On 04 Nov 2013, at 00:01, Stéphane Ducasse <stephane.ducasse@inria.fr> wrote:
But marcus we are doing update all the time because this is the way we push new updates daily. So I do not understand why this would be different?
It happens due to strange things: updates that access the web and URLs change, for example. Or in the past one problem was that it could overload squeak-source to load for hours and hours packages⦠And it takes quite some care to keep it going: when *really* bad things happen, we now (once or twice in a year) upload a fixed image that is than taken as the base. If you want to support âtake pharo2 and update it to pharo3â, (or even, take Pharo3 half a year agoâ¦), then we need to put *a huge* (very huge) effort into making sure that we never ever upload a hand-fixed image. This is a lot of work. And in addition, even that does not guarantee anything: you would need to actually test itâ¦
no you dont do anything wrong, unfortunately from what I have been told the update process is broken. Right now the best choices is to download directly from pharo website
The problem is that how images are used is shifting: People used to use an image for a long time, updating the base from time to time while their code was in the image.
These days, what people do is to have an automatic (and well defined) process that build a fresh image on -> base system is updated -> Own code commit
So e.g. I never retain images after I finish something. I commit, wait for the build system to tell me everything is green, and I throw the image away. Images are transient things.
This in turn means nobody uses updating, and this means that it is not tested. and everything not tested brakes after a whileâ¦. (In turn, everything we want to be sure works needs to be tested after every commit, but testing âupdating every old version to the newestâ is not really testable, anaywayâ¦)
When we move to an image-bootstrap for the development of Pharo itself (I guess in Pharo4), we should really check what and how (and if) we support updating existing images, or if we declare the image to be something that *always* be the result of a deterministic build processâ¦
Marcus