Resolves #1151 - send errors, error-logs and panics to sentry

[syphar]

This PR starts sending errors to sentry.

What is sent to sentry now (including backtraces):

• log::error!
• anyhow::Error, when ending up in the main handler
• panic! everywhere

I'm not 100% sure if there need to be more places changed?

open topics / question

• should all the error! calls in the codebase stay errors being sent to sentry? I could imagine some of them should be changed to warnings? Or should we exclude error-logs from sentry?
• I would love to switch the handling code in the ctry! to use capture_anyhow when it gets a fitting error type, but I wasn't sure about the best approach. (done, thanks to @jyn514)

[jyn514]

This is really cool :) do you know how I can test it?

[jyn514]

Hmm, why set this? The docs say:

When enabled, stack traces are automatically attached to all messages logged. Stack traces are always attached to exceptions; however, when this option is set, stack traces are also sent with messages. This option, for instance, means that stack traces appear next to all log messages.

and I don't think it's super useful to show a backtrace for the top-level error! macro, since it will always be the same.

[syphar]

I wanted to have automatic stack-traces for log::error! calls.

I might be wrong, but in testing the stacktrace looked correct, from the place where the log::error call was made:

( I had an additional log::error call in main)

[jyn514]

I might be wrong, but in testing the stacktrace looked correct, from the place where the log::error call was made:

Right - but my point is that usually we want the stacktrace of the error itself, not the error! call.

I guess that's an argument for getting rid of error! altogether and changing all those to capture_anyhow; I would be ok leaving this in until we do that.

[syphar]

This is really cool :) do you know how I can test it?

@jyn514 short answer to this question:
I used my personal sentry account to create a SENTRY_DSN, then manually when through the errors to see what ends up in sentry.

[syphar]

@jyn514 IMHO you can have another look :)

I replaced log::error! with report_error and sometimes log::error! with log::warn!.

Right now I'm not sure

• if all the replacements are ok to replace
• if the remaining error! calls can stay like this
• if there are duplications where the same error would lead to multiple logs/sentry errors in some cases. we could also just fix it when this happens.

Also, from a rust perspective, if there is a better way than the report_error(&anyhow!(err).context(format!( construct.
I only can imagine:

• a single macro for anyhow with context/format?
• extending std-err so we can call .context on it, returning an anyhow-err? ( failure had this, anyhow doesn't)

[jyn514]

if there are duplications where the same error would lead to multiple logs/sentry errors in some cases. we could also just fix it when this happens.

Fixing them as they occur seems fine - I'd rather have too many alerts than too few.

if there is a better way than the report_error(&anyhow!(err).context(format!( construct.

Another macro seems fine. I don't know how to do it through traits without messing with it on desktop, and I don't think it needs to block the PR.

[syphar]

@jyn514 this is probably ready for another review :)

[jyn514]

LGTM :) @Nemo157 feel free to review too if you like, I won't merge this until the next time I do a deploy (so probably no sooner than Saturday).