Add GC preserve to prevent segfaults by blegat · Pull Request #182 · JuliaInterop/JavaCall.jl · GitHub

The bug: In _jcall, only savedArgs is protected by GC.@preserve.

For most argument types, convert_arg returns (x, x) so preserving savedArgs implicitly preserves convertedArgs. But for array arguments (Vector{jint}, Vector{T<:JavaObject}, etc.), convert_arg returns different objects:

JavaCall.jl/src/convert.jl

Lines 98 to 121 in 7e1e4e0

    
                   function convert_arg(argtype::Type{Array{$x,1}}, arg) 
        
                       carg = convert(argtype, arg) 
        
                       sz=length(carg) 
        
                       arrayptr = $y(sz) 
        
                       arrayptr === C_NULL && geterror() 
        
                       $z(arrayptr, 0, sz, carg) 
        
                       # Wrap into JavaObjects so we can delete the local refs 
        
                       return carg, JavaObject{argtype}(arrayptr) 
        
                   end 
        
               end 
        
               eval( m) 
        
           end 
        
           function convert_arg(argtype::Type{Array{T,1}}, arg) where T<:JavaObject 
        
               carg = convert(argtype, arg) 
        
               sz = length(carg) 
        
               init = sz == 0 ? C_NULL : Ptr(carg[1])  
        
               arrayptr = JNI.NewObjectArray(sz, Ptr(metaclass(T)), init) 
        
               arrayptr === C_NULL && geterror() 
        
               for i=2:sz 
        
                   JNI.SetObjectArrayElement(arrayptr, i-1, Ptr(carg[i])) 
        
               end 
        
               return carg, JavaObject{argtype}(arrayptr) 
        
           end

In that case savedArgs[i] = carg, the Julia Vector{jint} while convertedArgs[i] is a new JavaObject wrapping the JNI array pointer.

This JavaObject has a finalizer that calls DeleteLocalRef. Since it's not in GC.@preserve, GC can finalize it during the ccall, deleting the JNI array while the JVM is reading from it. Similarly, obj has its pointer extracted via Ptr(obj) and is not referenced afterwards, so it could also be collected.

I couldn't write a self-contained reproducer that reliably segfaults (the GC timing race is hard to trigger in isolation), but the fix resolves a non-deterministic segfault I get when running https://github.com/JuliaConstraints/JaCoP.jl with the MathOptInterface's test suite that was kind of reproducible before this PR when running ~1200 tests creating many Java objects with array arguments. After this PR, the segfault disappears.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.90%. Comparing base (2f1824f) to head (3f0e8c1).
⚠️ Report is 21 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #182      +/-   ##
==========================================
+ Coverage   67.01%   67.90%   +0.88%     
==========================================
  Files          11       11              
  Lines        1043     1050       +7     
==========================================
+ Hits          699      713      +14     
+ Misses        344      337       -7

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

This seems reasonable. Thank you.

Add GC preserve to prevent segfaults

3f0e8c1

mkitti merged commit 45b1a19 into JuliaInterop:master Apr 13, 2026
8 checks passed

Add GC preserve to prevent segfaults by blegat · Pull Request #182 · JuliaInterop/JavaCall.jl · GitHub

Uh oh!

Add GC preserve to prevent segfaults - #182

Add GC preserve to prevent segfaults#182
mkitti merged 1 commit into
JuliaInterop:masterfrom
blegat:segfault

blegat commented Apr 13, 2026

Uh oh!

codecov-commenter commented Apr 13, 2026 •
edited
Loading

Uh oh!

Uh oh!

mkitti commented Apr 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

	function convert_arg(argtype::Type{Array{$x,1}}, arg)
	carg = convert(argtype, arg)
	sz=length(carg)
	arrayptr = $y(sz)
	arrayptr === C_NULL && geterror()
	$z(arrayptr, 0, sz, carg)
	# Wrap into JavaObjects so we can delete the local refs
	return carg, JavaObject{argtype}(arrayptr)
	end
	end
	eval( m)
	end

	function convert_arg(argtype::Type{Array{T,1}}, arg) where T<:JavaObject
	carg = convert(argtype, arg)
	sz = length(carg)
	init = sz == 0 ? C_NULL : Ptr(carg[1])
	arrayptr = JNI.NewObjectArray(sz, Ptr(metaclass(T)), init)
	arrayptr === C_NULL && geterror()
	for i=2:sz
	JNI.SetObjectArrayElement(arrayptr, i-1, Ptr(carg[i]))
	end
	return carg, JavaObject{argtype}(arrayptr)
	end

Add GC preserve to prevent segfaults by blegat · Pull Request #182 · JuliaInterop/JavaCall.jl · GitHub

Uh oh!

Add GC preserve to prevent segfaults - #182

Add GC preserve to prevent segfaults#182mkitti merged 1 commit intoJuliaInterop:masterJuliaInterop/JavaCall.jl:masterfrom blegat:segfaultblegat/JavaCall.jl:segfaultCopy head branch name to clipboard

Conversation

blegat commented Apr 13, 2026

Uh oh!

codecov-commenter commented Apr 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

mkitti commented Apr 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Add GC preserve to prevent segfaults#182
mkitti merged 1 commit into
JuliaInterop:masterfrom
blegat:segfault

codecov-commenter commented Apr 13, 2026 •
edited
Loading