Hi,
there is one more issue I wanted to discuss with you guys, but I completely forgot to mention it at the meeting we had on Monday.
It's the thing with parser implementation that came up after I finished writing libteam support. The problem is that the way the parsing is implemented now allows using template funcions (e.g. {ip(1,1)}) only within a command sequence. The functions can be only used after all machine and net configs were parsed.
This makes the libteam implementation practically unusable. It works correctly, but you have to enter the device names of slave interfaces to the teamd config (which is contained within netconfig) by hand.
We discussed it a while ago with Jirka and it seems, that there's no easy way around this :-/. The solution is to make the parser work on a line-by-line basis and process actions associated with certain lines right away. Now things are a little optimized, so the controller sends only one big request to each slave with the whole netconfig, which results in this not-very-intuitive behavior of recipes.
I personally think, that this could really simplify conditions of using the template functions and aliases from a obscure set of rules (that can be hard to get for people not involved int LNST development) to a simple rule "you can use whatever you see written above this line".
Do you think this is worth the work at the moment or I should leave it for now and get back to it later?
I don't know if my explanation is clear enough ... I can explain it to you in more detail in person sometime.
Radek