Thanks Guillermo.

I do not really know because what is cool is to be able to load the compiler without having the compiler. 
I know you want that :). 

Now with tanker we have the question of how to handle the binary (probably inside an mcz package)
and managing the fact that the code can be desynchronized from the binary.

Stef

Hi guys!

I was taking a look at tanker to use it to load packages on a mini image. And I found some difficulties :).

- First, the code does not load well in Pharo3. There are some classes/traits that changed and the class extensions do not want to load. I fixed that by adding some stub classes with the same name. And I was able to load the code

- Then, almost all tests fail. The same problem is in the CI job (https://ci.inria.fr/pharo-contribution/job/Tanker/PHARO=30,VERSION=bleedingEdge,VM=vm/248/#showFailuresLink). So, I was having a look at them. I fixed many, but many other things scape from my not so deep understanding of tanker:

  * it uses the old class builder. I changed that reference for the OldClassBuilderAdapter
  * it tries to decompile code when sources are not available... So far I replaced that by the #codeForNoSource which actually does something different...
  * there is a problem with extension methods and local selectors. I don't yet understand where they are being built. Probably it has to do with RPackage-Tanker integration.
  * then some tanker tests make assumptions about the order of literals, which has changed (probably because of opal)


Is it worth to invest some hours/days in tanker? :) Can I help? I have my changes but I cannot commit them, should we move it to Pharo or PharoExtras?

Guille