@snoop_inference taking a very long time?

johnomotani · October 17, 2025, 11:43am

Running @snoop_inference on a function from my package is taking a very long time - it’s been running for an hour since the function finished executing and hasn’t returned yet. Is this a known issue in any cases? Any suggestions for debugging? The function itself takes about 3 minutes to compile/run (mostly compile) and is admittedly pretty complicated - but that’s why I want to profile the compilation/inference!

I’m using Julia v1.12.0, SnoopCompileCore v3.1.1. Some simple ‘hello world’ type example in the REPL works fine, and completes almost instantaneously.

johnomotani · October 17, 2025, 12:42pm

I tried profiling the @snoop_inference call, and found that it seems to be taking a huge amount of time here

github.com/JuliaDebug/SnoopCompile.jl

SnoopCompileCore/src/snoop_inference.jl

a0a03f3b2


      
          for (i, ci) in pairs(cis)
              ci ∈ handled && continue
              # Follow the backedges to the root
              j = i
              be = ci
              while true
                  found = false
                  for k in backedges[j]
                      k < j && continue   # don't get caught in cycles
                      be = cis[k]
                      if be ∉ handled
                          j = k
                          found = true
                          break
                      end
                  end
                  found || break  # quit when no unhandled backedge found
              end
              be ∈ handled && continue
              # bt1, bt2 = get(backtraces, Core.Compiler.get_ci_mi(be), (nothing, nothing))
              # child = InferenceTimingNode(be, make_stacktrace(bt1, bt2), parent)

Maybe this should be expected if the tree is very large, but my original call is up to about 2 hours now, and still not finished…

johnomotani · October 17, 2025, 2:03pm

Maybe this is a bug?

github.com/JuliaDebug/SnoopCompile.jl

Inifinite loop in add_children!()

opened 02:03PM - 17 Oct 25 UTC

johnomotani

I have a case where I get stuck in an infinite loop here https://github.com/Juli…aDebug/SnoopCompile.jl/blob/a0a03f3b2943a6275cd58aa3dec64d172bda264e/SnoopCompileCore/src/snoop_inference.jl#L127-L155 I think what's happening is that `be ∉ handled` is true the first time it is tested, for `k == j`. That means that `j` never increments, and the loop never ends. I don't understand what's being done here, but if I change line 137 to ```julia k ≤ j && continue # don't get caught in cycles ``` instead of `k < j`, then there's no infinite loop, and the output looks reasonable. Is this a correct fix?

Topic		Replies	Views
Profiling compilation/inference Performance inference	10	1456	August 24, 2020
Making sense of output of Snoopcompile Tooling	0	281	September 2, 2021
New tools for reducing compiler latency Package Announcements	7	4037	January 28, 2021
What next after SnoopCompile? Performance	7	918	March 20, 2023
Measuring time of type inference Internals & Design	8	933	March 25, 2019

@snoop_inference taking a very long time?

Related topics