On 3 August 2013 18:44, David T. Lewis <lewis at mail.msen.com> wrote:
> Run the tests on an interpreter VM and all should be well.
>
> Many of the tests in the OSProcess and CommandShell test suites rely on
> cooperating Squeak images initiated with the #forkSqueak mechanism, which
> is not fully operational on Cog. For this reason, the test failures
> overstate the "problem". For most purposes, OSProcess and CommandShell
> are completely functional on Cog, but if you want to run the full test
> suites, use an interpreter VM on a Unix/Linux platform.

OK, that will explain the difference between everyone being very happy
with how OSProcess works, and the test failures. The jobs run a
package's tests on all of Interpreter, Cog and Cog MT VMs.

I'll have to think of a way to say "only run this test on this subset of VMs".

Thanks!

frank

> Dave
>
> On Thu, Jul 25, 2013 at 11:47:01AM +0100, Frank Shearar wrote:
>> Hi Dave,
>>
>> A number of OSProcess tests fail on CI
>> (http://build.squeak.org/job/ExternalPackages/287/#showFailuresLink).
>> I'm installing them like this:
>>
>>     Installer squeakmap
>>         update;
>>         addPackage: 'OSProcess (head)';
>>         install.
>>
>>     [(Smalltalk at: #AioEventHandler) aioPluginPresent]
>>         on: Warning
>>         do: [ "ignore the one-time warning dialog" ].
>>
>> and I run the tests like this:
>>
>>     HDTestReport runPackage: 'OSProcess-Tests'.
>>
>> which just runs the TestCases in that package using a custom runner.
>> For instance, #testSpawnTenHeadlessChildren fails with "Block raised
>> Warning: process accessor module not available". Could these be
>> failing for the same reason the FFI tests fail? For instance
>> FI.Tests.FFIPluginTests.testConstructedCharCall fails with "Unable to
>> find function address".
>>
>> Any thoughts?
>>
>> frank
>