<div dir="ltr">I'm a pretty big fan of the approach here. I hear Chris's concern. The only solution I see is to break up the tests by what they're testing. Tests associated with stuff we bless into the core should be run against the core, and everything should be run against the fully integrated "release" image (including core tests.)<div>
<br></div><div style>I know that some tests will fall into a grey area around what exactly is under test. Does anyone see any other gotchas to the approach I'm suggesting?</div><div><br></div><div style>I think Frank's doing stellar work here. Folks have been talking about exploding the image for a long time, and he's approaching a really solid solution.</div>
<div style><br></div><div style><br></div></div><div class="gmail_extra"><br><br><div class="gmail_quote">On Wed, Jun 12, 2013 at 3:45 AM, Frank Shearar <span dir="ltr"><<a href="mailto:frank.shearar@gmail.com" target="_blank">frank.shearar@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="HOEnZb"><div class="h5">On 12 June 2013 11:37, Edgar J. De Cleene <<a href="mailto:edgardec2005@gmail.com">edgardec2005@gmail.com</a>> wrote:<br>
><br>
><br>
><br>
> On 6/11/13 5:17 PM, "Jecel Assumpcao Jr." <<a href="mailto:jecel@merlintec.com">jecel@merlintec.com</a>> wrote:<br>
><br>
>> Yesterday we had another Squeak Community Hangout (some of the previous<br>
>> ones were on April 22, April 8, March 25, March 11, November 2012 and<br>
>> August 30). Though there were fewer people than normal, I thought it<br>
>> would be interesting to mention a bit of what was talking about for<br>
>> those who weren't there.<br>
>><br>
>> Bob Arning shared a scan of a very nice article from a late 1976 issue<br>
>> of Popular Science which had a picture of him running a program he had<br>
>> written for the Dazzler board on the Altair 8800 computer. This was at a<br>
>> store he owned at the time. He also shared a reimplementation of that<br>
>> same program in Squeak.<br>
>><br>
>> Edgar De Cleene mentioned his frustration with the lack of something<br>
>> between Pharo's "throw out the old stuff no matter what breaks" scheme<br>
>> and Squeak's extremely conservative position. I suggested that allowing<br>
>> a system to have several separate images working together could be a<br>
>> solution to that, as well as making good use of today's multicore<br>
>> machines. It would also be a way to add security to SqueakNOS. We also<br>
>> talked a bit about Spoon.<br>
>><br>
>> Hans-Martin Mosner asked if anybody had Squeak code for Elliptic Curve<br>
>> Cryptography. He mentioned that he is part of the Squeak special<br>
>> interest group on cryptography (<a href="http://wiki.squeak.org/squeak/5776" target="_blank">http://wiki.squeak.org/squeak/5776</a>), but<br>
>> nobody there has this stuff and he doesn't know any other Smalltalk that<br>
>> has it.<br>
>><br>
>> That was about it, though since these meetings are supposed to last 24<br>
>> hours it is possible that more people showed up when I wasn't there.<br>
>> Most of the conversations are about Squeak, of course, but in the past<br>
>> we have also talked about subjects such as DRAM bit geometry and the<br>
>> effect of alpha particles. And at one point yesterday we had a "show off<br>
>> your pet" moment. :-)<br>
>><br>
>> -- Jecel<br>
>><br>
>><br>
> Thanks for the excellent report.<br>
> Next time wish to talk about why we do not have a SqueakCore and<br>
> SqueakKernel as separate projects.<br>
> Also show more pets to far away friends :=)<br>
<br>
</div></div>Well, we kind've do. I just don't advertise it. The SqueakTrunk build<br>
on CI takes a hand-crafted "core" image, updates it and runs all its<br>
tests. Noone notices that Nebraska, Universes and XML-Parser are no<br>
longer in the Core because a second CI job, ReleaseSqueakTrunk, loads<br>
these in.<br>
<br>
My aim is to progressively rip more and more out of the hand-crafted<br>
base image, and keep loading those unloaded packages in the<br>
ReleaseSqueakTrunk. I also add the unloaded packages as test suites in<br>
the ExternalPackages job.<br>
<br>
Chris Muller has rightly raised concern that this core+extra image<br>
isn't itself tested. I think the answer to this is that (a) we often<br>
have no idea whether some package works currently because there are<br>
far too few tests, but (b) we can always run the full set of core<br>
tests + package tests on (a copy of) the resulting core+extra image.<br>
<br>
But right now it's extremely hard to unload packages because so many<br>
packages have circular dependencies and the like. Most of my work<br>
recently has been around cutting the easier parts of the tangle, and<br>
raising the profile of some of the less easy parts.<br>
<br>
frank<br>
<br>
> Edgar<br>
><br>
><br>
><br>
><br>
<br>
</blockquote></div><br><br clear="all"><div><br></div>-- <br>Casey Ransberger
</div>