Why is BasicBlock's copy constructor private?

Project Infrastructure LLVM Dev List Archives
1

Hi Trevor,

I am not 100% sure why you should not copy a BasicBlock, however I believe two times the same BasicBlock does not make any sense in a CFG. There would be inconsistencies as e.g if a BB A is the successor of B there are two equal basic blocks A and A' after copying A, but only one (A) can be the real successor. The other one (A') is not referenced, but still references B as the predecessor of A'.
So for me the reason is that BasicBlocks are not independent and can just be copied. They are connected to their environment and cannot be exactly the same. Even the statements in the basicblock always have to be different, as every register in LLVM can only be written once per function.

To copy a BasicBlock in LLVM you should probably create a new one and copy the statements into the new BasicBlock. This obviously does not work for boost.

Some basic functionality for graph traversal is already in LLVM (as you mentioned) like iterating over edges, children or doing stuff like depthfirstsearch or post order search.
What kind of graph algorithm would you like to execute that is not available in LLVM?

Tobi

2

Well, iterating over the edges was one of the first things I'd wanted to do. But I spent hours reading the source code and couldn't figure out how. AFAICT there are no actual "edge" objects that can be iterated over.

The edges also need to have IDs assigned to them so that, for example, if Basic Block "A" has a successor "B", the outgoing edge of A will have the same ID as the incoming edge of B.

But it wasn't just this functionality I needed from Boost. The sample code and documentation would also be a big help, as I was having trouble figuring out LLVM's graph functionality just from its source code. That's why I'm trying to convert LLVM's CFG into a Boost graph.

Trevor