Ivan Molodetskikh

I believe I'm hitting the same issue. I have the following code: ```rs let profiling_layer = if env::var_os("BXT_RS_PROFILE").is_some() { let (chrome_layer, guard) = tracing_chrome::ChromeLayerBuilder::new() .file("trace.json") .include_args(true) .include_locations(false) .build(); Box::leak(Box::new(guard)); Some(chrome_layer)...

I'm using: ```toml tracing = "0.1.36" tracing-chrome = "0.6.0" tracing-subscriber = "0.3.15" ``` `None` gives me `LevelFilter::current()` = off, `Some(_)` gives me `LevelFilter::current()` = trace.

I consent to re-license my contributions to build.rs.

It's become worse, actually (unless I rebased something wrong): https://beta.arewecompressedyet.com/?job=speed2-2019-10-04_114925-45585ee&job=all-intra-modes-importance-speed2-2019-10-04_115057-fe648e4

Probably enough to have them in the PR

On the one hand yes, but on the other hand I just had bisection pinpoint me to the exact small change leading to #1513 and that's super helpful to have.

Even something like this is UB I guess: ```rust impl ObjectSubclass for FooPrivate { const NAME: &'static str = "Foo"; type Type = Foo; type ParentType = glib::Object; type Instance...

That looks like an interesting idea. > `#[properties]` > `#[methods]` Are these needed? Maybe we could just say "don't put any extra stuff into `mod obj`"?

> It is quite typical to have member variables in GObjects that are not exposed as properties. The annotations I'm referring to are on the whole struct at once, not...

> Breaking change from old behavior that is not easy to detect FWIW I am currently in process of using this default exactly for log-and-return, so yeah, that'll be a...