[SR-14069] Symbol not in TBD file when initializing a static member variable with an inline function

[swift-ci]

Previous ID	SR-14069
Radar	rdar://problem/83405989
Original Reporter	scentini (JIRA User)
Type	Bug

Additional Detail from JIRA

Votes	0
Component/s	Compiler
Labels	Bug, CxxInterop
Assignee	None
Priority	Medium

md5: 0bf3fb4aec12d9cfe2a92ea0cc046d58

Issue Description:

inline int increment(int t) { return t + 1; }

struct Incrementor { static int incrementee; };

int Incrementor::incrementee = increment(41);

inline int initializeStaticVar() { return Incrementor::incrementee; }

When using the initializeStaticVar() function from Swift, we get an error:

<unknown>:0: error: symbol '_ZN11Incrementor11incrementeeE' (_ZN11Incrementor11incrementeeE) is in generated IR file, but not in TBD file

<unknown>:0: error: please submit a bug report and include the project, and add '-Xfrontend -validate-tbd-against-ir=none' to squash the errors

The issue arises when using an inline function to initialize the static member variable.

[zoecarver]

@swift-ci create

[swift-ci]

Comment by Puyan Lotfi (JIRA)

I've added the repro steps to a branch here:

https://github.com/plotfi/cxx-interop-test/tree/tbd-SR-14069

[swift-ci]

Comment by Puyan Lotfi (JIRA)

I think in SwiftDeclConverter::VisitVarDecl in ImportDecl.cpp we may want to record somewhere that the var_decl isStatic. Later in swift::validateTBD we can know that specific externally visible symbol encountered in validateSymbols() in TBD.cpp comes from an inline definition or a static class variable definition. The cause of the verifier error here is that Incrementor::incrementee is not being added to the symbols list passed to validateSymbols because the TBDGenVisitor::visitFile does not encounter it as a top level Decl.

[plotfi]

@drodriguez @bulbazord

[drodriguez]

Initially it was unrelated to C++, but #40414 did fix a small bug in which the TBD validation was performed more than needed and was surfacing some of these errors incorrectly (in the case of that PR, when optimizations were enabled, but also -enable-testing, IIRC). It does not fix the problem, if there was one, it just does the validation only when it is supposed to happen.