Bruno Haible <br...@clisp.org> writes: >> The culprit is pretty clearly this code in lib/vasnprintf.c: >> >> case 's': >> [...] >> tmp_length = strlen (a.arg[dp->arg_index].a.a_string); >> break; > > How did you find this? I thought that valgrind only runs on modern > platforms, which all have snprintf.
I don't know. Somehow the GNU PSPP current Git tree selected this case when I configured it on a modern GNU/Linux system, and so it came up when I was running a test under valgrind. But it only did it after I had done some manual editing of Makefiles, etc. produced by configure, and much source hacking. And then I was unable to reproduce it with "gnulib-tool --create-testdir". So to make sure that it could really happen in a simple test case, I forced !USE_SNPRINTF by editing vasnprintf.c manually. The one interesting thing I did to the Makefiles was to set CFLAGS (not AM_CFLAGS) manually, to a value that used -O0 instead of -O2 (so that debugging with GDB would be easier). I don't know how this could cause such a problem, though. Thank you for the fix! One oddity of the fixed version (at least as it appeared in my email client) is that the comments include a number of usages of the ligature for "fi", in the word "specified", instead of separate letters "f" and "i". I guess that you must have cut and pasted from an ISO document that included those ligatures. -- "Because computer source code is an expressive means for the exchange of information and ideas about computer programming, we hold that it is protected by the First Amendment." --Hon. Boyce F. Martin, Jr., for the 6th Circuit Court, Junger vs. Daley
