IsAllocator redesigned and intercommunicating with GetAllocType by keremsahn · Pull Request #1072 · compiler-research/CppInterOp

keremsahn · 2026-07-23T18:15:18Z

IsAllocator now returns an AllocType value instead of bool, and it recognizes user added attrbiutes with specific keyword:
cppAllocAllocTypeEnum
For example

[[clang::annotate("cppAllocMalloc")]] void* mergeFunc();

cppAllocMalloc attribute is added to mergeFunc with specific signature, and IsAllocator recognizes cppAllocMalloc attribute as AllocType::Malloc.

Also traverser now asks IsAllocator whether the callee is labeled with an attribute before recursively analyzing it
@Vipul-Cariappa @vgvassilev

codecov · 2026-07-23T18:19:59Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.80%. Comparing base (9bbdbb2) to head (3d7c271).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1072      +/-   ##
==========================================
+ Coverage   87.74%   87.80%   +0.06%     
==========================================
  Files          23       23              
  Lines        6429     6454      +25     
==========================================
+ Hits         5641     5667      +26     
+ Misses        788      787       -1

Files with missing lines	Coverage Δ
include/CppInterOp/CppInterOpTypes.h	`96.93% <ø> (ø)`
lib/CppInterOp/CppInterOp.cpp	`90.64% <100.00%> (+0.09%)`	⬆️

Files with missing lines	Coverage Δ
include/CppInterOp/CppInterOpTypes.h	`96.93% <ø> (ø)`
lib/CppInterOp/CppInterOp.cpp	`90.64% <100.00%> (+0.09%)`	⬆️

🚀 New features to boost your workflow:

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

github-actions

clang-tidy made some suggestions

github-actions · 2026-07-23T18:38:39Z


+#include <optional>
+
 namespace Cpp {


warning: included header optional is not used directly [misc-include-cleaner]

Suggested change

namespace Cpp {

usplus

t

github-actions · 2026-07-23T18:38:39Z

 }

-bool IsAllocator(ConstFuncRef Fn) {
+std::optional<AllocType> IsAllocator(ConstFuncRef Fn) {


warning: no header providing "Cpp::AllocType" is directly included [misc-include-cleaner]

tr); ^

github-actions · 2026-07-23T18:38:39Z

    return RecursiveASTVisitor::TraverseDecl(D);
  }

+  std::optional<AllocType> join(std::optional<AllocType> a,


warning: method 'join' can be made static [readability-convert-member-functions-to-static]

Suggested change

std::optional<AllocType> join(std::optional<AllocType> a,

);static

github-actions · 2026-07-23T18:38:39Z

+            !COCE->isInfixBinaryOp())
+          return handleExpr(COCE->getArg(0));
+      }
+      return handleCall(const_cast<CallExpr*>(CE));


warning: do not use const_cast to remove const qualifier [cppcoreguidelines-pro-type-const-cast]

} ^

github-actions · 2026-07-23T18:38:39Z

@@ -0,0 +1,13 @@
+#ifndef UNITTESTS_CPPINTEROP_TESTATTRIBUTEMERGE_H
+#define UNITTESTS_CPPINTEROP_TESTATTRIBUTEMERGE_H


warning: header guard does not follow preferred style [llvm-header-guard]

Suggested change

#define UNITTESTS_CPPINTEROP_TESTATTRIBUTEMERGE_H

#ifndef GITHUB_WORKSPACE_UNITTESTS_CPPINTEROP_APINOTES_TESTATTRIBUTEMERGE_H

#define GITHUB_WORKSPACE_UNITTESTS_CPPINTEROP_APINOTES_TESTATTRIBUTEMERGE_H

keremsahn · 2026-07-25T14:39:14Z

@Vipul-Cariappa @vgvassilev

I found an important issue while developing recursion support in deallocation. In this PR attribute injection is done in static function not in the API itself, but in recursive cases function in cache can be evaluated wrong even though the main function is always evaluated true, and the false evaluation is injected to the recursively analyzed functions. Therefore attribute injection mechanism should be moved to API function, which is not called in recursive cases

Vipul-Cariappa

Initial quick review.
This looks fine to me. But I have a few questions. I have not looked at the tests yet, nor at the record support part of the changes.
Is it possible for this PR to be independent of #1083? This PR was opened before that.

Vipul-Cariappa · 2026-08-21T12:26:54Z

-AnalyzeAllocType(const clang::FunctionDecl* Fn,
+AnalyzeAllocType(clang::FunctionDecl* Fn,


Why is this change required?

First version of this PR was injecting attributes after analyzing, and I found serious problems with that and removed the feature, but forgot to change const and const_cast. I reversed them git diff is clear now.

Also it is not dependent to any other PR now, it would be better to merge this PR first

Vipul-Cariappa · 2026-08-21T12:28:21Z

+      if (attrName == "cppAllocUnknown")
+        return INTEROP_RETURN(AllocType::Unknown);


Not required. Why should we support "unknown" attribute?

You are right, I put it there to be consistent with enum

We also do not need Null, I guess? It is only useful in recursion, to not effect other returns

Vipul-Cariappa · 2026-08-21T12:29:39Z

+    for (const auto* attr : FD->specific_attrs<clang::AnnotateAttr>()) {
+      llvm::StringRef attrName = attr->getAnnotation();


Can you check if this same code also works with APINotes? If not, what additional changes are required?

No, I need another loop like

for (const auto* attr : FD->specific_attrs<clang::SwiftAttrAttr>())

Vipul-Cariappa · 2026-08-21T12:32:38Z

 }

-bool IsAllocator(ConstFuncRef Fn) {
+std::optional<AllocType> IsAllocator(ConstFuncRef Fn) {


Why introduce std::optional? Can we represent no attribute available with Unknown?
You can condition on Unknown in AnalyzeAllocType.

You are also right about that, I put optional to distinguish no-attr functions with None-Unknown attred functions, for recursive base case in analyzer, but it makes sense to remove cppAllocUnknown and we can use this value for no-attr situation

github-actions

clang-tidy made some suggestions

github-actions · 2026-08-21T23:39:33Z

 }

-bool IsAllocator(ConstFuncRef Fn) {
+AllocType IsAllocator(ConstFuncRef Fn) {


warning: no header providing "Cpp::AllocType" is directly included [misc-include-cleaner]

tr); ^

Vipul-Cariappa

Can we have one test case for a templated function and one for methods?

One more test where the annotation is not present in the original function. For example:

int* foo() { return new int; }
__attribute__((annotate("cppAllocNew"))) int* foo();

Vipul-Cariappa · 2026-08-31T05:15:22Z

+  // Nullopt is returned because when analyzer calls this API it should know
+  // whether an attribute injected before or FD has a meaningful attribute
+  return INTEROP_RETURN(AllocType::Unknown);


Is this comment relevant?

No, I forgot to delete it after removing attribute injection

Vipul-Cariappa · 2026-08-31T05:27:11Z

+    void* __attribute__((cf_returns_retained)) CFAllocFunc();
+
+    void __attribute__((annotate("cppAllocNone"))) NoneFunc();
+    void __attribute__((annotate("cppAllocNew"))) NewFunc();


A void function, that returns New!!

😅 Since i did not implement or call any of them and the functions are just for attribute holding, I did not care but I can change it if you think it is required

keremsahn · 2026-08-31T22:08:31Z

Can we have one test case for a templated function and one for methods?

One more test where the annotation is not present in the original function. For example:
int* foo() { return new int; }
__attribute__((annotate("cppAllocNew"))) int* foo();

I added one test for a method, but testing redeclaration is not responsibility of this function because this function is just there to recognize whether the given decl has the target attributes, but redeclaration is more about clang's merging mechanism and which decl you take and give to test, so lookup. These tests are already written in CppJIT and they pass because CppJIT's lookup mechanism takes the last decl, in despite of CppInterOp's GetNamed that takes the canonical decl which will not include the attribute if you redeclare the function after implementing it.

And for the templates, I am aware they do not enter in both analyzer and IsAllocator because they are FunctionTemplateDecl, but CppJIT gives the instantiated version so both works fine except caching when you call function with different T parameters. We can discuss it but I do not add it for now because I know it will not pass

github-actions

clang-tidy made some suggestions

github-actions · 2026-08-31T22:30:12Z

 }

-bool IsAllocator(ConstFuncRef Fn) {
+AllocType IsAllocator(ConstFuncRef Fn) {


warning: no header providing "Cpp::AllocType" is directly included [misc-include-cleaner]

RETURN(nullptr); ^

IsAllocator now returns std::optional<AllocType> instead of bool, and reads back cppAllocX clang::annotate attributes in addition to the native ones it already understood (RestrictAttr, OwnershipAttr, the CFReturnsRetained family). A hand-annotated or API-notes-annotated function is therefore recognised without walking its body at all. nullopt means no signal was found, which is deliberately distinct from AllocType::None -- the latter is a definite answer, the former tells the analyzer to keep looking. AllocationTraverser::handleCall consults IsAllocator on a callee before recursing into it, so a call to a declaration-only function that already carries an attribute short-circuits instead of degrading to Unknown. GetAllocType's parameter changed from ConstFuncRef to FuncRef and the .td was updated to match. Also fixes <optional> leaking into the C-only branch of CppInterOpTypes.h, which broke the plain-C build, and pins TracingTests.cpp to -std=c++17 since the interpreter otherwise defaults to C++14 and cannot parse std::optional in the public headers. Tests cover the full IsAllocator branch matrix, attribute merging across redeclarations (both a single function and an overload set, via a pragma clang attribute push block in a preincluded header), and the handleCall short-circuit on a body-less annotated callee.

…wnership attribute when our specific signature is inserted which is cppAlloc`EnumValue` and EnumValue can not be null or unknown, furthermore I changed return type to raw enum from optional wrapped, because we can use Unknown value to pass no-attr information

…ed and unnecessary comment lines are removed in IsAllocator

github-actions Bot reviewed Jul 23, 2026

View reviewed changes

keremsahn force-pushed the attr-design branch 3 times, most recently from 8f98ada to fc0e7da Compare July 23, 2026 21:47

keremsahn force-pushed the attr-design branch from fc0e7da to 62e8599 Compare July 27, 2026 16:34

keremsahn force-pushed the attr-design branch 2 times, most recently from d5e66ef to 98115cd Compare August 20, 2026 19:55

Vipul-Cariappa reviewed Aug 21, 2026

View reviewed changes

github-actions Bot reviewed Aug 21, 2026

View reviewed changes

keremsahn force-pushed the attr-design branch 3 times, most recently from 0d585c4 to f89d189 Compare August 27, 2026 08:55

Vipul-Cariappa reviewed Aug 31, 2026

View reviewed changes

keremsahn force-pushed the attr-design branch from 6114323 to 39499fb Compare August 31, 2026 22:13

github-actions Bot reviewed Aug 31, 2026

View reviewed changes

keremsahn force-pushed the attr-design branch from 39499fb to f8925d1 Compare September 1, 2026 17:37

keremsahn added 2 commits September 1, 2026 20:37

keremsahn force-pushed the attr-design branch from f8925d1 to fd2ba8d Compare September 1, 2026 17:37

IsAllocator now can process FunctionTemplateDecl, also some tests add…

3d7c271

…ed and unnecessary comment lines are removed in IsAllocator

keremsahn force-pushed the attr-design branch from fd2ba8d to 3d7c271 Compare September 1, 2026 17:47

keremsahn requested a review from Vipul-Cariappa September 1, 2026 18:47

	std::optional<AllocType> join(std::optional<AllocType> a,
	);static

		@@ -0,0 +1,13 @@
		#ifndef UNITTESTS_CPPINTEROP_TESTATTRIBUTEMERGE_H
		#define UNITTESTS_CPPINTEROP_TESTATTRIBUTEMERGE_H

	#define UNITTESTS_CPPINTEROP_TESTATTRIBUTEMERGE_H
	#ifndef GITHUB_WORKSPACE_UNITTESTS_CPPINTEROP_APINOTES_TESTATTRIBUTEMERGE_H
	#define GITHUB_WORKSPACE_UNITTESTS_CPPINTEROP_APINOTES_TESTATTRIBUTEMERGE_H

		AnalyzeAllocType(const clang::FunctionDecl* Fn,
		AnalyzeAllocType(clang::FunctionDecl* Fn,

		if (attrName == "cppAllocUnknown")
		return INTEROP_RETURN(AllocType::Unknown);

		for (const auto* attr : FD->specific_attrs<clang::AnnotateAttr>()) {
		llvm::StringRef attrName = attr->getAnnotation();

Uh oh!

Conversation

keremsahn commented Jul 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov Bot commented Jul 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

github-actions Bot left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions Bot Jul 23, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions Bot Jul 23, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions Bot Jul 23, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions Bot Jul 23, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions Bot Jul 23, 2026

Choose a reason for hiding this comment

Uh oh!

keremsahn commented Jul 25, 2026

Uh oh!

Vipul-Cariappa left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

keremsahn Aug 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

github-actions Bot left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions Bot Aug 21, 2026

Choose a reason for hiding this comment

Uh oh!

Vipul-Cariappa left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

keremsahn Aug 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

keremsahn commented Aug 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions Bot Aug 31, 2026

Choose a reason for hiding this comment

Uh oh!

keremsahn commented Jul 23, 2026 •

edited

Loading

codecov Bot commented Jul 23, 2026 •

edited

Loading

keremsahn Aug 21, 2026 •

edited

Loading

keremsahn Aug 31, 2026 •

edited

Loading

keremsahn commented Aug 31, 2026 •

edited

Loading