On Tue, Sep 22, 2015 at 04:14:39PM +0100, Simon Hobson wrote:
[...]

> But one thing I did pick up on, one of the reasons given
> for not having any subdivision was a desire to not have to
> have documented and stable APIs. I find that "a tad
> off-putting" because in the projects I used to work in
> (many many years ago, working as a very junior engineer in
> a shipyard supplying the navy with bespoke vessels) that
> would have been one of the earliest parts to be nailed
> down - split the "blob" into small parts, each doing
> something understandable and testable, and have them all
> communicating via fixed* and documented interfaces.

> * Fixed, as in "can be changed if it has to, but it'll need all the change control that goes with it".

> Reading that the ability to change internal APIs on a whim is seen as a positive attribute suggests to me that this isn't something that's been designed before it's been built.
> I know our methods weren't what you might call "agile", but they were intended to give some expectation of reliability.

I understand that systemd developers guarantee some APIs,
such as DBus, and others are less constrained. I think the
idea is to keep stable the interfaces most important to
external developers.


--
Joel Roth