Conversation
Sorry, something went wrong.
The C standard (before C23) says that giving invalid pointers, even with length=0, to memcpy() is undefined behaviour. R returns an invalid pointer for vectors of length 0, so avoid calls to memcpy() altogether in such cases.
Codecov ReportAll modified and coverable lines are covered by tests ✅ Additional details and impacted files @@ Coverage Diff @@
## master #6820 +/- ##
=======================================
Coverage 98.64% 98.64%
=======================================
Files 79 79
Lines 14642 14650 +8
=======================================
+ Hits 14444 14452 +8
Misses 198 198 ☔ View full report in Codecov by Sentry. |
Sorry, something went wrong.
|
Generated via commit 4773ceb Download link for the artifact containing the test results: ↓ atime-results.zip
|
Sorry, something went wrong.
MichaelChirico
left a comment
There was a problem hiding this comment.
LGTM, feel free to merge whether or not you want to make my suggested change.
Do we have any way of preventing this from recurring in the future?
Sorry, something went wrong.
Co-authored-by: Michael Chirico <chiricom@google.com>
Co-authored-by: Michael Chirico <chiricom@google.com>
|
Maybe not literally preventing, but a sanitizer-enabled CI check should be able to catch similar problems as soon as they land on master. Then why does rocker/r-devel-san-clang built using the same Debian Testing clang miss this, while our own container doesn't? Not sure. Maybe due to -O3 optimizing some of it away? |
Sorry, something went wrong.
* Avoid memcpy()-ing 0-length vectors The C standard (before C23) says that giving invalid pointers, even with length=0, to memcpy() is undefined behaviour. R returns an invalid pointer for vectors of length 0, so avoid calls to memcpy() altogether in such cases. * Use early returns for 0-length cases Co-authored-by: Michael Chirico <chiricom@google.com>
* Avoid memcpy()-ing 0-length vectors The C standard (before C23) says that giving invalid pointers, even with length=0, to memcpy() is undefined behaviour. R returns an invalid pointer for vectors of length 0, so avoid calls to memcpy() altogether in such cases. * Use early returns for 0-length cases Co-authored-by: Michael Chirico <chiricom@google.com>
* Avoid `memcpy()`-ing 0-length vectors (#6820) * Avoid memcpy()-ing 0-length vectors The C standard (before C23) says that giving invalid pointers, even with length=0, to memcpy() is undefined behaviour. R returns an invalid pointer for vectors of length 0, so avoid calls to memcpy() altogether in such cases. * Use early returns for 0-length cases Co-authored-by: Michael Chirico <chiricom@google.com> * Avoid memcpy() of empty vector (#6911) * Avoid memcpy() of empty vector * in getidcols() too * Add a test case Distilled from the `melt()` call here: https://github.com/gdemin/expss/blob/d498caf72a9938ac71b9e06fd61485cd0347d607/tests/testthat/test_subtotal.R#L161-L166 * NEWS entry --------- Co-authored-by: Michael Chirico <chiricom@google.com> Co-authored-by: Michael Chirico <michaelchirico4@gmail.com>
The C standard (before C23) says that giving invalid pointers, even with length=0, to memcpy() is undefined behaviour. R returns an invalid pointer for vectors of length 0, so avoid calls to memcpy() altogether in such cases.
Fixes: #6819