Replace TE->Scalars with VL, as suggested.

I think this is the cause of the crash: https://reviews.llvm.org/D144895
@ABataev could you take a look ?

Hi - I noticed it doesn't appear that getIFuncList was actually made private by this one.

Thanks, let me fix it.

Yeah sorry about that, I reverted it here: b5f239363a3 Revert "[NFC][IR] Make Module::getGlobalList() private"

And what is this needed for / what will this be used for?

This is to make it easier to track changes to the attributes for Checkpointing, but regardless I think this is generally useful e.g., for debugging, as you can watch all changes from a single point.

It funnels all changes to attributes through a single function, which makes it really easy to track all attribute changes from a single point.

Rebase

Removed commented out lines from ModuleTest.cpp

Rebase

Rebase

Rebase

Addressed comments

Rebase

Removed newline

Rebase

BB->eraseFromParent() is a better option.

If the intention is to make it behave like a container, then the erase method should take an iterator.
But who would possibly use it if it requires a BB reference to call the method? It is kind of easier and safer to just I->eraseFromParent().
I don't mind adding this, just don't see how it can be useful.

I->eraseFromParent() now calls BasicBlock::erase(BB). Also BasicBlock::erase(BB) now calls InstList.erase() directly instead of calling BasicBlock::erase(FromIt, ToIt).

It is exactly the same. There is no super strong reason for having both functions.
The main reason is that the BB behaves like a list as it implements begin(), end(), size(), splice() etc. so not having an erase(I) function looks strange.
Up until now one could simply do BB->getInstList().erase(I), instead of I->eraseFromParent(), so this function is covering this case.

Rebase

Rebase

Fixed a couple more files that were missing from the original patch.

Both examples should now be fixed. Updated the doc too. Thanks for your help @aeubanks .