[cfe-dev] [Frontends] Ask "meaningful things" from Sema
Whisperity via cfe-dev
cfe-dev at lists.llvm.org
Wed Feb 5 03:17:09 PST 2020
Currently, to my understanding, Sema analyses the source code to build the
AST. Much of its methods are non-const, and a friend into the AST building
process. This is not a problem in itself, as a semantic error will
hard-break the compilation process.
However, Clang has evolved into something much much more than just a simple
compiler, and there are several potential checks that naturally could
belong to Sema only during "AST building time". Still, we would benefit if
we could call them from front-end code. For example, static analysis, Tidy,
project-wide visualisations, metrics, refactoring suggestions - a plethora
of developer tooling armed (potentially) with the ASTImporter could explore
more options in these efforts.
(My side-step idea is to just manually create C++ code and spawn another
compiler and somehow fetch diagnostic results, but that's way too much work
for something that should(?) be easily accessible.)
The most striking example that's on top of my mind's stack is checking
whether a type A expr could be implicit-converted to a type B value. While
of course, everyone could reimplement the logic based on the standard... it
would be better if we used the same logic as per which the compilation
breaks in case someone wants to compile an expression that can't be
converted. (And then language dialects and releases would be "implicitly"
So, given this particular example, I might be looking for a function with a
signature as follows:
/** True if 'LHSVariableType Var = (RHSExprType)foo;' is legit,
according to the current standard being used. **/
bool tryCanImplicitlyConvert(const QualType RHSExprType, const QualType
LHSVariableType, bool WithAllowingUserDefinedConversion = true);
- Am I missing something and does this already exist with a different name?
- If not, why doesn't this exist? Is it a historical reason, or a
complexity issue? Sema is a Lovecraftian horror already, I wouldn't be
surprised if this is the reason...
- How would one (or (a subset of) the community) go about
implementing/refactoring to allow this, if possible?
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the cfe-dev