Improve wording of uninitialized read warning #321
Conversation
|
Check-perf-impact results: (000e2892abb21ddd1ae413a5c86d7d95) ❓ No new benchmark data submitted. ❓ |
psalz
force-pushed
the
improve-uninit-read-warning
branch
from
58591aa to
442b951
Compare
Pull Request Test Coverage Report for Build 12411400550Details
💛 - Coveralls |
| bool is_pure_consumer_access = true; | ||
| for(size_t i = 0; i < access_map.get_num_accesses(); ++i) { | ||
| const auto [b, mode] = access_map.get_nth_access(i); | ||
| if(b == bid && is_producer_mode(mode)) { | ||
| is_pure_consumer_access = false; | ||
| break; | ||
| } | ||
| } |
There was a problem hiding this comment.
This is also just a heuristic btw; there could be an uninitialized read and an initialized write on the same buffer and it would print the no_init advice. Unfortunately I don't think this can be avoided unless we do the check on a per-accessor basis (which would mean separately evaluating each range mapper, followed by a region map query).
There was a problem hiding this comment.
This is still better than what we have now! ... and we should at some point improve the buffer_access_map interface so this can just be a std::any_of.
The previous warning suggested to use
no_initeven for pure-consumer (read_only) accesses, which never made any sense.