Evan Phoenix

@johanbrandhorst `type Tuple[T, ...]` isn't valid either, there are no varadic templates. So we'll end up with something like: ``` type Tuple1[T] struct { A T } type Tuple2[T1, T2]...

@eregon Good context. Perhaps another way to look at this is, by having a simple autowarm by default that isn't sophisticated, it's better than the default (strictly 2 seconds) without...

I think the send overhead is probably fine since the user would likely be measuring the results against previous results generated the same way. For super tight code, it might...

Both ARM64 and AMD64. Yup, benchmark all written and results provided here: https://github.com/lab47/lz4decode/actions/runs/7428506656

The reason this happens is that puma always uses chunked encoding because it's more efficient, server wise, when the body is being returned in sections via #each. rack-streaming-proxy assumes that...

Ah! If you're using puma, I'd suggest you use rack.hijack, which puma supports natively. That will allow you to very efficiently stream the output from your program directly to the...

@raggi Ok, thanks for the clarification. Puma already disables it's own chunked encoding if any Transfer-Encoding header is present, since that's the only safe assumption.

Hi! Sure, that makes sense and we should clarify that in the README. What I was really speaking to is that embedding JSON in syslog means putting the JSON into...

I understand and I do sympathize with your case. I think it's important to say where loss should take place. In your case, where the load of log transmission results...

A really good question. I think that the elements of the syslog message (facility, severity, etc) should not be lost. My preference would be to inject them at the top...