Good idea to try rebar3! I tried it a few years back and got stumped at building some NIFs and a few other issues. Most of those can be overcome with a bit of work.
I saw you used the pc compiler for NIFs, and I think rebar3 can set overrides now too https://github.com/blt/port_compiler#use-with-existing-dependency for individual dependencies. Perhaps the issue I encountered might be solved that way. The other issue was our "raw" dependencies - fauxton and docs. The approach I had tried was to make them look like Erlang apps. I don't recall how successful that was. But I think there are some newer things like https://rebar3.readme.io/docs/custom-dep-resources which we could leverage there. One more issue I recently discovered was that rebar3 is not Erlang 20 compatible. We ship our releases with the Erlang 20 as the minimum version and if we update to use rebar3 our release branch might be left behind. If the issue is just using Tag:Error:Stack catch patterns, we could potentially have our own fork of rebar3, as we already do for rebar2. Cheers, -Nick On Tue, Nov 23, 2021 at 5:39 PM Adam Kocoloski <kocol...@apache.org> wrote: > > Bah, ignore the message below. The issue isn’t that EUnit has a problem with > nested directories, it’s that it expects each test module to have a `_tests` > suffix, not a `_test` one. We’re inconsistent about adhering to that rule in > our codebase I’ll submit a PR to fix the module names so no one else bumps > into that issue. > > Adam > > > On Nov 23, 2021, at 3:48 PM, Adam Kocoloski <kocol...@apache.org> wrote: > > > > Hi, > > > > TL;DR - I think it would be a good idea to move EUnit tests back into > > `test` instead of `test/eunit`. > > > > Recently I’ve been spending some time porting specific libraries to use > > rebar3 and I thought it might be a nice project to hack on supporting > > rebar3 in CouchDB proper during the holiday weekend. I noticed a few small > > issues, but one in particular is that it was difficult to get rebar3 to > > pick up unit tests that were in subdirectories of `test`. > > > > Eventually I figured out that you could tell the compiler to do recursive > > compilation on these test directories and that worked for running the > > end-to-end test suite, but then I found that rebar3’s `—app` flag was still > > broken, e.g. if I ran `rebar3 eunit —app couch_log` it would only find > > tests in the actual source modules but not any of the modules in the > > test/eunit directory. I thought this might be an inconsistency bug in > > rebar3, but I realized that when you invoked rebar3 this way it was really > > just calling `eunit:test({application, couch_log})`, and EUnit doesn’t > > allow any configurability of its search path when invoked this way. > > > > Bottom line, it feels like we’re going against the grain of eunit itself > > when we try to move the eunit tests to anywhere else other than `test`. > > Currently we have a small handful of apps following this pattern: > > > > chttpd > > couch > > couch_epi > > couch_log > > couch_prometheus > > couch_replicator > > > > Of those, only the first two have any ExUnit unit tests that (I think) > > motivated the creation of separate eunit and exunit tests in the first > > place. Would it be OK if we moved those test modules back into `test` so we > > can stay on the “paved path” for EUnit usage going forward? > > > > Cheers, Adam >
