Fix: Save flash by not inlining SubscriptionInterval::copy()

[MaEtUgR]

Solved Problem

When porting the fix from #23384 to 1.15 with #23392 I saw that it uses 2.5 kilobytes more flash which is totally unexpected.

Big thanks to @niklaut who decompiled the binary and found that the unexpected impact comes from inlining the functions everywhere.

Solution

Move updated(), update(), copy() from SubscriptionInterval into a separate cpp file such that they are not inlined everywhere.
Saves 17.3 kilobytes of flash 😮

Changelog Entry

Fix: Save flash memory by not inlining SubscriptionInterval::copy()

Alternatives

Disadvantage is probably slightly more run time but I'd assume it's not worth that much flash 👀

Test coverage

I'm testing this now.

Context

[dagar]

Nice one!

[dagar]

As a quick check take a look at top before and after on the same board/config.

[MaEtUgR]

As a quick check take a look at top before and after on the same board/config.

ok, doing that.

[MaEtUgR]

@dagar ok, I tested two boards with logging on SD card started.
It's not super scientific but I don't see much of a difference.
v6x with random configuration:
before:

after:

v4 with standard VTOL config:
before:

after:

[dakejahl]

epic! I wonder how much more flash is out there waiting to be saved... 🤑

[MaEtUgR]

For completeness, I made another presumably close to worst case CPU load test: Original Pixracer with standard VTOL 1002 loaded and this time enabled all the logging options from SDLOG_PROFILE:

Ran the logger (while not armed like before) and the CPU load increased by ~1.56%

Before:

After:

[dagar]

epic! I wonder how much more flash is out there waiting to be saved... 🤑

Quite a lot, but at some point there will be performance tradeoffs. Take a look at bloaty, you can use it to see things like code size per file, symbol, or even how much template instantiations cost.
https://github.com/google/bloaty/blob/main/doc/using.md