[cfe-dev] Use of Smart Pointers in LLVM Projects

[David Blaikie]

There seems to be some uncertainty about the use of smart pointers
(previously OwningPtr, now std::unique_ptr and std::shared_ptr
predominantly) in the LLVM project as a whole, so here's a thread to
discuss/clarify/etc the project preferences/direction with regard to
smart pointer usage.

For some context, see discussions in LLVM r212403 and Clang r213307.

The basic question here seems to be whether smart pointer ownership is
a direction we want to take the LLVM project in general.

I've seen others contribute and have myself contributed many patches
moving towards smart pointer ownership (both in the pre-C++11 days of
OwningPtr, and much moreso in the post-C++11 world with
std::unique_ptr and std::shared_ptr being usable inside containers, as
return values, etc, allowing many more opportunities).

std::unique_ptr's been used in LLD as far back as r153620.
std::unique_ptr appeared in LLVM shortly after the C++11 switch with
Ahmed's work to migrate the project from OwningPtr to std::unique_ptr
(starting with r202609 and ending with r211259). Originally OwningPtr
was added in r45261.
Something in the order of 60 changes across clang and LLVM mention
unique_ptr in their subject and migrate various APIs to use unique_ptr
for ownership. Many of which remove uses of explicit delete or helpers
like DeleteContainerPointers (and removing explicit dtors in many of
those cases).

Are people OK with/prefer the use of owning smart pointers in APIs?
Are there places where you've found them to be too noisy/burdensome
and would rather use raw pointers or some other abstraction? Would you
prefer pre-commit review of such changes to adequately consider
alternatives (such as?)?

Thanks,
- David