Skip to content

[R-pkg-devel] duplicating gcc-UBSAN warning

2 messages · Brad Eck, Ivan Krylov

Brad Eck
# 
Dear List

I'm trying to duplicate a warning CRAN is seeing on the additional checks
for gcc-UBSAN.  That is using:

CRAN's gcc-UBSAN:
* using R Under development (unstable) (2024-12-04 r87420)
* using platform: x86_64-pc-linux-gnu
* R was compiled by
    gcc-14 (GCC) 14.2.0
    GNU Fortran (GCC) 14.2.0
* running under: Fedora Linux 36 (Workstation Edition)

And gave warnings like.
    input2.c:863:5: warning: ?__builtin_strncpy? output may be truncated
copying 225 bytes from a string of length 255 [-Wstringop-truncation]

To try and duplicate, I added a src/Makevars with
PKG_CFLAGS = -Wstringop-truncation

And ran in two different docker images:
Docker #1 (fedora:40)
* using R version 4.4.2 (2024-10-31)

* using platform: aarch64-redhat-linux-gnu

* R was compiled by

    gcc (GCC) 14.2.1 20240912 (Red Hat 14.2.1-3)

    GNU Fortran (GCC) 14.2.1 20240912 (Red Hat 14.2.1-3)

* running under: Fedora Linux 40 (Container Image)

Docker #2 (thanks to Tomas Kalibera)

* using R Under development (unstable) (2024-12-02 r87417)

* using platform: aarch64-unknown-linux-gnu

* R was compiled by

    gcc (Debian 14.2.0-8) 14.2.0

    GNU Fortran (Debian 14.2.0-8) 14.2.0

* running under: Debian GNU/Linux trixie/sid

The flag appears in 00install.out, but the warning doesn't get raised.

Any ideas on how to duplicate?

Thanks,

Brad

PS checks for my package are here:
https://www.stats.ox.ac.uk/pub/bdr/memtests/gcc-UBSAN/epanet2toolkit/00check.log
Ivan Krylov
# 
? Thu, 12 Dec 2024 14:46:08 +0000
Brad Eck <bradleyjeck at gmail.com> ?????:
In my experiments with gcc 14.2.0-8 on Debian for amd64 (the
debian:testing container), the warning only appears when all of the
following flags are present:

-fsanitize=address -O2 -Wstringop-truncation

-Wall can be used instead of -Wstringop-truncation because the former
includes the latter. -fsanitize=address,undefined also works, but not
just -fsanitize=undefined.

Remove the optimisation or the AddressSanitizer, and GCC stops being
able to prove that truncation occurs. It looks like you will either
need to find a container with R pre-built with sanitizers enabled (are
there Rocker containers for aarch64?) or build R yourself [*].