[ORC] Add Value support for capturing runtime JITed code results in clang-repl. #145263
Conversation
| #define BUILTIN_TYPES \ | ||
| X(bool, Bool) \ | ||
| X(char, Char_S) \ | ||
| X(int8_t, SChar) \ | ||
| X(uint8_t, Char_U) \ | ||
| X(uint8_t, UChar) \ | ||
| X(int16_t, Short) \ | ||
| X(uint16_t, UShort) \ | ||
| X(int32_t, Int) \ | ||
| X(uint32_t, UInt) \ | ||
| X(int64_t, LongLong) \ | ||
| X(uint64_t, ULongLong) \ | ||
| // X(long, Long) \ | ||
| // X(unsigned long, ULong) \ | ||
| // X(float, Float) \ | ||
| // X(double, Double) \ | ||
| // X(long double, LongDouble) |
There was a problem hiding this comment.
Is the idea here to define as many types as possible while remaining agnostic to the major data models?
Should Value just work in terms of raw sizes / signedness ([u]int8, [u]int16, ..., [u]int64) and then the compiler can map those into whatever data model it's using?
(I'm imagining the case where clang-repl might be used to inject code into some tiny microcontroller where the data model might be incompatible with the above)
There was a problem hiding this comment.
I’m using fixed-size types so they work across all platforms. In setValue, I cast long and long long to fixed sizes based on whether they’re 4 or 8 bytes—but I’m not sure that’ll work correctly on microcontrollers with different type sizes.
| if (!SPSArgList<SPSExecutorAddr>::serialize(OB, V.getOpaqueType())) | ||
| return false; |
There was a problem hiding this comment.
Are we passing around references to an opaque type class here? This type would only exist in the controller, right? Not the executor?
There was a problem hiding this comment.
Yes, but usually the Value class just stores the type and raw data together, which is why we only keep the address — even though I was thinking of tracking the type info on the controller side.
Adds a
Valueclass to support capturing the results of JIT-compiled expressions in clang-repl. This allows runtime values to be stored and passed through the ORC executor.