[Vm-dev] VM Maker: VMMaker.oscog-nice.1733.mcz

[Holger Freyther]

> On 19 Mar 2016, at 01:23, Nicolas Cellier <nicolas.cellier.aka.nice at gmail.com> wrote:



Dear Nicolas,


> Stephane, that's my fault and I apologize.This time, i've not put enough effort before the commit.
> Instead i've put excessive confidence in the simulator and my previous work in own branch.
> I can only commit nightly, so effectively, it can cost a day before I correct my mistakes.
> 
> HOWEVER, may I remind you the combinatorial:
>  (spur,v3) x (interpreter,cogit) x (32,64bits) x (mac,win,linux,...) x (squeak,pharo,newspeak)
> not speaking of ARM, android free bsd etc...
> How do you conceivably expect that a single contributor like me would perform tests for each and every configuration?

the burden should never be on a single contributor but luckily computing time has become so cheap that many projects let computers take care of it (nobody needs to wait for a batch job, nobody needs hours to wait for a compiler, etc). Specially with github and travis-ci (or semaphoreci, wercker, ... or tons of other services) it becomes very easy to execute a matrix of configurations before merging something to master. This includes running tests on Linux and OSX and maybe on the Android emaulator if you want to.

Before working on Telecommunication I was a WebKit reviewer and we had a strict policy on regressions. No user should experience an issue that has been already resolved once. This leads to a policy where bug fixes come with tests[1]. I would be very happy if the VM development community could pick-up similar policies. I am happy to help with travis-ci integration.

have a nice weekend
	holger


[1] https://webkit.org/contributing-code/#regression-tests