mir-algorithm: Assert error for CI master

[jmh530]

I recently fixed up mir-algorithm's CI and have been getting persistent issues with failures on ldc-master. It fails on ubuntu, windows, and macos. For whatever reason, not a problem for dmd-master or ldc-latest.

Below is the output for x86_64 ubuntu. This run is based on commit 9e1df01. The full log is here (and you can see the other runs, at least as long as the link still works).
https://github.com/libmir/mir-algorithm/actions/runs/21124059607/job/62376521508

It looks like this code in dmd (https://github.com/dlang/dmd/blob/master/compiler/src/dmd/expressionsem.d) is different from ldc (https://github.com/ldc-developers/ldc/blob/master/dmd/expressionsem.d), but I have no idea how it works or anything.

Run dub test --arch=$ARCH --combined
  
             Generating test runner configuration 'mir-algorithm-test-default' for 'default' (library).
    Starting Performing "unittest" build using /opt/hostedtoolcache/ldc2/9e1df017/x64/ldc2-9e1df017-linux-x86_64/bin/ldc2 for x86_64.
    Building mir-algorithm ~master: building configuration [mir-algorithm-test-default]
Mir warning: can't compute hash. Expexted `size_t toHash() scope @safe const pure nothrow @nogc` method for Algebraic!(Ion_)
Warning: multivariate cubic spline with derivatives was not tested!!!
Warning: multivariate cubic spline with derivatives was not tested!!!
core.exception.AssertError@/home/runner/work/ldc/ldc/dmd/expressionsem.d(11753): Assertion failure
----------------
??:? _d_assert [0x56244480f09f]
??:? ExpressionSemanticVisitor::visit(AssignExp*) [0x562444995066]
??:? void dmd.statementsem.statementSemanticVisit(dmd.statement.Statement, dmd.dscope.Scope*).visit.VisitStatement(dmd.statement.Statement) [0x562444825524]
??:? void dmd.statementsem.statementSemanticVisit(dmd.statement.Statement, dmd.dscope.Scope*).visit.VisitStatement(dmd.statement.Statement) [0x562444824ac6]
??:? Semantic3Visitor::visit(FuncDeclaration*) [0x5624448fe80e]
??:? Semantic3Visitor::visit(TemplateInstance*) [0x5624448fa84a]
??:? void dmd.templatesem.templateInstanceSemantic(dmd.dtemplate.TemplateInstance, dmd.dscope.Scope*, dmd.expression.ArgumentList) [0x562444782fc4]
??:? void dmd.templatesem.functionResolve(ref dmd.declaration.MatchAccumulator, dmd.dsymbol.Dsymbol, dmd.location.Loc, dmd.dscope.Scope*, dmd.root.array.Array!(dmd.rootobject.RootObject).Array*, dmd.mtype.Type, dmd.expression.ArgumentList, void delegate(const(char)*) scope) [0x562444a4768e]
??:? dmd.expression.Expression dmd.opover.pickBestBinaryOverload(dmd.dscope.Scope*, dmd.root.array.Array!(dmd.rootobject.RootObject).Array*, dmd.dsymbol.Dsymbol, dmd.dsymbol.Dsymbol, dmd.expression.BinExp, out bool) [0x562444a5610b]
??:? dmd.expression.Expression dmd.opover.opOverloadAssign(dmd.expression.AssignExp, dmd.dscope.Scope*, dmd.mtype.Type[2]) [0x562444a5646b]
??:? ExpressionSemanticVisitor::visit(AssignExp*) [0x56244498f1ce]
??:? void dmd.statementsem.statementSemanticVisit(dmd.statement.Statement, dmd.dscope.Scope*).visit.VisitStatement(dmd.statement.Statement) [0x562444825524]
??:? void dmd.statementsem.statementSemanticVisit(dmd.statement.Statement, dmd.dscope.Scope*).visit.VisitStatement(dmd.statement.Statement) [0x562444824ac6]
??:? Semantic3Visitor::visit(FuncDeclaration*) [0x5624448fe80e]
??:? ConditionalDeclaration::accept(Visitor*) [0x56244492167d]
??:? Semantic3Visitor::visit(AttribDeclaration*) [0x56244491ee79]
??:? Semantic3Visitor::visit(Module*)laration*) [0x56244479e899]
??:? mars_tryMain(Param&, Array<char const*>&) [0x562444a80fe5]
??:? cppmain() [0x562444afc49f]
??:? void rt.dmain2._d_run_main2(char[][], ulong, extern (C) int function(char[][])*).runAll() [0x5624496178cc]
??:? _d_run_main2 [0x5624496176ad]
??:? _d_run_main [0x5624496174cc]
??:? main [0x56244942c8a0]
??:? [0x7fbfffe2a1c9]
??:? __libc_start_main [0x7fbfffe2a28a]
??:? _start [0x562444b01414]
Error /opt/hostedtoolcache/ldc2/9e1df017/x64/ldc2-9e1df017-linux-x86_64/bin/ldc2 failed with exit code 1.

[kinke]

I'm pretty sure I've seen that specific assert being violated in the past (in private Symmetry code most likely):

ldc/dmd/expressionsem.d

Lines 11752 to 11753 in b6e3298

	// SliceExp cannot have static array type without context inference.
	assert(exp.e1.op != EXP.slice);

I'm also pretty sure this is purely from upstream - but there aren't any DMD builds with enabled assertions AFAIK, so users don't see it. And for the prebuilt LDC packages, it's only the master package coming with enabled assertions.

So looks like the violated assertion isn't crucial (code still working), but the expectation of the author not met.

A reduced testcase would be very helpful, to prove that it fails with a DMD debug build too and then file it upstream.

[jmh530]

@kinke Thanks for the prompt reply.

I really have no sense from the message above where the error is coming from. To help narrow down where the problem is, I was thinking I would start a new project that just the minimal number of files for mir-algorithm and keep adding files until I trigger the error with ldc-master. Does this make sense to you? I'm also not really keen to build my own DMD with a debug build.

[kinke]

In the interest of a usable minimal test case (ideally, a little single file), you'll almost certainly need to dustmite it, checking that the assert is still hit with an LDC master build.