`sentry::capture_error` does not capture the backtrace even when the feature is enabled

[Tuetuopay]

Environment

How do you use Sentry?
Self-hosted 21.5.0 and 9.1.2

Which SDK and version?
Rust 0.18 and 0.22 (because of Tokio 0.2 and 1.0)

Steps to Reproduce

1. Install sentry integration with backtraces enabled
2. Call sentry::capture_error
3. Go to sentry, see the absence of stacktrace

• Cargo.toml

sentry = { version = "0.22", features = ["backtrace"] }

• main.rs

fn main() {
    sentry::init(("https://my-dsn", sentry::ClientOptions {
        attach_stacktrace: true,
        ..Default::default()
    });
    println!("Sentry installed");

    let error = "thisisnotanumber".parse::<u32>().unwrap_err();
    sentry::capture_error(&error);
}

Expected Result

The stacktrace would be shown in the web ui.

Actual Result

The web UI shows no stacktrace for the issue.

Fix suggestion

The above screenshot with backtrace was obtained by slightly tweaking the condition in the AttachStacktraceIntegration, which checks whether the event contains exceptions before attaching the stacktrace. However, in the case of events made from capture_error, they do not contain a backtrace, resulting in no backtrace being attached to the event. The tweak is as follows:

diff --git a/sentry-backtrace/src/integration.rs b/sentry-backtrace/src/integration.rs
index 49f6826..e6c3a5a 100644
--- a/sentry-backtrace/src/integration.rs
+++ b/sentry-backtrace/src/integration.rs
@@ -73,7 +73,7 @@ impl Integration for AttachStacktraceIntegration {
         mut event: Event<'static>,
         options: &ClientOptions,
     ) -> Option<Event<'static>> {
-        if options.attach_stacktrace && event.exception.is_empty() {
+        if options.attach_stacktrace && event.threads.is_empty() {
             let thread = current_thread(true);
             if thread.stacktrace.is_some() {
                 event.threads.values.push(thread);

While this works, it may not be ideal as I feel like this is not really what the threads field is for. A fix that may be cleaner would be to record the backtrace in sentry_core::error::event_from_error and/or exception_from_error.

[jfirebaugh]

I would also expect events captured via SentryLogger where the log level is Exception (and potentially Event as well) to include a backtrace. Currently they do not. I'm not sure if this would be addressed by the same change as for capture_error. Let me know if you would prefer me to open a separate issue.