argh! Of course! That makes so much sense. I hadn't thought of it like that

Exception for sure. It would be up to the UI in a particular domain to decide to display 0 or “n/a” or other visualization.

In a language like Scala you might use an “Option” here, and for any aggregate that is not defined for multiple items

Or you make a special “nothing” object which is basically the same idea

For it to be defined, you need to be able to combine the “nothing” average with an existing average and get the existing thing back out

I could see NaN here instead of a divide by zero exception, since 0/0 is NaN.

You might, but often this sort of aggregation is done in error rather than intentionally.

eg. “7-day average of daily average latencies” to give you a smoothing function

Better than percentiles, you actually want the histogram. Just looking at key percentiles throws away a LOT of data. (granularity of the histogram is an interesting topic, but the idea is the same: Retain your data when aggregating)

Agree, raw data is best, but in a choice between average and a list of percentiles, I prefer the latter

Oh certainly agreed, a list of percentiles (ie. a histogram 😉 ) is very useful for understanding the shape of a data set (ie. the distribution). Average does have its uses though, in that it doesn’t throw away any data, whereas a single percentile throws away most of it. For latency tracking, a “trimmed mean” is often a really good statistic, as it has the benefits of Average (single number to interpret, not throwing away most of the data), but also deals w/ the most common problem with averages: Outliers.

Usually for latency I'd be satisfied with 50, 90, 99, up to 6 nines

Yes, it is often the case that these are what get looked at, though I think that is more of a tooling artifact than being driven by our goals. Usually we want to go faster for everyone, not just those that happen to feel these percentiles. From my observations, looking at specific percentiles tends to cause the latency histogram to have “bumps” and consolidate just below these signposts, primarily because latency improvements for requests in the faster 50% of the population don’t get “counted” when we look at what changes moved the needle. Don’t get me wrong, any attention to latency is good, but I think we can do better than just a handful of percentiles

"it depends" Maybe I'm spoiled by Gil Tene's talks on latency, but in certain scales those percentiles are a real pain, I care about the high percentiles, to an extent where I don't care about the bumps as much as I care about the tail in plenty of circumstances

I’m only getting started with Nix myself, but w/ my limited knowledge I’d use a shell.nix in the project’s directory. I’m currently trying https://github.com/nix-community/lorri to keep the current shell up-to-date wrt shell.nix.

It looks like the https://github.com/NixOS/nixpkgs/blob/nixos-20.09/pkgs/development/interpreters/clojure/default.nix uses JDK11 for the clojure/`clj` commands.

My editor is Calva and I'm doing a deps.edn project.

This file https://github.com/github/gitignore/blob/218a941be92679ce67d0484547e3e142b2f5f6f0/Clojure.gitignore suggests that I should make a new file and reference it for deps.edn projects.

Anybody have a different/better idea?

I always just fill in the gitignore when i see files i don’t want tracked

i don’t think any of those entries except for .cpcache are useful for me

Same, but if there's a template I'd like it to be generally useful.

I added

# Calva
.calva/
.clj-kondo/
.lsp/

sure. that template looks geared towards the stuff that lein makes so it doesn’t seem that useful to me

i think you want to ignore .clj-kondo/.cache and .lsp/.cache . Both of those directories can have useful stuff you want checked in

no idea about .calva

that was such a confusing message. you want to ignore .clj-kondo/.cache. not just the root .clj-kondo since there are useful config files in there

here are my additions at this point

# Calva
.calva/output-window/
.clj-kondo/.cache/
.lsp/.cache/

thanks for the tips!

If you haven't already, you should check out Gibo https://github.com/simonwhitaker/gibo for gitignore boilerplates. Starting up a project I do gibo dump clojure > .gitignore and get

pom.xml
pom.xml.asc
*.jar
*.class
/lib/
/classes/
/target/
/checkouts/
.lein-deps-sum
.lein-repl-history
.lein-plugins/
.lein-failures
.nrepl-port
.cpcache/

I often use the the gitignore file generated by this site: https://gitignore.io/

If you create a Clojure project with deps-new, you get this https://github.com/seancorfield/deps-new/blob/develop/resources/org/corfield/new/app/root/.gitignore (it's the same for a lib)

Interestingly enough, it's straight up an error in CLJS: https://clojure.atlassian.net/browse/CLJS-254

Oh, and the Q&A where I found the link ties back to the issue 792, heh.

https://groups.google.com/g/clojure-dev/c/6EEfdPhrWgk - no definitive answers, but some opinions from well known people.

Ah, found it, by Alex Miller: https://clojurians-log.clojureverse.org/clojure/2020-07-27/1595872486.471000 Where the linked article by Stuart Sierra says: > Use each clause at most once.

Yeah, but he said it quite literally recently somewhere

Thanks for these references btw