https://github.com/usx95 commented:
Great work on implementing flow history tracing! This would significantly
improves the diagnostics. A few suggestions for enhancement:
1. **Simplify source highlighting and improve wording**: Rather than
highlighting both sides of each step in the flow chain, we should only
highlight the destination (left-hand side). Additionally, make the diagnostic
text configurable based on what we know: use `"variable 'x'"` when we have a
named variable, `"function call result"` for call expressions, `"temporary
object"` for temporaries, etc. (Apologies for not mentioning this in the
original issue!)
For example, given:
The desired diagnostic output would be:
For chained member/arrow operations, we should highlight each step in the
chain (with distinct ranges):
Would produce:
Notice how:
- We only highlight the destination (`a`, `b`, `e`, function call, `s`,
`x`) in each step of the flow
- Each note traces back to the original source (`tgt` or the temporary)
- For chained operations, we highlight progressively longer prefixes
showing the flow through the chain in a single statement.
- Function calls and member accesses are described appropriately when they
appear in the flow
- All involved entities use distinct source ranges (avoiding duplicate
reports for the same range)
2. **Extend to other error types**: This tracing mechanism would be valuable
for:
- Use-after-return errors
- Lifetime annotation suggestions
- Dangling field/global warnings
Consider refactoring the tracing logic to be reusable across these
different diagnostic paths.
The implementation looks solid overall. Will take a deeper look at the tracking
logic after the functional changes.
Feel free to ask follow up questions.
https://github.com/llvm/llvm-project/pull/188467
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
