dhcpcd-discuss

Hi Koichi

On 31/10/2016 05:08, Okamoto, Koichi wrote:
> Hello Roy san,
> 
> I send the following my struggle bad timing.
> http://roy.marples.name/archives/dhcpcd-discuss/2016/1418.html
> 
> Because, you expressed as follows:
>> I don't normally work in branches, but I have just created a dhcpcd-6 branch
>> which will only get fixes I need serious from this point onwards.
> 
> But, if possible, I would like to get your help and to know your comment for my patch
> so that I can prepare more effective patch in the future in time.
> 
> This modification can work "dhcpcd -k wlan0" and "dhcpcd -n wlan0" control command
> without issue, that dhcpcd operates all existing network node as dhcp client
> on Linux-3.10.26 even if the allowance setting were only for wlan0 and eth0.

I'm not sure the patch is fully correct. Consider a second interface,

   Step1. dhcpcd -M -b wlan0
   Step2. dhcpcd --exit wlan0
   Step3. dhcpcd -M -S ip_address=192.168.XX.YY/24
            -S routers=192.168.XX.1
            -S domain_name_servers="192.168.XX.ZZ 192.168.XX.AA"
            wlan0
   Step4. dhcpcd -M -n eth0

In step 4, we don't want the options from step 3 to be applied, but I
think your patch changes it so they will.

> 
> The below was your lecture:
> http://roy.marples.name/archives/dhcpcd-discuss/2016/1335.html
> 
>> The third idea is to re-think how dhcpcd handles options, basically parsing
>> dhcpcd.conf. This is a fairly radical idea as depending on the outcome existing
>> command line options and configs may not work. This idea might not even get
>> off the ground .... it's just an idea.
>> Commentary on this is welcome.
> 
> I wish dhcpcd new version (branch-7 or lator) can keep the behavior of a control command
> via control socket as well as dhcpcd.conf. That makes version up easy and may also enlarge
> user ratio of latest version.
> 
> If you will replace all command control part and all configuration file settings on dhcpcd-7
> from dhcpcd-6 latest version, we might consider ignoring control command issue I reported.
> 
> I would like to know your opinion too.

The control socket will be a compilable option - it's sole reason for
existance was to to allow starting/stopping of an interface for the
master daemon.

Accepting extra options on it has been problematic since it was first
written ...... mainly because the syntax is not powerful enough to fully
turn off what was turned on and also introduces some potential conflicts
with what was set earlier.

For example, if dhcpcd was started with -4, should it allow a subseqent
call with -6 to work? (ok, that's quite easy to answer, but you get the
idea).

Anyway, neither the control socket nor dhcpcd.conf are going away in
dhcpcd-7 so the reported issue still applies.
I'm really busy with other things, so if I don't reply back in a
suitable time-frame feel free to file a ticket here so I don't forget:
http://roy.marples.name/projects/dhcpcd/login?g=/projects/dhcpcd/tktnew&anon

Roy