Continue analysing other namespaces after a failure

[danielcompton]

I noticed shadow-cljs imports were failing. Looking at a recent build, I could see the error was:

Execution error (ClassNotFoundException) at java.net.URLClassLoader/findClass (URLClassLoader.java:476).\norg.graalvm.polyglot.PolyglotException$StackFrame\n",
  :clojure.main/triage
  {:clojure.error/class java.lang.ClassNotFoundException,
   :clojure.error/line 476,
   :clojure.error/cause
   "org.graalvm.polyglot.PolyglotException$StackFrame",
   :clojure.error/symbol java.net.URLClassLoader/findClass,
   :clojure.error/source "URLClassLoader.java",
   :clojure.error/phase :execution},

Even if we can't analyse that one namespace, there are probably other namespaces that could be analysed. It seems like it would be good to have an option to keep analysing other namespaces. At the end of the analysis, cljdoc could still show that there were errors, and in the namespace sidebar show the failed namespace in red.

This would let the rest of the library continue to be analysed, even if one part fails.

I'm not well versed in how this all works, so maybe this isn't possible/a good idea.

[lread]

Hi @danielcompton thanks for raising this issue!

There is something raised around this cljdoc/cljdoc#228.

My guess was that this might get weird/tricky for load-time analysis.
When you analyze ns x it may require nses y and z, then z may include nses c and d.
And d might be the one that fails. Last time I looked, I think it was for cljs, it was a challenge to detect where the failure had occurred.

At one point I was of the opinion that if your code can't load it should fail analysis.
I also felt that we would not be doing devs/readers a favour by including partial API docs.
But maybe that's a bit of a hard stance, I dunno.
Happy to discuss.

Related issues are: cljdoc/cljdoc#565, cljdoc/cljdoc#543 and cljdoc/cljdoc#628

[danielcompton]

And d might be the one that fails. Last time I looked, I think it was for cljs, it was a challenge to detect where the failure had occurred.

You could still track failures by the entrypoint. If x can't be loaded because of a problem in d, it's still a problem for x.

At one point I was of the opinion that if your code can't load it should fail analysis.

Yeah, I can see it both ways, you might be right about it being confusing if there are only partial docs. People don't always read everything to realise there has been a partial failure.