[PATCH] D53483: [analyzer] Restrict AnalyzerOptions' interface so that non-checker objects have to be registered

[Artem Dergachev via Phabricator via cfe-commits]

NoQ accepted this revision.
NoQ added inline comments.


================
Comment at: include/clang/StaticAnalyzer/Core/AnalyzerOptions.h:146-149
+/// If you'd like to add a new -cc1 flag, add it to
+/// include/clang/Driver/CC1Options.td, add a new field to store the value of
+/// that flag in this class, and initialize it in
+/// lib/Frontend/CompilerInvocation.cpp.
----------------
Szelethus wrote:
> NoQ wrote:
> > Nono, don't add more -cc1 flags :)
> Code review is there to stop adding unnecessary -cc1 flags. Are we sure we wouldn't even like to document it? I myself will add at least 2 more -cc1 flags in the future (-analyzer-config-help, -analyzer-checker-option-help), that undoubtedly belong there.
But these flags wouldn't define new analyzer options(?)


================
Comment at: lib/StaticAnalyzer/Core/AnalyzerOptions.cpp:172
+                     .getAsInteger(10, *V);
+  assert(!HasFailed && "analyzer-config option should be numeric");
+  (void)HasFailed;
----------------
Szelethus wrote:
> NoQ wrote:
> > I guess @xazax.hun's comment also applies to this assert.
> Which is sadly only going to be addressed in followup patches :/ Note that I merely moved code around in this patch, but have organized things in a way that emitting proper diagnostics will be super easy going forward. Since analyzer configs are mainly meant for developers, I think this is acceptable before I fix it (in any case, it has always been like this).
Hmm, right. I guess @xazax.hun's comment also applies to all other asserts all over this code.


https://reviews.llvm.org/D53483