+2011-07-28 Paul Eggert <eggert@cs.ucla.edu>
+
+ Integer signedness and overflow and related fixes. (Bug#9079)
+
+ * bidi.c: Integer size and overflow fixes.
+ (bidi_cache_size, bidi_cache_idx, bidi_cache_last_idx)
+ (bidi_cache_start, bidi_cache_fetch_state, bidi_cache_search)
+ (bidi_cache_find_level_change, bidi_cache_ensure_space)
+ (bidi_cache_iterator_state, bidi_cache_find, bidi_cache_start_stack)
+ (bidi_find_other_level_edge):
+ Use ptrdiff_t instead of EMACS_INT where either will do.
+ This works better on 32-bit hosts configured --with-wide-int.
+ (bidi_cache_ensure_space): Check for size-calculation overflow.
+ Use % rather than repeated addition, for better worst-case speed.
+ Don't set bidi_cache_size until after xrealloc returns, because it
+ might not return.
+ (bidi_dump_cached_states): Use ptrdiff_t, not int, to avoid overflow.
+ (bidi_cache_ensure_space): Also check that the bidi cache size
+ does not exceed that of the largest Lisp string or buffer. See Eli
+ Zaretskii in <http://debbugs.gnu.org/cgi/bugreport.cgi?bug=9079#29>.
+
+ * alloc.c (__malloc_size_t): Remove.
+ All uses replaced by size_t. See Andreas Schwab's note
+ <http://debbugs.gnu.org/cgi/bugreport.cgi?bug=9079#8>.
+
+ * image.c: Improve checking for integer overflow.
+ (check_image_size): Assume that f is nonnull, since
+ it is always nonnull in practice. This is one less thing to
+ worry about when checking for integer overflow later.
+ (x_check_image_size): New function, which checks for integer
+ overflow issues inside X.
+ (x_create_x_image_and_pixmap, xbm_read_bitmap_data): Use it.
+ This removes the need for a memory_full check.
+ (xbm_image_p): Rewrite to avoid integer multiplication overflow.
+ (Create_Pixmap_From_Bitmap_Data, xbm_load): Use x_check_image_size.
+ (xbm_read_bitmap_data): Change locals back to 'int', since
+ their values must fit in 'int'.
+ (xpm_load_image, png_load, tiff_load):
+ Invoke x_create_x_image_and_pixmap earlier,
+ to avoid much needless work if the image is too large.
+ (tiff_load): Treat overly large images as if
+ x_create_x_image_and_pixmap failed, not as malloc failures.
+ (gs_load): Use x_check_image_size.
+
+ * gtkutil.c: Omit integer casts.
+ (xg_get_pixbuf_from_pixmap): Remove unnecessary cast.
+ (xg_set_toolkit_scroll_bar_thumb): Rewrite to avoid need for cast.
+
+ * image.c (png_load): Don't assume height * row_bytes fits in 'int'.
+
+ * xfaces.c (Fbitmap_spec_p): Fix integer overflow bug.
+ Without this fix, (bitmap-spec-p '(34359738368 1 "x"))
+ would wrongly return t on a 64-bit host.
+
+ * dispnew.c (init_display): Use *_RANGE_OVERFLOW macros.
+ The plain *_OVERFLOW macros run afoul of GCC bug 49705
+ <http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49705>
+ and therefore cause GCC to emit a bogus diagnostic in some cases.
+
+ * image.c: Integer signedness and overflow and related fixes.
+ This is not an exhaustive set of fixes, but it's time to
+ record what I've got.
+ (lookup_pixel_color, check_image_size): Remove redundant decls.
+ (check_image_size): Don't assume that arbitrary EMACS_INT values
+ fit in 'int', or that arbitrary 'double' values fit in 'int'.
+ (x_alloc_image_color, x_create_x_image_and_pixmap, png_load)
+ (tiff_load, imagemagick_load_image):
+ Check for overflow in size calculations.
+ (x_create_x_image_and_pixmap): Remove unnecessary test for
+ xmalloc returning NULL; that can't happen.
+ (xbm_read_bitmap_data): Don't assume sizes fit into 'int'.
+ (xpm_color_bucket): Use better integer hashing function.
+ (xpm_cache_color): Don't possibly over-allocate memory.
+ (struct png_memory_storage, tiff_memory_source, tiff_seek_in_memory)
+ (gif_memory_source):
+ Use ptrdiff_t, not int or size_t, to record sizes.
+ (png_load): Don't assume values greater than 2**31 fit in 'int'.
+ (our_stdio_fill_input_buffer): Prefer ptrdiff_t to size_t when
+ either works, as we prefer signed integers.
+ (tiff_read_from_memory, tiff_write_from_memory):
+ Return tsize_t, not size_t, since that's what the TIFF API wants.
+ (tiff_read_from_memory): Don't fail simply because the read would
+ go past EOF; instead, return a short read.
+ (tiff_load): Omit no-longer-needed casts.
+ (Fimagemagick_types): Don't assume size fits into 'int'.
+
+ Improve hashing quality when configured --with-wide-int.
+ * fns.c (hash_string): New function, taken from sxhash_string.
+ Do not discard information about ASCII character case; this
+ discarding is no longer needed.
+ (sxhash-string): Use it. Change sig to match it. Caller changed.
+ * lisp.h: Declare it.
+ * lread.c (hash_string): Remove, since we now use fns.c's version.
+ The fns.c version returns a wider integer if --with-wide-int is
+ specified, so this should help the quality of the hashing a bit.
+
+ * emacs.c: Integer overflow minor fix.
+ (heap_bss_diff): Now uprintmax_t, not unsigned long. All used changed.
+ Define only if GNU_LINUX.
+ (main, Fdump_emacs): Set and use heap_bss_diff only if GNU_LINUX.
+
+ * dispnew.c: Integer signedness and overflow fixes.
+ Remove unnecessary forward decls, that were a maintenance hassle.
+ (history_tick): Now uprintmax_t, so it's more likely to avoid overflow.
+ All uses changed.
+ (adjust_glyph_matrix, realloc_glyph_pool, adjust_frame_message_buffer)
+ (scrolling_window): Use ptrdiff_t, not int, for byte count.
+ (prepare_desired_row, line_draw_cost):
+ Use int, not unsigned, where either works.
+ (save_current_matrix, restore_current_matrix):
+ Use ptrdiff_t, not size_t, where either works.
+ (init_display): Check for overflow more accurately, and without
+ relying on undefined behavior.
+
+ * editfns.c (pWIDE, pWIDElen, signed_wide, unsigned_wide):
+ Remove, replacing with the new symbols in lisp.h. All uses changed.
+ * fileio.c (make_temp_name):
+ * filelock.c (lock_file_1, lock_file):
+ * xdisp.c (message_dolog):
+ Don't assume PRIdMAX etc. works; this isn't portable to pre-C99 hosts.
+ Use pMd etc. instead.
+ * lisp.h (printmax_t, uprintmax_t, pMd, pMu): New types and macros,
+ replacing the pWIDE etc. symbols removed from editfns.c.
+
+ * keyboard.h (num_input_events): Now uintmax_t.
+ This is (very slightly) less likely to mess up due to wraparound.
+ All uses changed.
+
+ * buffer.c: Integer signedness fixes.
+ (alloc_buffer_text, enlarge_buffer_text):
+ Use ptrdiff_t rather than size_t when either will do, as we prefer
+ signed integers.
+
+ * alloc.c: Integer signedness and overflow fixes.
+ Do not impose an arbitrary 32-bit limit on malloc sizes when debugging.
+ (__malloc_size_t): Default to size_t, not to int.
+ (pure_size, pure_bytes_used_before_overflow, stack_copy_size)
+ (Fgarbage_collect, mark_object_loop_halt, mark_object):
+ Prefer ptrdiff_t to size_t when either would do, as we prefer
+ signed integers.
+ (XMALLOC_OVERRUN_CHECK_OVERHEAD): New macro.
+ (xmalloc_overrun_check_header, xmalloc_overrun_check_trailer):
+ Now const. Initialize with values that are in range even if char
+ is signed.
+ (XMALLOC_PUT_SIZE, XMALLOC_GET_SIZE): Remove, replacing with ...
+ (xmalloc_put_size, xmalloc_get_size): New functions. All uses changed.
+ These functions do the right thing with sizes > 2**32.
+ (check_depth): Now ptrdiff_t, not int.
+ (overrun_check_malloc, overrun_check_realloc, overrun_check_free):
+ Adjust to new way of storing sizes. Check for size overflow bugs
+ in rest of code.
+ (STRING_BYTES_MAX): Adjust to new overheads. The old code was
+ slightly wrong anyway, as it missed one instance of
+ XMALLOC_OVERRUN_CHECK_OVERHEAD.
+ (refill_memory_reserve): Omit needless cast to size_t.
+ (mark_object_loop_halt): Mark as externally visible.
+
+ * xselect.c: Integer signedness and overflow fixes.
+ (Fx_register_dnd_atom, x_handle_dnd_message):
+ Use ptrdiff_t, not size_t, since we prefer signed.
+ (Fx_register_dnd_atom): Check for ptrdiff_t (and size_t) overflow.
+ * xterm.h (struct x_display_info): Use ptrdiff_t, not size_t, for
+ x_dnd_atoms_size and x_dnd_atoms_length.
+
+ * doprnt.c: Prefer signed to unsigned when either works.
+ * eval.c (verror):
+ * doprnt.c (doprnt):
+ * lisp.h (doprnt):
+ * xdisp.c (vmessage):
+ Use ptrdiff_t, not size_t, when using or implementing doprnt,
+ since the sizes cannot exceed ptrdiff_t bounds anyway, and we
+ prefer signed arithmetic to avoid comparison confusion.
+ * doprnt.c (doprnt): Avoid a "+ 1" that can't overflow,
+ but is a bit tricky.
+
+ Assume freestanding C89 headers, string.h, stdlib.h.
+ * data.c, doprnt.c, floatfns.c, print.c:
+ Include float.h unconditionally.
+ * gmalloc.c: Assume C89-at-least behavior for preprocessor,
+ limits.h, stddef.h, string.h. Use memset instead of 'flood'.
+ * regex.c: Likewise for stddef.h, string.h.
+ (ISASCII): Remove; can assume it returns 1 now. All uses removed.
+ * s/aix4-2.h (HAVE_STRING_H): Remove obsolete undef.
+ * s/ms-w32.h (HAVE_LIMITS_H, HAVE_STRING_H, HAVE_STDLIB_H)
+ (STDC_HEADERS): Remove obsolete defines.
+ * sysdep.c: Include limits.h unconditionally.
+
+ Assume support for memcmp, memcpy, memmove, memset.
+ * lisp.h, sysdep.c (memcmp, memcpy, memmove, memset):
+ * regex.c (memcmp, memcpy):
+ Remove; we assume C89 now.
+
+ * gmalloc.c (memcpy, memset, memmove): Remove; we assume C89 now.
+ (__malloc_safe_bcopy): Remove; no longer needed.
+
+ * lisp.h (struct vectorlike_header, struct Lisp_Subr): Signed sizes.
+ Use EMACS_INT, not EMACS_UINT, for sizes. The code works equally
+ well either way, and we prefer signed to unsigned.
+
2011-07-27 Lars Magne Ingebrigtsen <larsi@gnus.org>
* gnutls.c (emacs_gnutls_read): Don't message anything if the peer
HCoop / bpt / emacs.git / blobdiff