GWT 2.8 generates code 14% bigger than GWT 2.7

[Óscar Frías Barranco]

Hi.

We are trying to move our website from GWT 2.7 to 2.8 but we have found that GWT 2.8 generates code which is significantly heavier.

For example, these are the sizes for one of the permutations:
GWT 2.7 + Closure Compiler:  590 KB
GWT 2.7 without Closure Compiler:  664 KB
GWT 2.8 (no Closure Compiler available):  674 KB

So the difference between 2.7 and 2.8 is +14% code size.

We would like to keep our site up to date with the last version of GWT but we don't like increasing our website code in this manner because it means higher downloads, higher waiting times, worse user experience, etc.

Do you plan to bring Closure Compiler (-XenableClosureCompiler option) back to GWT? It was very useful for us.

Kind regards,
Oscar

[Jens]

Do you plan to bring Closure Compiler (-XenableClosureCompiler option) back to GWT? It was very useful for us.

No. It was experimental and had some rough edges that haven't been fully implemented.

You would need to execute it manually using the GWT output as input to Closure Compiler (and provide an extern file to protect special GWT variables). Alternatively try build your own GWT SDK and rollback the commit that removed closure compiler from GWT.

-- J.

[Kirill Prazdnikov]

Do you have performance measurements ? 

Did the performance better for 2.7 ?  

Did you measured real transfer traffic (web server supports real-time GZ compression over HTTP) ?

Thanks

[Óscar Frías Barranco]

Do you plan to bring Closure Compiler (-XenableClosureCompiler option) back to GWT? It was very useful for us.

No. It was experimental and had some rough edges that haven't been fully implemented.

And why don't keep it experimental too in 2.8 instead of removing it?

[Kirill Prazdnikov]

Hi, 

I think you can 

  - enable single permutation, single script linker

  - run the tool from maven build: apply gwt maven plugin, and then apply required version of Closure Compiler

[Jens]

And why don't keep it experimental too in 2.8 instead of removing it?

Why keeping it if it's only half baked into GWT and you can run it externally as well? 

But I guess part of the reason to delete it is because GWT is in the process of cleaning itself up to move forward to GWT 3.0. Given that it is very likely that GWT 3.0 will be based on the upcoming J2CL compiler, you will probably need to run Closure Compiler manually as part of your build process (Gradle, Maven, whatever) anyways if you switch to GWT 3.0.

-- J.

[Óscar Frías Barranco]

And why don't keep it experimental too in 2.8 instead of removing it?

Why keeping it if it's only half baked into GWT and you can run it externally as well? 

Do you have a step by step guide to run the closure compiler externally ?

[Jens]

Do you have a step by step guide to run the closure compiler externally ?

Basically you would need to replicate what has been done programmatically in the removed ClosureJsRunner class. See commit

https://github.com/gwtproject/gwt/commit/162ccc9c9112a09bf9ea046da95760f5f1886b72

Looks like you have to

- define your own gwt_externs file to tell Closure Compiler which special GWT variables not to rename

- maybe create a hack.js file that exports gwtOnLoad function so it does not get removed by Closure Compiler. Maybe its not needed if also passing in the <module>.nocache.js file to Closure compiler. no idea.

- Give Closure Compiler the GWT output (your *.cache.js files including split points) in the correct ordering as split point files have dependencies (initial fragment, leftover, split points)

-- J.

[Óscar Frías Barranco]

Thanks for your help, but unfortunately this does not look like a simple process to implement for GWT users like us.  For the moment we will remain in GWT 2.7 which generates a smaller code for our website, and hoping that future GWT versions will be able to reduce or at least maintain the same code size as version 2.7

[Kirill Prazdnikov]

Hi Jens,

What if we have SSO linker ?
Then no hacks is needed, right ?
Then just run the Google Closure Compiler on the output.

[Slava Pankov]

@Jens

I've tried WITHOUT any success to use closure compiler externally with GWT 2.8

See my question here: https://groups.google.com/forum/#!searchin/google-web-toolkit/closure$20compiler%7Csort:date/google-web-toolkit/k_kjIv9Klsg/LZAZiUf9BAAJ

Still want to find out exact steps to get it working :-(

[Kirill Prazdnikov]

Hi

 

I've tried WITHOUT any success to use closure compiler externally with GWT 2.8

Did you try SSO ? Did you run Closure Compiler on SSO output script ?

[Ignacio Baca Moreno-Torres]

But IMO if you really care user experience you have other places to improve your page. I thinks updating to the last version is always a good idea, if any security or compatibility blocker problem is detected in your version you will need to update, and having a old version won't help. Also this helps move forward the community and the lib (GWT), because everyone focused in the last version, also focus the issues and improvements in that version.

The small step back eliminating the closure compiler just increase a bit the code size of your app. But 100k?! did you see the trends and the average code size of webs? http://httparchive.org/trends.php?s=All&minlabel=Nov+15+2014&maxlabel=Nov+1+2016, you should note that even google.com uses 500k in 13request to load, and it's ""just"" a text inbox in the center of your window (http://httparchive.org/viewsite.php?pageid=70036564).

I might be wrong, but probably trying to embed the app.nocache.js in your index.html will gain some milliseconds, probably just that dummy change will get more ms than the cache-ables 100k you have lose without the closure pass. If your app is quite big, various MBs, you should take care than the first split point load a minimum as possible to show something to the user. But, in your case, less than a 1M, I think that using split point is not a good idea. Your code will get cached, and most of static apps already has more than 500k of JS, so your single page app having a 500k-1M of JS is pretty reasonable.

You can go forward and apply the progressive web concepts (https://developers.google.com/web/progressive-web-apps/checklist). Pretty sure you get there without the closure compiler ;). Service workers will allow maximum control to preload other split point or resources, and to absolutely control how your app caches, so you can use previous version, load the next one and in the next page reload the new app version will be used. This is not easy at all, but... some time we focus on the optimization we have no control of (like the closure pass) and we don't apply other we have control of, but that requires our effort.

So, 😀 update to GWT 2.8! it's awesome!, try other solutions to improve page loads and share how you do it!

[Frank]

I am also interested in answers to this question.

When Closure became available in GWT I did many real world tests on real world projects. And in the end the code compiled with closure was slower in every case (although smaller). So I never understood why anyone would want to use that closure compile on top of gwt compile.

Op donderdag 24 november 2016 15:05:50 UTC+1 schreef Kirill Prazdnikov:

[Kirill Prazdnikov]

I have some glue. Google JS Compiler used in chrome is able to detect some patterns and jit. But some patterns are not detected.

Ex: for are deceted, do ... while sometimes not.

It happens so that GWT produce patterns detectable for JS Compiler. And Closure can break that structure.

I am happy that GWT dropped cl suppert and focus on JsInterop (where is JsIndexer?)

28 нояб. 2016 г. 11:04 пользователь "Frank" <frank....@gmail.com> написал:

--
You received this message because you are subscribed to a topic in the Google Groups "GWT Users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/google-web-toolkit/IIQk4xH0-g8/unsubscribe.
To unsubscribe from this group and all its topics, send an email to google-web-toolkit+unsub...@googlegroups.com.
To post to this group, send email to google-web-toolkit@googlegroups.com.
Visit this group at https://groups.google.com/group/google-web-toolkit.
For more options, visit https://groups.google.com/d/optout.

[Slava Pankov]

No, I have not tried SSO. But in 2.7 it was working fine without SSO, I will try SSO and let you know.

[Etienne Basso]

Hi Ignacio,

This is not only about loading time over the network, but also about script parsing by the web browser. 

In 98% of the case this is not an issue, but I have a very specific use case were the code is directly embedded on a special device which doesn't even have a real CPU but an ASIC emulating one. This device is really really not powerful and GWT is the only framework providing decent speed.  I noticed significant differences in the script parsing delay as js sources are getting bigger.

In my case 100KB are a huge deal.

[Kirill Prazdnikov]

Do you use SSO ? SSO script is smaller then no SSO. Did you run CC on it ?

29 нояб. 2016 г. 17:29 пользователь "Etienne Basso" <ebas...@gmail.com> написал:

--

[Slava Pankov]

SSO is not working for my project, I cannot successfully compile with it.

Put the following to my gwt.xml, but no luck:

<inherits name="com.google.gwt.core.Core" />

<collapse-all-properties />

<add-linker name="sso" />

[Kirill Prazdnikov]

Please try single permutation

  <set-property name="user.agent" value="safari"/>

[Slava Pankov]

Still not working for me.

  <inherits name="com.google.gwt.core.Core" />

  <collapse-all-properties />

  <set-property name="user.agent" value="safari"/>

  <add-linker name="sso" />

[ERROR] The module must not have multiple fragments when using the Single Script Linker.

Probably the problem is that I have many languages (i18n) in my app.

Any other ideas?

To post to this group, send email to google-we...@googlegroups.com.

[Etienne Basso]

Yes I use SSO and Cc from gwt 2.7
I also use other optimizations like XdisableClassMetadata and XdisableCastChecking.

I also use smaller css class names:
<set-configuration-property name="CssResource.obfuscationPrefix" value="empty" />

In my very specific case this makes a big difference.

[Etienne Basso]

Yes I confirm this. For using SSO I had to set only one language.

I do multiple build manually for each language until I find a better solution.