https://gcc.gnu.org/bugzilla/show_bug.cgi?id=38470
--- Comment #16 from Manuel López-Ibáñez <manu at gcc dot gnu.org> --- (In reply to Eric Gallager from comment #15) > That has since been closed as fixed. So are the chances of this one being > fixed next somewhat better now? Not really. PR23608 fixes the case where the variable is actually a "const" by assuming the initial value cannot be changed. There is no propagation of values; we still warn for "i = 5; i < 5u;" To fix the general case, someone has to implement some kind of reduced dataflow within the FE. Something that once you see "x >= 0", records this info somewhere, keeps track of any subsequent jumps, then when it sees "x < Unsigned", it tries to see if "x >= 0" was recorded and there is a (unconditional?) path from there to here. Doing this seems to me to require a non-trivial amount of work. The simplest case (x >= 0 && x < Unsigned) could perhaps be handled by some special (folding?) function that when seeing such expression, sets TREE_NO_WARNING for x, or converts the right-hand to (unsigned)x< Unsigned, or something clever that I cannot even imagine. If someone is interested in trying the above, look at warn_logical_operator in c-common.c, which is one of those warnings that looks at logical && expressions and tries to figure out whether to warn. You would need to create a similar function (and call it from wherever warn_logical_operator is called from), but you may need to rewrite the expressions or set TREE_NO_WARNING or do something else to avoid warning later. This seems much less work, but still far from trivial and it will only catch this very specific case.
