[LEDE-DEV] [RFC 1/1] base-files/etc/rc.common: allow services to overwrite PROCD functions

[Zefir Kurtisi]

On 11/20/2017 08:46 PM, Felix Fietkau wrote:
> On 2017-11-20 14:25, Zefir Kurtisi wrote:
>> On 11/20/2017 01:27 PM, Felix Fietkau wrote:
>>> On 2017-11-20 11:20, Zefir Kurtisi wrote:
>>>> PROCD services can't overwrite those functions defined in the
>>>> USE_PROCD body (like stop()), since the initfile is sourced
>>>> before.
>>>>
>>>> This change moves the sourcing below and with that allows
>>>> PROCD init scripts to overwrite them.
>>>>
>>>> Signed-off-by: Zefir Kurtisi <zefir.kurtisi at neratec.com>
>>> This seems broken to me. USE_PROCD is defined by the init script, so you
>>> can't test for it before the script has been sourced.
>>>
>>> - Felix
>>>
>> This was my initial assumption, but in fact it works as expected (for PROCD and
>> legacy services). Running 'INIT_TRACE=y /etc/init.d/log shows that USE_PROCD is
>> set correctly during execution.
>>
>> Why it does I don't understand, since when
>> * /etc/init.d/log starts with '#!/bin/sh /etc/rc.common'
>> * in /etc/rc.common, the initscript is sourced again
>>
>> ... I expected this to be a circular dependency, but what it does is:
>> 1. interpreting all the variables set in initscript (including USE_PROCD=1)
>> 2. interpreting all the variables set in rc.common
>>    (starting with [ -n "$USE_PROCD" ])
>> 3. executing the functions
>>
>> Could not find enough information about how shebanging a shell-script is executed.
>>
>> Does somebody know and could please clarify?
> I just did an experiment and your claimed step 1 does not happen in my
> test. rc.common gets interpreted first, and USE_PROCD is unset where it
> would be needed.
> 
> - Felix
> 
Thank you Felix for taking a look.

:-/ need to find out why it does differently on my side, but anyway realized that
the approach tried might not be the best in town.

What I need to achieve is a 'logread --follow --reconnect' functionality, i.e.
currently if someone is running a 'logread -f' in a console and log service is
restarted, the logread instance stops logging. I'll try to add that functionality
to logread instead.


Thanks,
Zefir