• For a single ExecutionSession instance we should have unique JITDylib names

For now the JITDylib names are only used for debugging output (I don't have any plans to use them for anything else yet, either), but I think it would be reasonable to add this restriction. It would be an API contract though, so this should be verified by an assertion, rather than returning a runtime error. I would be inclined to add another method to ExecutionSession:

/// Returns the JITDylib with the given name, or nullptr if no such dylib exists.
JITDylib *getDylibByName(StringRef Name);

For now that could be implemented as a linear search through the JITDylibs list.

Then createJITDylib could be written as:

JITDylib &createJITDylib(StringRef Name) {
assert(!getDylibByName(Name) && "A dylib with that name already exists");
...
}

Good point. I agree that the API may enforce unique dylib names and it should be checked with assertions.

In case of createJITDylib(), though, the name must be considered user input (see lli's -jd param as an example). We may deduplicate here, so that subsequent code can relay on uniqueness (and use assertions).

Fix: https://reviews.llvm.org/D62139

In case of createJITDylib(), though, the name must be considered user input
(see lli's -jd param as an example). We may deduplicate here, so that
subsequent code can relay on uniqueness (and use assertions).

Good point. I think the answer is to document the contract and adds checks to lli. I have done this in r361322.

Praveen -- if you're happy with the fix please mark this as RESOLVED/FIXED. :)

Closing note:
Assertion added to the createJITDylib method. And the corresponding use cases in lli tool has be updated. (Thanks lang).