Changing mindset of core LO developers to the status of master -- was test infrastructure ideas appreciated ...

[Wols Lists]

On 11/06/15 15:23, Bjoern Michaelsen wrote:
> On Thu, Jun 11, 2015 at 05:09:08PM +0300, Tor Lillqvist wrote:
>>> So what happens if I write a patch that works fine on linux, so I apply
>>> it to master, and the Windows build promptly blows up ...
>>
>> Instead of "master", read "bleeding". Does that make it OK?
> 
> Not if we care about testing and CI on that branch.
> 
This is my point exactly - master should always pass testing.

Because not all developers can test all configs, we then have the branch
"bleeding" which is explicitly "works for me, does it break someone else?".

That way, master always works, which is what novice (and I guess most
experienced, too) devs want AND EXPECT, but we have a branch dedicated
to ensuring (as far as possible) that we don't get nasty surprises.

The Mac devs don't want the linux guys breaking their build, the linux
devs can't develop in splendid isolation, and the Windows guys can't
blow everything up with some MS peculiarity.

Why should *I* have to spend precious dev time tracking down a bug in
somebody else's code for a system I don't even possess, when all I want
is for "make" to run to completion ...

Cheers,
Wol