[Lldb-commits] [lldb] [lldb][rpc] Fix build failures when building lldb-rpc-gen (PR #151603)

[Chelsea Cassanova via lldb-commits]

chelcassanova wrote:

> Yes, as a user, if you want to point to a preexisting executable, you'd still set it in LLDB_RPC_GEN_EXE, but that variable is empty if the user didn't set anything, and then when the cmake functions either use the user supplied value, or a locally compiled one, the resolved value is set in lldb_rpc_gen_exe.

Noted, I'll change how we set the variable for the tool.

> But this seems to assume that everybody's cross builds are done in some specific sequence, with regards to how LLVM_HOST_TRIPLE and LLVM_DEFAULT_TARGET_TRIPLE are set (in my case, I just cross build with one single step, and both of those are set to the same in that single stage) - so I'm not really sure if we can use them like this for detecting what kind of situation we are in.

> The safest, with regards to cross builds, probably is to default to not enabling this when cross compiling, unless either LLDB_CAN_USE_LLDB_RPC_SERVER or LLDB_RPC_GEN_EXE is set by the user.

Given how many configurations there are to account for here this does look to be the best option.

> Yes, but it's also (theoretically) possibly to be doing a universal build for 2 architectures, while neither of them is the native host one; so you can't generally assume that one of them is the host architecture. So for that case it's probably fine to just pick the first/any architecture out of the ones included.

Hm, I hadn't thought about it that way. In that case then we can modify this to process files for the first architecture for the header files.

https://github.com/llvm/llvm-project/pull/151603