[PATCH] D77866: [analyzer][CallAndMessage] Add checker options for each bug category

Tue Jun 2 04:58:36 PDT 2020

Szelethus added a comment.

In D77866#2068394 <https://reviews.llvm.org/D77866#2068394>, @NoQ wrote:

> We (me, Valeriy, Devin)'ve just been talking about this and mostly of agreed that `core.CallAndMessage` should ideally be removed (at least as a `Checkers.td` entity) and individual checks moved to the relevant package (null dereference, use of uninitialized value, etc.). It won't hurt backwards compatibility too much because so far there was no reasonable outcome that one might expect from disabling `core.CallAndMessage`, given that it's a weird mix of completely unrelated checks. The benefit is that disabling, say, null dereference checker will actually disable all null dereference warnings (as opposed to having to additionally disable the ones that come from `CallAndMessage`).

Luckily this should be super easy to implement after this patch, and it is totally what I, and most folks on Ericsson's end want this checker to look like. I trust that you've glanced over the comments that justify the tiptoeing creating smaller checkers, and it is a result of a conversation we had that the price on the user would have to pay the way we implement our database is too great.

I'll initialize a new discussion on this on our end, but speaking from **my** perspective, we should just deal the cost of this and be done with it. However, since sooooo many reports originate from this checker, I wonder whether we should go all the way and make sense out of the hash generation we currently have.

> Also a regular reminder that we wouldn't have had this conversation if we had hashtags instead of packages (#CallAndMessage #NullDereference #C++ #AllPlatforms #UndefinedBehavior #SecurityCritical #PathSensitive #Stable #2012).

Could you detail this a bit?

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D77866/new/

https://reviews.llvm.org/D77866