On 01.09.2016, at 05:37, Chris Muller <asqueaker at gmail.com> wrote:

> Hi Tobias,
> 
>> Where's the problem with calling the image with 'patch.st' in the first place?
> 
> Per our other conversation in box-admins, I want to control the RFB
> configuration from the Linux side.  Once the server image builds
> itself and saved once, it is never saved again.

I still do not get this. 

> 
> If we can't save the image, how to do a quick-patch?  In today's
> SqueakSource, we RFB into the image, make the method changes, save the
> image.  (Shudder).  After a week nobody remembers what the heck we're
> running in production, and its not easy to see even when one gets
> RFB'd into the image.
> 
> In this new SqueakSource server setup, the built image is never
> resaved, so its validity could actually be easily confirmed with a
> hashsum.
> 
> The image is never resaved, but could still suffer a daemontools
> restart, in which case the patch must be reapplied automatically.
> This is done by the administrator exporting the method changes as
> patch.st, instead of saving the image.

Well, then we should write the startup script in a manner that it always is started with
a script, which _then_ gets amended outside[1].
IMHO, The hook is providing a script in the first place, not relying a magic constant file name.

> 
>> It looks a little bit out of place for me here.
> 
> It should be private.  I'll do that.
> 
>> Also, Why Kill MNU and Halt from the list?
> 
> MNU is covered by Error.  Halt because its sometimes necessary to use
> it to diagnose and fix a production problem.

Yeah, but to my knowledge, run: has up to now be intended for non-interacive use. 
_Not_ catching Halt defeats this purpose.

Best regards
	-Tobias

> 
> Best,
>  Chris
> 

[1]: I am still not comfortable with dropping on the image concept just like that :(