Here’s the CLI command….

npx shadow-cljs -A:cljs:git-deps:test compile test-ci && node cljs-out/test-ci/node-tests.js

and the config

{:target           :node-test
                                :compiler-options {:optimizations :whitespace}
                                :output-to        "cljs-out/test-ci/node-tests.js"
                                ;:ns-regexp        "^\b([a-z\\-]+)\b(?<!render|ignoreme)-test"
                                ; cannot use autorun if return code is needed. see shadow docs link above
                                :autorun          true}

I’ve tried this without autorun as well but same problem

@thheller if there’s nothing obvious then I’ll try a local docker repro next. I’d guess its a docker env problem. CI is often painful

again. in what context is the CLI command called?

this is relevant since that dictates how the error code is handled

I'm also not a docker expert or any container expert. so no idea what the rules of those even are and if they even check error codes

@thheller I got to the bottom of this one. I was using node fs/mkdirSync and that was failing in CI (not on localhost). If an fs/foo call throws an exception, the return code from the tests is zero. once I fixed the fs call, the tests properly returned non-zero for actual test fails

is this interesting enough for an issue?

node -e "require('fs').mkdirSync(null)" echo $! does not have a 0 exit code?

do you have an "uncaughtException" handler that prevents the actual exit?

I’m running a chain of promises and the fs fail is in the last one. maybe that’s the key? I’m using promesa for the chain

if you want a repro I can add a sample test to an issue

I would totally buy a shadow cljs shirt.

hehe nice