Thanks Bill for taking the time to look into this, the patch, and the
enlightening comments.
I incorporated it and it resolved the valgrind issues. An old compiler issue
resurfaced (Wstringop-truncation warning from strncopy) whose dodgy fix had
introduced the current bug but it is now fixed.
As to valgrind, I adapted a Github actions workflow kindly provided by Martin
R. Smith which reproduced the errors from CRAN.
Georgi Boshnakov
==============================
From: Bill Dunlap <williamwdun...@gmail.com>
Sent: 07 August 2021 03:09
To: Georgi Boshnakov <georgi.boshna...@manchester.ac.uk>
Cc: r-package-devel@r-project.org
Subject: Re: [R-pkg-devel] can't reproduce 'Additional issues' on CRAN with
valgrind
I think the following change to str.c, in str_strcpy_internal(), fixes the
valgrind issues. It was easier to track down after the calloc() was changed to
malloc(), since before it only happened after a realloc().
$ diff -u rbibutils/src/str.c rbibutils-fixed/src/str.c
--- rbibutils/src/str.c 2021-06-16 04:13:48.000000000 -0700
+++ rbibutils-fixed/src/str.c 2021-08-06 18:38:56.763556500 -0700
@@ -217,8 +217,8 @@
unsigned long size = str_initlen;
assert( s );
if ( minsize > str_initlen ) size = minsize;
- // 2021-06-16 was: s->data = (char *) malloc( sizeof( *(s->data) ) *
size );
- // changing to calloc() to avoid this kind of error from valgrind:
+ s->data = (char *) malloc( sizeof( *(s->data) ) * size );
+ // tried changing to calloc() to avoid this kind of error from
valgrind:
// > bibConvert(fn_med, bib, informat = "med")
// ==16041== Conditional jump or move depends on uninitialised
value(s)
// ==16041== at 0x10CCF2A3: xml_processtag (xml.c:174)
@@ -249,8 +249,9 @@
//
// TODO:
// The data is not really left uninitialised and there may be a
better way to let the compiler know.
+ // WWD: fixing str_strcpy_internal takes care of memory misuse.
//
- s->data = (char *) calloc( size, sizeof( *(s->data) ) );
+ // s->data = (char *) calloc( size, sizeof( *(s->data) ) );
if ( !s->data ) {
error("Error. Cannot allocate memory in str_initalloc, requested %lu
characters.\n\n", size );
// error("\n"); // error( EXIT_FAILURE );
@@ -552,9 +553,9 @@
// Georgi: this fixes the warning about truncation in strncpy
// strcpy cannot be used here since at least one of the calls below
// passes a non-NULL terminated 'p'
- strncpy( s->data, p, n + 1);
- // strncpy( s->data, p, n );
- // s->data[n] = '\0';
+ // strncpy( s->data, p, n + 1); // WWD: ???
+ strncpy( s->data, p, n );
+ s->data[n] = '\0';
s->len = n;
}
I don't know why valgrind isn't working for you.
-Bill
On Mon, Aug 2, 2021 at 4:23 PM Georgi Boshnakov
<mailto:georgi.boshna...@manchester.ac.uk> wrote:
Thanks for looking into this. I probably didn’t make it clear that I am not
questioning the errors on CRAN but my own configuration. I have pretty good
idea where the error comes from, since the only change from the previous CRAN
version was in medin.c (function medin_readf from line 94 and most probably
line 144-145). I think also that all errors have medin.c on the trace.
It would be very helpful if somebody can spot from the description of my
configuration in the original email where I have gone wrong in the setup of R
with valgrind.
>realloc() does not initialize memory. str.c contains a comment about
>replacing malloc() with calloc() to avoid…
// The data is not really left uninitialised and there may be a
better way to let the compiler know.
//
s->data = (char *) calloc( size, sizeof( *(s->data) ) );
if ( !s->data ) {
error("Error. Cannot allocate memory in str_initalloc, requested
%lu characters.\n\n", size );
// error("\n"); // error( EXIT_FAILURE );
}
s->data[0]='\0';
s->dim=size;
s->len=0;
My comment is indeed sloppy but the first byte is initialised to zero and the
rest is used for writing only
(valgrind has no way to know, of course, and it is fair to question how a human
can possibly be sure).
Thanks again,
Georgi Boshnakov
From: Bill Dunlap <mailto:williamwdun...@gmail.com>
Sent: 02 August 2021 19:48
To: Georgi Boshnakov <mailto:georgi.boshna...@manchester.ac.uk>
Cc: mailto:r-package-devel@r-project.org
Subject: Re: [R-pkg-devel] can't reproduce 'Additional issues' on CRAN with
valgrind
I ran the tests of rbibutils-2.2.2, using the latest devel version of R
configured to use valgrind, with
R --debugger=valgrind --debugger-args=--track-origins=yes --quiet -e
'testthat::test_package("rbibutils")'
I saw a lot of 'conditional jump depends on uninitialized value' errors:
==27280== Conditional jump or move depends on uninitialised value(s)
==27280== at 0x11C420B7: UnknownInlinedFun (xml.c:178)
==27280== by 0x11C420B7: xml_parse (xml.c:241)
==27280== by 0x11C514EF: medin_processf (medin.c:683)
==27280== by 0x11C67ADE: UnknownInlinedFun (bibcore.c:479)
==27280== by 0x11C67ADE: bibl_read.part.0 (bibcore.c:862)
==27280== by 0x11C68C66: bibprog (bibprog.c:36)
==27280== by 0x11C69327: any2xml_main (any2xml.c:127)
==27280== by 0x307639: do_dotCode (dotcode.c:1814)
==27280== by 0x2B4600: bcEval.lto_priv.0 (eval.c:7128)
==27280== by 0x2DADB7: Rf_eval (eval.c:740)
==27280== by 0x2DD28E: R_execClosure.lto_priv.0 (eval.c:1910)
==27280== by 0x2DE1EC: Rf_applyClosure (eval.c:1836)
==27280== by 0x2DAFD3: Rf_eval (eval.c:863)
==27280== by 0x2D7F22: do_begin (eval.c:2530)
==27280== Uninitialised value was created by a heap allocation
==27280== at 0x483DFAF: realloc (in
/usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so)
==27280== by 0x11C38DB2: str_realloc.part.0 (str.c:90)
==27280== by 0x11C3B63D: UnknownInlinedFun (str.c:85)
==27280== by 0x11C3B63D: UnknownInlinedFun (str.c:543)
==27280== by 0x11C3B63D: UnknownInlinedFun (str.c:551)
==27280== by 0x11C3B63D: UnknownInlinedFun (str.c:547)
==27280== by 0x11C3B63D: UnknownInlinedFun (str.c:607)
==27280== by 0x11C3B63D: str_segcpy (str.c:585)
==27280== by 0x11C4DDFA: medin_readf (medin.c:144)
==27280== by 0x11C67A96: UnknownInlinedFun (bibcore.c:471)
==27280== by 0x11C67A96: bibl_read.part.0 (bibcore.c:862)
==27280== by 0x11C68C66: bibprog (bibprog.c:36)
==27280== by 0x11C69327: any2xml_main (any2xml.c:127)
==27280== by 0x307639: do_dotCode (dotcode.c:1814)
realloc() does not initialize memory. str.c contains a comment about replacing
malloc() with calloc() to avoid similar problems that valgrind found. It
states that valgrind is mistaken, that that memory really is initialized. Hmm.
I also see the error about reading off the end of a malloc'ed array:
==27280== Invalid read of size 1
==27280== at 0x11C420A8: UnknownInlinedFun (xml.c:174)
==27280== by 0x11C420A8: xml_parse (xml.c:241)
==27280== by 0x11C514EF: medin_processf (medin.c:683)
==27280== by 0x11C67ADE: UnknownInlinedFun (bibcore.c:479)
==27280== by 0x11C67ADE: bibl_read.part.0 (bibcore.c:862)
==27280== by 0x11C68C66: bibprog (bibprog.c:36)
==27280== by 0x11C69327: any2xml_main (any2xml.c:127)
==27280== by 0x307639: do_dotCode (dotcode.c:1814)
==27280== by 0x2B4600: bcEval.lto_priv.0 (eval.c:7128)
==27280== by 0x2DADB7: Rf_eval (eval.c:740)
==27280== by 0x2DD28E: R_execClosure.lto_priv.0 (eval.c:1910)
==27280== by 0x2DE1EC: Rf_applyClosure (eval.c:1836)
==27280== by 0x2DAFD3: Rf_eval (eval.c:863)
==27280== by 0x2D7F22: do_begin (eval.c:2530)
==27280== Address 0x125028af is 0 bytes after a block of size 10,927 alloc'd
==27280== at 0x483DD99: calloc (in
/usr/lib/x86_64-linux-gnu/valgrind/vgpreload_memcheck-amd64-linux.so)
==27280== by 0x11C38DF2: str_initalloc (str.c:253)
==27280== by 0x11C3B66A: UnknownInlinedFun (str.c:541)
==27280== by 0x11C3B66A: UnknownInlinedFun (str.c:551)
==27280== by 0x11C3B66A: UnknownInlinedFun (str.c:547)
==27280== by 0x11C3B66A: UnknownInlinedFun (str.c:607)
==27280== by 0x11C3B66A: str_segcpy (str.c:585)
==27280== by 0x11C4DDFA: medin_readf (medin.c:144)
On Mon, Aug 2, 2021 at 10:11 AM Georgi Boshnakov
<mailto:georgi.boshna...@manchester.ac.uk> wrote:
An update to my package rbibutils triggered 'Additional issues' on CRAN from
valgrind, CRAN Package Check Results for Package rbibutils
(http://r-project.org)<https://cran.r-project.org/web/checks/check_results_rbibutils.html>.
However, running the checks with -use-valgrind on my machine gives zero errors.
I am on Ubuntu (the latest stable version), my R-devel installation was updated
earlier today and I rebuild it adding the
--with-valgrind-instrumentation=2 to the call to configure.
Valgrind coming with Ubuntu is 3.5.0, so I installed its latest version from
github (also no errors).
rbibutils has no dependencies, except that testthat is used for testing.
What am I missing? Do I need to have a separate library for the instrumented
R-devel version?
As a side note, is there a way to find out if R has been built with valgrind?
It seems that 'capabilities()' and 'R CMD config' do not offer this information.
Thanks,
Georgi Boshnakov
[[alternative HTML version deleted]]
______________________________________________
mailto:R-package-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-package-devel
______________________________________________
R-package-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-package-devel
