Without this fix, the command to link temacs failed due to an
undefined symbol __builtin_isnan. This is because
/usr/include/iso/math_c99.h #defines isnan(x) to
__builtin_isnan(x), but the bundled gcc, which identifies itself
as gcc 3.4.3 (csl-sol210-3_4-branch+sol_rpath), does not have
a __builtin_isnan.
* configure.in (isnan): Remove now-unnecessary check.
* src/floatfns.c (isnan): #undef, and then #define to a clone of
what's in data.c.
(Fisnan): Always define, since it's always available now.
(syms_of_floatfns): Always define isnan at the Lisp level.
(window_scroll_line_based, Frecenter):
Check that an Emacs fixnum is in range before assigning it to 'int'.
(Frecenter, Fmove_to_window_line): Use EMACS_INT, not int, for
values converted from Emacs fixnums.
(Frecenter): Don't wrap around a line count if it is out of 'int'
range; instead, treat it as an extreme value.
(Fset_window_configuration, compare_window_configurations):
Use ptrdiff_t, not int, for index that might exceed 2 GiB.
(Freplace_match): Use ptrdiff_t, not int, for indexes that can
exceed INT_MAX. Check that EMACS_INT value is in range before
assigning it to the (possibly-narrower) index.
(match_limit): Don't assume that a fixnum can fit in 'int'.
(position_indentation): Now takes ptrdiff_t, not int.
(Fvertical_motion): Don't wrap around LINES values that don't fit
in 'int'. Instead, treat them as extreme values. This is good
enough for windows, which can't have more than INT_MAX lines anyway.
* xml.c (parse_region): Don't call xmlCleanupParser after parsing,
since this reportedly can destroy thread storage.
* emacs.c (shut_down_emacs): Call xmlCleanupParser on shutdown.
The previous code assumed that file offsets (off_t values) fit in
EMACS_INT variables, which is not true on typical 32-bit hosts.
The code messed up by falsely reporting buffer overflow in cases
such as (insert-file-contents "big" nil 1 2) into an empty buffer
when "big" contains more than 2**29 bytes, even though this
inserts just one byte and does not overflow the buffer.
(Finsert_file_contents): Store file offsets as off_t
values, not as EMACS_INT values. Check for overflow when
converting between EMACS_INT and off_t. When checking for
buffer overflow or for overlap, take the offsets into account.
Don't use EMACS_INT for small values where int suffices.
When checking for overlap, fix a typo: ZV was used where
ZV_BYTE was intended.
(Fwrite_region): Don't assume off_t fits into 'long'.
* buffer.h (struct buffer.modtime_size): Now off_t, not EMACS_INT.
* font.c (APPEND_SNPRINTF): Remove.
(font_unparse_xlfd):
* xterm.c (x_io_error_quitter):
Use esnprintf, not snprintf. That way, we don't have to worry
about porting to ancient platforms that lack snprintf.
(x_term_init): Use sprintf, not snprintf.
src/xdisp.c (produce_stretch_glyph): No longer static, compiled also
when HAVE_WINDOW_SYSTEM is not defined. Support both GUI and TTY
frames. Call tty_append_glyph in the TTY case. (Bug#9402)
src/term.c (tty_append_glyph): New function.
(produce_stretch_glyph): Static function and its prototype deleted.
src/dispextern.h (produce_stretch_glyph, tty_append_glyph): Add
prototypes.
(print_object): Make sprintf buffer a bit bigger, to avoid potential
buffer overrun. Don't assume list length fits in 'int'. Treat
print length of 0 as 0, not as infinity; to be consistent with other
uses of print length in this function. Don't overflow print length
index. Don't assume hash table size fits in 'long', or that
vectorlike size fits in 'unsigned long'.
(Fend_kbd_macro): Don't mishandle MOST_NEGATIVE_FIXNUM by treating
it as a large positive number.
(Fexecute_kbd_macro): Don't assume repeat count fits in int.
* macros.h (executing_kbd_macro_iterations): Now EMACS_INT, not int.
(make_terminal_frame, set_term_frame_name): Print it.
(x_report_frame_params): In X, window IDs are unsigned long,
not signed long, so print them as unsigned.
(validate_x_resource_name): Check for implausibly long names,
and don't assume name length fits in 'int'.
(x_get_resource_string): Don't blindly alloca invocation name;
use SAFE_ALLOCA. Use esprintf, not sprintf, in case result does
not fit in int.
Don't assume XINT result fits in int, or that XFLOAT_DATA * 10
fits in int, when using sprintf. Use single snprintf to count
length of string rather than counting it via multiple sprintfs;
that's simpler and more reliable.
(APPEND_SPRINTF): New macro.
(font_unparse_fcname): Use it to avoid sprintf buffer overrun.
(generate_otf_features) [0 && HAVE_LIBOTF]: Use esprintf, not
sprintf, in case result does not fit in int.
