[cfe-dev] Using the Sema Object to Generate Implicit Definitions
Gregory Malecha via cfe-dev
cfe-dev at lists.llvm.org
Thu Oct 15 19:37:07 PDT 2020
Thanks! Worked like a charm.
On Thu, Oct 15, 2020 at 8:17 PM Richard Smith <richard at metafoo.co.uk> wrote:
> On Thu, 15 Oct 2020 at 14:23, Gregory Malecha via cfe-dev <
> cfe-dev at lists.llvm.org> wrote:
>
>> Hello --
>>
>> I've been trying to use the Sema object to generate implicit definitions,
>> e.g. default constructors, destructors, etc, in a tool that I have but I'm
>> running into some issues. The code that I have is a DeclVisitor with the
>> following for the CXXRecord case.
>>
>> void VisitCXXRecordDecl(CXXRecordDecl *decl, bool is_specialization) {
>> if (not(decl->isImplicit() or decl->isAnonymousStructOrUnion())) {
>> ci_->getSema().ForceDeclarationOfImplicitMembers(decl);
>> }
>> // ...
>> }
>>
>> This works great, and always works to get me constructor definitions with
>> empty bodies but marked as default.
>>
>> To generate the bodies, I have (roughly) the following code for
>> constructors, destructors, and methods:
>>
>> if (not decl->getBody() && decl->isDefaulted()) {
>> if (decl->isMoveAssignmentOperator()) {
>> ci_->getSema().DefineImplicitMoveAssignment(decl->getLocation(),
>> decl);
>> } else if (decl->isCopyAssignmentOperator()) {
>> ci_->getSema().DefineImplicitCopyAssignment(decl->getLocation(),
>> decl);
>> }
>> }
>>
>> In most cases, this seems to do the trick, but when the implementation of
>> the operator uses something like memcpy I get a segfault (stack trace
>> below) due (it seems) to the fact that TUScope is null when my code runs.
>>
>> #0 0x00007ffff5ff733a in clang::Sema::PushOnScopeChains(clang::NamedDecl*, clang::Scope*, bool) () from /usr/lib/libclang-cpp.so.10#1 0x00007ffff601dc18 in clang::Sema::LazilyCreateBuiltin(clang::IdentifierInfo*, unsigned int, clang::Scope*, bool, clang::SourceLocation) () from /usr/lib/libclang-cpp.so.10#2 0x00007ffff63599ac in clang::Sema::LookupBuiltin(clang::LookupResult&) () from /usr/lib/libclang-cpp.so.10#3 0x00007ffff636e713 in clang::Sema::LookupName(clang::LookupResult&, clang::Scope*, bool) () from /usr/lib/libclang-cpp.so.10#4 0x00007ffff61512e4 in ?? () from /usr/lib/libclang-cpp.so.10#5 0x00007ffff6151643 in ?? () from /usr/lib/libclang-cpp.so.10#6 0x00007ffff6159481 in clang::Sema::DefineImplicitCopyAssignment(clang::SourceLocation, clang::CXXMethodDecl*) () from /usr/lib/libclang-cpp.so.10
>>
>> I can avoid this problem if I put the preprocessor in "incremental
>> processing mode" using `enableIncrementalProcessing()` from
>> BeginSourceFileAction in my ASTFrontendAction, but doing this causes clang
>> to not generate the bodies of template specializations. I assume that this
>> is because you can't generate specializations until you have the complete
>> module which makes sense.
>>
>> Is there some way that I can get both of these, i.e. generate all
>> implicit declarations *with their bodies* and get template
>> specializations? Perhaps by using Sema to generate the bodies and then
>> somehow finishing the file to get clang to generate the specializations? Or
>> maybe I can manually call something in Sema to generate the bodies of the
>> templates?
>>
>
> You can call Sema::ActOnEndOfTranslationUnit() to get it to perform the
> pending actions that it's been saving up, such as template instantiations.
>
>
>> Thanks in advance.
>>
>> --
>> gregory malecha
>> gmalecha.github.io
>> _______________________________________________
>> cfe-dev mailing list
>> cfe-dev at lists.llvm.org
>> https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
>>
>
--
gregory malecha
gmalecha.github.io
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20201015/5bd5a70c/attachment-0001.html>
More information about the cfe-dev
mailing list