Add interface for lookup up of externally stored type descriptors #69556
Conversation
augusto2112
force-pushed
the
builtin-interface
branch
from
5f92b33 to
3eaf708
Compare
|
@slavapestov this is my approach for looking up descriptors that live outside of reflection metadata. With this approach we can keep the entire type information building logic inside remote mirrors, and only implement in LLDB the logic to look up the descriptors. |
|
@swift-ci smoke test |
|
@swift-ci smoke test |
augusto2112
force-pushed
the
builtin-interface
branch
from
3eaf708 to
77b4b00
Compare
|
@swift-ci smoke test |
augusto2112
force-pushed
the
builtin-interface
branch
from
77b4b00 to
8046460
Compare
|
@swift-ci smoke test |
Separate the functionality that reads descriptors from reflection information into its own class.
augusto2112
force-pushed
the
builtin-interface
branch
2 times, most recently
from
0c212b3 to
d739887
Compare
|
@swift-ci smoke test |
| struct DescriptorFinder { | ||
| virtual ~DescriptorFinder(){}; | ||
|
|
||
| virtual std::unique_ptr<BuiltinTypeDescriptorInterface> |
There was a problem hiding this comment.
Performance-wise, you could design the base class to already have all the data members needed by all implementations, maybe with an opaque element for implementation-specific data, then you could pass a stack-allocated object through this interface and don't need to use unique_ptr here.
There was a problem hiding this comment.
Personally I don't think the performance benefits would be worth the extra complexity. There's already indirection going on today when using RemoteRef, and there will be a pretty big difference in size between the DWARF version and the reflection version, so accommodating both on the base class would look a bit ugly imo.
There was a problem hiding this comment.
If this is only used in the debugger and similar clients, I think that's reasonable. If it's used in-process then I'm not sure.
There was a problem hiding this comment.
This is only used for out of process inspection.
augusto2112
force-pushed
the
builtin-interface
branch
from
d739887 to
3793b29
Compare
|
@swift-ci smoke test |
| NormalizedReflectionNameCache; | ||
|
|
||
| /// Cache for built-in type descriptor lookups. | ||
| std::unordered_map<std::string /* normalized name */, |
There was a problem hiding this comment.
Not for this patch, but since we have LLVM here, this should probably be a StringMap for better efficiency. Same for all the other caches.
augusto2112
force-pushed
the
builtin-interface
branch
from
3793b29 to
5327590
Compare
|
@swift-ci test |
|
@swift-ci smoke test |
Currently, TypeRefBuilder knows how to parse type descriptors from reflection metadata. We aim to store the same information that lives in type descriptors externally, as an effort to support embedded Swift debugging. In order to take advantage of all of the existing type information parsing code that exists today in remote mirrors, add an interface for external lookup of type descriptors, and replace all the usages of type descriptors with their generic counterpart (this patch only adds support for builtin descriptors, but follow up patches will add support for other descriptor types).
augusto2112
force-pushed
the
builtin-interface
branch
from
5327590 to
f09f518
Compare
|
@swift-ci smoke test |
Currently, TypeRefBuilder knows how to parse type descriptors from reflection metadata. We aim to store the same information that lives in type descriptors externally, as an effort to support embedded Swift debugging. In order to take advantage of all of the existing type information parsing code that exists today in remote mirrors, add an interface for external lookup of type descriptors, and replace all the usages of type descriptors with their generic counterpart (this patch only adds support for builtin descriptors, but follow up patches will add support for other descriptor types).