src/dom/__package__.js and src/html/__package__.js and src/html/style/__package__.js) and only like 9 coming from the default tapestry stuff plus 2 from my own stuff.. essentially the resultant pages are now visible and responsive to a client in under 3s (not 3s to excute, 3s before the page is actively and perceptively loading... by the same metric, were faster than cnn.. heh). anyway removing these low hanging fruits (just the dojo stuff until i can figure out how to bake all the tapestry stuff into a single file build) literally tripled more than doubled our speed. 20 requests down to 9 requests (even if the resultant js is now 250k) makes for a) less to download (after all, the resultant dojo.jhs is compressed src) b) less handshakes and things that might succumb to network latency in roundtriping so header overhead is minimalized c), more pages of code that will be also be later cached in browsers.
It's antithetical to common sense.. you figure a rally compressed minimalist dojo.js initially will improve performance and then the stuff can be downloaded as needed.. well, it turns out most of its needed on page load anyway so youve only seperated the work, not reduced it. Plus, the dojo.require mechanism is synchronous/blocking, so you dont even get the much vaunted, perceptual advantage of asynchrnocity in ajax... The trick wil be to now try plugging in a custom dojo build instead of having to rebuild tapestry itself each time.. the last thing i want to do is branch and maintain (sync) a tapestry release! Using dojoSource, dojoPath doesnt seem to work on the normal places.. (well, i havent tried just sticking the .js in the /src folder of my classes and using classpath:/js/dojo. i will try it if i must, but it sounds evil. Is there anyway to get the context:/ options working or am i missing something hideously obvious? Anyway, forgive the deluge and thanks in advance for any help, Josh ps:I hope your Halloween went well (if it went at all, that is) On 10/31/06, Josh Long <[EMAIL PROTECTED]> wrote:
Sounds good so far! Thatll good to know.. what would building/baking the tapestry files into dojo.js or some file where its compressed look like? That is, i have the tapestry head on disk and im in /js/ and i see tapestry, dojo, and a few other directories. i also see a tapestry profile .js -- how would i use that setup to get a build with tapestry.namespace, and tapestry.* etc... I know what to do to get a dojo build from head to produce a custom profile... what's it look like iwth the tapestry js files baked in? What do i modify? What's the equivalent of ant -Dprofile=foo release compress in the tapestry tree with the tapestry packages baked in? Thanks, Josh On 10/31/06, Jesse Kuhnert <[EMAIL PROTECTED]> wrote: > > Oh..And don't forget the "intern-strings" ant build option. That will > embed > the templates/css into the main dojo.js file as well. > > On 10/31/06, Jesse Kuhnert <[EMAIL PROTECTED] > wrote: > > > > The dojo build system can compress its own and your (or tapestry) js > > packages all into one file. You'll probably have to dig around in > > manual.dojotoolkit.org to find more. > > > > When you are really getting serious about optimizations though I would > > check out the jslinker provided with dojo as well. It will condense > your js > > into ~exactly~ only the functions/code your app uses. I haven't > actually run > > it myself but I know Andy has. > > > > > > > > On 10/31/06, Josh Long <[EMAIL PROTECTED]> wrote: > > > > > > I'm experiencing a ridiculous slow down in speed on a front page > because > > > the > > > initial page uses a few widgets whcih themselves use other dojo > > > packages, > > > few of which are included in the default dojo.js that comes bundled > with > > > > > > tapestry. So, i built a dojo.js equivalent to te kitchen sink build > with > > > a > > > few extra packages (dojo.dom, dojo.regexp, and more are a few > packages i > > > added for good measure.. anyway, this has worked out pretty well > thus > > > far: > > > the front page loads all of the code from dojo/* in one js file > (well, > > > i do > > > see two xmlhttp requests for the templates of a few widgets, of > course, > > > but > > > everything else is readily available.) instead of in 40 seperate, > non > > > compressed, header-incurring requests. > > > > > > I have the tapestry source code checked out and have one more thing > i > > > want > > > to do but im not sure what the right mechanism is: how do i get all > the > > > tapestry packages to be in one fle that i can load manually like i > do > > > dojo.js? that way id have only two files that i could tune > individually > > > that > > > load the code specific to my app and theyd all be cacheable (ie, not > > > paged > > > trough xmlhttp, which is a problem fr example in FF pre 1.5 . .) .js > > > files and > > > theyd all require only one round trip and no headers.. we have > pretty > > > bad > > > latency are really looking to streamline individual requests to the > > > server. > > > > > > Any help would definitely be appreciated... > > > > > > thanks, > > > > > > Josh > > > > > > > > > > > > -- > > Jesse Kuhnert > > Tapestry/Dojo/(and a dash of TestNG), team member/developer > > > > Open source based consulting work centered around > > dojo/tapestry/tacos/hivemind. http://blog.opencomponentry.com > > > > > -- > Jesse Kuhnert > Tapestry/Dojo/(and a dash of TestNG), team member/developer > > Open source based consulting work centered around > dojo/tapestry/tacos/hivemind. http://blog.opencomponentry.com > >
