On 11 Sep 2016, at 19:33, Caleb James DeLisle <[email protected]> wrote:
On 11/09/16 19:21, Vincent Massol wrote:
Hi Thomas,
On 11 Sep 2016, at 17:16, Thomas Mortagne <[email protected]> wrote:
Are those jetty/hsqldb or WAR sizes ?
I’ve used http://snippets.xwiki.org/xwiki/bin/view/Extension/Size+of+XE+Distributions which measures the Jetty/HSQLDB packaging.
Apart from the size another important issue is the number of external dependencies we have which among other things makes dependency handling a pain for extensions.
Here are some numbers from the 8.3M1 WAR:
The WEB-INF/lib takes 172M from which only 12M is pure XWiki jars.
In the WAR some of the biggest features are (size are the feature and the dependencies used only by that feature): * Full text search (close to 70M) ** Solr (~30M), including probably totally useless Jetty (yes in the WAR) stuff (~2M) ** Tika (more than 33M probably not far from 40M, lots of small stuff I was too lazy to count since Tika has an insane dependency tree full of stuff used only by it)
70M for text search, wow, that’s huge indeed. Are there are ways to take only some parts of SOLR (We use solr-core (36MB) and solr-solrj (654K))? Maybe we could ask on their list.
For Tika I see only 2 jars: * tika-core: 602K * tika-parsers: 782K
So I see only a total of ~1.5MB. Have I missed some to get 33MB+ you counted?
* Jython and all its standard libraries (37M): used by code and python macros
Could be good to not use the code macro in the core (I’ve actually always made sure to not use it in the extensions I wrote) and bring the code macro as flavor dependency only. I’ve just checked quickly and there are very few places where we use the code macro in platform. Should be easy to remove those. Then we could move the macro to contrib and bundle it in the XE flavor for now.
+1
37MB for what is usually just {{code}} macro is a bit annoying.
* Old GWT Editor (~26M)
+1
* Groovy+Ivy (7M + 1M): not really used in the XE flavor but required to write a scheduler and many contrib extensions use it and don't declare it as dependency
-0 this is kind of nice to have {{groovy}} always available for debugging and 7M is a lot by some standards but it's pretty small as a percentage.
You’d still have it in the “XE” flavor. The idea would be to remove it from the base flavor only, not “XE”, i.e not consider it core (i.e. moved to contrib). Note that a nice side-effect is that it makes it much simpler to ensure that no core module uses it by mistake. Thanks -Vincent
Also +1 the others below...
Thanks, Caleb
Yep we need to get rid of Groovy in the core. One area is indeed the scheduler extension and we know we need to replace the xproperty using groovy by an xproperty using wiki syntax to allow all script languages to be used. I don’t know of other places requiring absolutely Groovy in the platform. Do you?
Things that would not be too hard to move to Contrib that would also take several external jars in their luggage (which is also nice for extensions, the less core extensions the less dependencies issues): * Office server related stuff (would probably be a good idea anyway to synchronize it more with LibreOffice than XWiki)
Yep +1. Might need some additional extension points defined to integrate cleanly with platform (in create dialog + in WYSIWYG editor).
* Chart macro and renderer
Yep, I don’t know any platform code using that. I’m going to send a proposal email on this and I could move it.
* Rendering syntaxes not enabled by default (we already started this with markdown and mediawiki which are the most used, all the others are probably never used or very rarely)
Agreed, we have a few more to move out.
* Captcha macro/renderer
Yep
* Formula macro/renderer
Yep
8.3M1 still includes CKEditor and Tour extensions jars which should not be there anymore in 8.3 final.
You mean that we should start distributing the platform distribution instead of XE and make XE a flavor (probably under another name, I’ll send a proposal mail about that) that users would pick at runtime, right?
[snip]
Thanks -Vincent