Backport fix for #496 and #361 - event clients lost after dev restart

[mliszcz]

Summary of changes comparing to original commits:

• Correction for Client subscribing to events can be blind for up to 200 seconds if the target device is restarted (device restart) #496 (backport of Fix #496 blind event clients after dev restart (attempt 2) #573):
  • 6c22521 - test only
  • 8712a7a - Event names are passed instead of event type enums
    to avoid changing set_client_lib signature
  • 7115111 - I needed to provide a new struct next to EventPar
    and implement a set of overloads that use the new struct
    and attribute_name field instead of attr_id
• Correction for missing attr_conf event after "Reset to default value" #361 (dependent on previous one, backport of Send attribute configuration event to all clients after device restart #694):
  • 3950063 - test only
  • b306962 - no changes

[t-b]

Overall looks good.

The only thing is get_event_param/set_event_param. According to 1 "You can not ... add an overload (BC, but not SC: makes &func ambiguous), adding overloads to already overloaded functions is ok (any use of &func already needed a cast).". But the compliancy tool says good its good
compat-report.zip.

I would tend to believe the tool's output.

[mliszcz]

@t-b thanks for pointing that out! I focused on binary compatibility but have not thought about source compatibility.

The tool is correct, the change is ABI compatible, because function signature is mangled into symbol's name in C++ ABI. So these two overloads are separate symbols pointing to different places in the code block of the library.

But, as documented in the link you referenced, any device server that does something like this, will stop compiling:

auto setter = &Device::set_event_param;

While these methods are intended to be internal for the library (not documented in doxygen), nothing prevents users from calling them from their device server code.

Do you think we need to address this somehow?

[bourtemb]

But, as documented in the link you referenced, any device server that does something like this, will stop compiling:

auto setter = &Device::set_event_param;

While these methods are intended to be internal for the library (not documented in doxygen), nothing prevents users from calling them from their device server code.

Do you think we need to address this somehow?

get_event_param & set_event_param methods are not documented in Doxygen (because under /// @privatesection line).

so users are not supposed to use it.
So I think it is a source code incompatibility we can tolerate.

[t-b]

If users are not supposed to use it we can break it ;)