feat(core): Emit hooks for transaction start/finish

[AbhiPrasad]

Builds on top of #7370

For now closes #7262 (but we still need to add replay hooks).

[github-actions]

size-limit report 📦

Path	Size
@sentry/browser - ES5 CDN Bundle (gzipped + minified)	20.21 KB (0%)
@sentry/browser - ES5 CDN Bundle (minified)	62.89 KB (0%)
@sentry/browser - ES6 CDN Bundle (gzipped + minified)	18.85 KB (0%)
@sentry/browser - ES6 CDN Bundle (minified)	55.8 KB (0%)
@sentry/browser - Webpack (gzipped + minified)	20.6 KB (0%)
@sentry/browser - Webpack (minified)	67.32 KB (0%)
@sentry/react - Webpack (gzipped + minified)	20.62 KB (0%)
@sentry/nextjs Client - Webpack (gzipped + minified)	48.39 KB (+0.08% 🔺)
@sentry/browser + @sentry/tracing - ES5 CDN Bundle (gzipped + minified)	27.39 KB (+0.16% 🔺)
@sentry/browser + @sentry/tracing - ES6 CDN Bundle (gzipped + minified)	25.65 KB (+0.13% 🔺)
@sentry/replay ES6 CDN Bundle (gzipped + minified)	43.15 KB (0%)
@sentry/replay - Webpack (gzipped + minified)	37.16 KB (0%)
@sentry/browser + @sentry/tracing + @sentry/replay - ES6 CDN Bundle (gzipped + minified)	61.05 KB (+0.07% 🔺)
@sentry/browser + @sentry/replay - ES6 CDN Bundle (gzipped + minified)	54.31 KB (+0.01% 🔺)

[JonasBa]

Thank you @AbhiPrasad!

[Mr0grog]

Apologies if this is the wrong place to ask (LMK where is better!), but I saw this in the release notes and started to use it to clean up some messy code I have around spans. However, the issue this is based on says:

This is special case - but we need this to unblock profiling so let's . We keep these APIs completely internal, we can change it later on.

…so I’m wondering if I should actually avoid using these events. (Side note: it would be really helpful to not include private APIs in release notes, or at least call out what things are not for public use. 🙏 )

[AbhiPrasad]

@Mr0grog we need to include some stuff about internals because we have downstream SDK consumers (https://github.com/getsentry/sentry-electron and https://github.com/getsentry/sentry-react-native for example).

Curious about your use case - maybe we do end up documenting this properly. What are you doing exactly with hooks?

We only recommend only using the API that is directly referenced via our documentation, but anything exposed as public API (class methods, exports) will not break until the next major version.

[Mr0grog]

Oh, we have some terrifyingly horrible hacks to try and make adding spans work well: https://gist.github.com/Mr0grog/bf88c0e003fc07cb2ca6271f426ba65e

The transaction finishing event is useful because we want most spans that never finished to get included in the transaction (probably an exception occurred that prevented us from getting to the span.finish() call), so the helpers patch Transaction to get notified and close the span unless you specifically opt out. BUT while I poking at the events added here I realized they probably aren’t a good solution anyway (it seems like both the hub is not guaranteed to be present on the transaction and a client is not guaranteed to be on the hub, so for something totally focused on the transaction/span world, a hook on the client is probably the wrong place). 🤷

[AbhiPrasad]

it seems like both the hub is not guaranteed to be present on the transaction and a client is not guaranteed to be on the hub

This is typed this way because JavaScript lol, but for for standard use you can pretty much always consider a hub/client to exist on a transaction that has been sampled.

[AbhiPrasad]

Oh, we have some terrifyingly horrible hacks to try and make adding spans work well

We need to make this better, let me take this back to the team.

[Mr0grog]

This is typed this way because JavaScript lol, but for for standard use you can pretty much always consider a hub/client to exist on a transaction that has been sampled.

Ah, good to know! So we could check transaction.sampled, and generally assume those should be present. 👍