I totally agree with you on this.

For a while I had a CI job that was trying to execute Sista on the latest
version of Pharo and VM, and sent me a mail when something had gone wrong.
I disabled it when there was problems on the CI, I will make it work again.
This way I am notified and I can update the Sista code as Pharo 7 is
developed. That helps a lot.

On the Pharo CI the time to run tests matters hence I will try to find
another solution to test full block and the other bytecode set. I did not
do that because I hope I can change Pharo to use full block by default
soon, but since they're broken...

Regards



On Fri, Sep 22, 2017 at 10:58 AM, Holger Freyther <[email protected]>
wrote:

>
> > On 22. Sep 2017, at 13:33, Clément Bera <[email protected]> wrote:
>
>
> Hi Clement,
>
>
> > Sista will work for the release of Pharo 7 as it did for Pharo 6 and 6.1
> but it won't work every day during the alpha of Pharo 7 and no sane
> framework developer can guarantee that. Sista does not work currently in
> Pharo 7 due to some changes breaking the full block closures, I am going to
> fix it but even then in the incoming months there is a high chance that
> another change in Pharo 7 breaks it again, so I would never recommend to
> use it with the alpha.
>
>
> thank you for your response. I didn't expect it to work (after all you
> label it "Alpha" and I ran into issues already) but my intention for the
> mail was a bit different. Let me try to briefly explain my motivation. In
> my experience:
>
> (1) Software not executed stops to work (or never "worked")
> (2) The earlier breakage is detected the less expensive it is to fix.
>
> My assumption is that Opal is part of Pharo and that we look at a
> regression of the Opal compiler and/or how it interacts with the rest of
> the image. As it might be easier to understand/fix it now than later I
> brought it up. From my point of view I wonder if we want to add ~50s to the
> CI process to execute:
>
> CompilationContext bytecodeBackend: OpalEncoderForSistaV1.
> CompilationContext usesFullBlockClosure: true.
> OpalCompiler recompileAll.
>
>
> Is there any value in doing so?
>
>
> cheers
>
>         holger
>
>
> PS
>
> (2) The intuitive explanation is that the cost of analysis is a lot lower
> the earlier the problem is detected. There are fewer potential changes that
> lead to the breakage and the people involved will remember what/why they
> did reducing the cost of fixing. Ideally things are found to be broken and
> fixed before integration.
>
> Maybe an analogy is old-school aviation navigation.. the earlier a mistake
> is found in navigation the closer to the original course...
>

Reply via email to