StackOverflow error related to `malli.core/register-function-schema`

[samuelludwig]

Pardon the sparse issue, I've been scratching my head looking at this for a bit now, and have been unable to replicate it outside of our company's repo. The repo is Clojure-only, and we are on malli v0.20.

I've been speaking to @opqdonut about this in the Clojure slack over the last couple days.

We have 27 function-schemas across our repo, and maybe around 60 schemas of varying complexity (nothing breaking more than 34 lines (mostly map-schemas with little nesting) from my current measure).

I've been experiencing a StackOverflow, from the malli.core/register-function-schema function, when I do one of the following with malli dev-mode active:

• execute cider-ns-reload-all on a file containing a function schema definition (malli.core/=> ...) (of any complexity)
• execute (malli.dev/start!) after an exception has already occurred from the command above
• execute (malli.instrument/instrument!) (in some circumstance which I forget how to replicate atm)

This StackOverflow stops happening, and will not happen again until I start a new REPL, when I do one of the following:

• execute (malli.core/-deregister-function-schemas! :clj)
• instrument the malli.core namespace with the FlowStorm debugger (I believe this re-evals the malli namespaces, so perhaps thats why, it may be indirectly doing the same thing as the -deregister-...)
• *turn off dev-mode (this does cause the error to happen upon start!ing it again, however)

After the overflow occurs, calls to (malli.core/function-schemas :clj) return an illegal-argument exception, with the message: No implementation of method: :-form of protocol: #'malli.core/Schema found for class: malli.core$__EQ__GT$reify$reify__40157

This reify appears in the stack traces quite a bit, I'm not sure where it might be connected responsibility-wise, but it seems to get called ~7k times when this error occurs. I've seen malli.registry/composite-registry also called a significant amount of times during error.

When instrumenting malli.* namespaces from the beginning and stepping through with FlowStorm, I did notice a lot of flipping between a -vmap function and a check-children! function, which seemed to be failing? Stepping through has not revealed a clear cause among schemas or namespaces yet (credit to its name, the stack overflow does indeed contain quite a bit of executed code).

Here is a snippet of the overflow:

#error {
 :cause nil
 :via
 [{:type java.lang.StackOverflowError
   :message nil
   :at [malli.core$_lookup invokeStatic "core.cljc" 320]}]
 :trace
 [[malli.core$_lookup invokeStatic "core.cljc" 320]
  [malli.core$_lookup invoke "core.cljc" 320]
  [malli.core$_lookup_BANG_ invokeStatic "core.cljc" 331]
  [malli.core$_lookup_BANG_ invoke "core.cljc" 329]
  [malli.core$schema invokeStatic "core.cljc" 2553]
  [malli.core$schema invoke "core.cljc" 2531]
  [malli.core$_schema_schema$reify__447771$fn__447772 invoke "core.cljc" 2045]
  [malli.impl.util$_vmap invokeStatic "util.cljc" 28]
  [malli.impl.util$_vmap invoke "util.cljc" 23]
  [malli.core$_vmap invokeStatic "core.cljc" 266]
  [malli.core$_vmap invoke "core.cljc" 266]
  [malli.core$_schema_schema$reify__447771 _into_schema "core.cljc" 2045]
  [malli.core$_lookup invokeStatic "core.cljc" 327]
  [malli.core$_lookup invoke "core.cljc" 320]
  [malli.core$_lookup_BANG_ invokeStatic "core.cljc" 331]
  [malli.core$_lookup_BANG_ invoke "core.cljc" 329]
  [malli.core$schema invokeStatic "core.cljc" 2553]
  [malli.core$schema invoke "core.cljc" 2531]
  [malli.core$_schema_schema$reify__447771$fn__447772 invoke "core.cljc" 2045]
  [malli.impl.util$_vmap invokeStatic "util.cljc" 28]
  [malli.impl.util$_vmap invoke "util.cljc" 23]
  [malli.core$_vmap invokeStatic "core.cljc" 266]
  [malli.core$_vmap invoke "core.cljc" 266]
  [malli.core$_schema_schema$reify__447771 _into_schema "core.cljc" 2045]
  [malli.core$_lookup invokeStatic "core.cljc" 327]
  [malli.core$_lookup invoke "core.cljc" 320]
  [malli.core$_lookup_BANG_ invokeStatic "core.cljc" 331]
  [malli.core$_lookup_BANG_ invoke "core.cljc" 329]
  [malli.core$schema invokeStatic "core.cljc" 2553]
  [malli.core$schema invoke "core.cljc" 2531]
  [malli.core$_schema_schema$reify__447771$fn__447772 invoke "core.cljc" 2045]
  [malli.impl.util$_vmap invokeStatic "util.cljc" 28]
  [malli.impl.util$_vmap invoke "util.cljc" 23]
  [malli.core$_vmap invokeStatic "core.cljc" 266]
  [malli.core$_vmap invoke "core.cljc" 266]
  [malli.core$_schema_schema$reify__447771 _into_schema "core.cljc" 2045]
  [malli.core$_lookup invokeStatic "core.cljc" 327]
  [malli.core$_lookup invoke "core.cljc" 320]
  [malli.core$_lookup_BANG_ invokeStatic "core.cljc" 331]
  [malli.core$_lookup_BANG_ invoke "core.cljc" 329]
  [malli.core$schema invokeStatic "core.cljc" 2553]
  [malli.core$schema invoke "core.cljc" 2531]
  [malli.core$_schema_schema$reify__447771$fn__447772 invoke "core.cljc" 2045]

Let me know if there's anything I could run/try to get some more-useful data/insights!

[opqdonut]

My guess would be that cider-ns-reload-all redefines some protocols, but there is some data or code that's referring to the old protocols, and this is causing an infinite loop for some reason.

One thing that might help would be to add some debug prints to -lookup and -lookup! to see what kind of data it's looping on.

[samuelludwig]

Alright, peering into -lookup! calls, I see some normal-seeming arglists, but then a lot of the following arglist (for >1000 subsequent calls, until I'm assuming the overflow seems to happen):

0. malli.core$__EQ__GT$reify$reify__40289@1fae51b2
1. malli.core$__EQ__GT$reify$reify__40289@1fae51b2
2. nil
3. false
4. nil

Following one of the executions, it looks like this is returned by -instrument-f at some point, but I can't seem to step into that invocation easily

I did note that, in the function-schemas list during one of these calls, the value for every fn-schema that is outside of the namespace I run the reload-all from seems to be

• malli.core$_proxy_schema$reify$reify__<nums>@<hexnums>

I'll add that these ...reify... values all seem to have the type of :malli.core/schema

[samuelludwig]

Pardon the screenshot. This is the call-tree from the ns-reload (which should be equivalent to (require (quote <my-ns>) :reload-all)). I think this shows the loop it appears to be getting trapped in:

• -lookup! -> reify--91775.-into-schema -> -vmap -> -schema-schema/reify--91775/fn--91776 -> schema -> -lookup! (repeat)