Skip to content

[R-pkg-devel] truncating strings on purpose

5 messages · Brad Eck, Ivan Krylov, Iris Simmons

#
Dear List

The latest update to my epanet2toolkit package bounced from CRAN due to
some warnings about string truncation in C code from gcc.

Found the following significant warnings:
    input2.c:863:5: warning: ?__builtin_strncpy? output may be truncated
copying 225 bytes from a string of length 255 [-Wstringop-truncation]
    input2.c:302:15: warning: ?__builtin_strncpy? output may be truncated
copying 78 bytes from a string of length 1024 [-Wstringop-truncation]
    report.c:795:9: warning: ?__builtin_strncpy? output may be truncated
copying 967 bytes from a string of length 1024 [-Wstringop-truncation]

The package wraps a simulation engine that parses inputs from and
writes results to text files. Where the truncation warnings appear, I
am truncating on purpose using `strncpy` to avoid overflows later.  I
believe this works fine from an application point of view but the
warnings remain.

Any suggestions on how to carry out a truncation so that gcc won't
complain?

Thanks in advance,

Brad

PS - CRAN check info is here
https://win-builder.r-project.org/incoming_pretest/epanet2toolkit_1.0.6_20241209_221743/specialChecks/

PPS the latest CRAN submission is from this branch
https://github.com/bradleyjeck/epanet2toolkit/tree/lto-crash
#
? Wed, 11 Dec 2024 12:12:11 +0000
Brad Eck <bradleyjeck at gmail.com> ?????:
GCC documentation suggests setting the length argument to _one less_
than sizeof destination buffer in addition to terminating the string:
https://gcc.gnu.org/onlinedocs/gcc-14.2.0/gcc/Warning-Options.html#index-Wstringop-truncation

        strncpy(trunc_s1, s1, sizeof trunc_s1 - 1);
        trunc_s1[sizeof trunc_s1 - 1] = '\0';
#
You could use memcpy instead. That gets rids of the type checking though.

You could also define something like char * my_strncpy ( char *
destination, const char * source, size_t num ) { return (char *)
memcpy(destination, source, num) } to keep the type checking.
On Wed, Dec 11, 2024, 07:12 Brad Eck <bradleyjeck at gmail.com> wrote:

            

  
  
2 days later
#
Hi Ivan, List,

Thanks for the suggestion, and for the pointer on duplicating the error.
The good news is that I am able to trigger the warning locally now. The bad
news is that following the gcc docs on avoid the warning did not seem to
work:


gcc -std=gnu99 -fsanitize=undefined -fno-omit-frame-pointer -std=gnu11
-I"/usr/local/lib/R/include" -DNDEBUG   -I/usr/local/include
-fsanitize=address
-O2 -Wstringop-truncation -fPIC  -g -O2  -c input2.c -o input2.o

input2.c: In function ?inperrmsg?:

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

  863 |     strncpy(trunc_tok, tok, sizeof trunc_tok - 1);

      |     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~




I could also try memcpy, but i don't so much like the idea of going away
from string functions for string handling.


Other suggestions are also welcome.


Brad
On Wed, Dec 11, 2024 at 1:29?PM Ivan Krylov <ikrylov at disroot.org> wrote:

            

  
  
2 days later
#
Dear Brad, Kurt,

? Fri, 13 Dec 2024 18:08:27 +0000
Brad Eck <bradleyjeck at gmail.com> ?????:
I have asked on Libera.Chat's #gcc and was told that the "middle-end"
warnings [1] (which include -Wstringop-truncation) should not count as
failures when Address Sanitizer is enabled. (Many thanks to Sam James
and Arsen Arsenovi? for the triage!) While we don't literally use
-Werror, de-facto the result is the same: the package doesn't pass the
check if the warning is emitted.

The reason for that is that the "middle-end" warnings require
optimisation to avoid false positives in the unreachable code paths,
and sanitizers change optimisation too much for the warnings to remain
reliable [2]. A similar false positive has been reported before [3].

Can we use -Wno-stringop-truncation during incoming checks with ASan?