[PATCH] D81315: [analyzer] Warning for default constructed unique pointer dereferences

Artem Dergachev via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Mon Jun 29 08:36:49 PDT 2020


NoQ added inline comments.


================
Comment at: clang/include/clang/StaticAnalyzer/Checkers/Checkers.td:577-583
+  CheckerOptions<[
+    CmdLineOption<Boolean,
+                  "CheckSmartPtrDereference",
+                  "Enable check for SmartPtr null dereferences",
+                  "false",
+                  InAlpha>,
+  ]>,
----------------
Szelethus wrote:
> This goes against D81750 -- Sorry for not bringing this up earlier, but you can't emit diagnostics from a checker that is hidden :/
> 
> The proper solution would be to create a non-hidden checker, and emit a diagnostic under its name. You can take inspiration from `MallocChecker`, `NullabilityChecker`, and a variety of other sound in the stack of this patch: D77845
Aha, ok, that's what i meant at the end of D81315#inline-760088. @Szelethus insists, so let's turn ourselves into a separate checker right away!

In fact, let's *not* take the inspiration from poor-man's solutions in `MallocChecker` etc., but instead let's try a full-on modular approach:
- Make a separate .cpp file and a separate Checker class for emitting diagnostics.
  - Remove the checker option and instead add a separate checker entry in Checkers.td.
- Subscribe the new checker on PreCall only and move all the bug reporting logic there.
- Keep all the modeling logic in the old checker (it's called SmartPtr//Modeling// for a reason!).
- Put common functionality into a header file shared between the two checkers.
  - In particular, the GDM trait should be only accessed through such getter.
    - Take some inspiration from `Taint.h`.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D81315/new/

https://reviews.llvm.org/D81315





More information about the cfe-commits mailing list