GitLab

Without this change, on typical 64-bit hosts error ("...%d...", N)
was used to print both 32- and 64-bit integers N, which relied on
undefined behavior.
* lisp.h, src/m/amdx86-64.h, src/m/ia64.h, src/m/ibms390x.h (pEd):
New macro.
* lisp.h (error, verror): Mark as printf-like functions.
* eval.c (verror): Use vsnprintf, not doprnt, to do the real work.
Report overflow in size calculations when allocating printf buffer.
Do not truncate output string at its first null byte.
* xdisp.c (vmessage): Use vsnprintf, not doprnt, to do the real work.
Truncate the output at a character boundary, since vsnprintf does not
do that.
* charset.c (check_iso_charset_parameter): Convert internal
character to string before calling 'error', since %c now has the
printf meaning.
* coding.c (Fdecode_sjis_char, Fdecode_big5_char): Avoid int
overflow when computing char to be passed to 'error'.  Do not
pass Lisp_Object to 'error'; pass the integer instead.
* nsfns.m (Fns_do_applescript): Use int, not long, since it's
formatted with plain %d.