The save_file() function in save.c stored the savefile's device number,
inode number, creation time, and modification time in the file. The
restore() function read them back, and apparently used to compare them
to protect against cheaters.
Unfortunately, the types and sizes of these numbers differ from system
to system, which ruins the Roguelike Restoration Project's fine
portability work. So they have been removed from the savefile.
This BREAKS SAVEFILE COMPATIBILITY, but old files can be converted by
excising the chunk starting at offset 0x22 with length sizeof(ino_t) +
sizeof(dev_t) + 2 * sizeof(time_t). That's 0x14 on i686 and 0x20 on
x86_64, at least with current versions of Linux and glibc.
The spell-choosing and prayer-choosing routines, when the one-line
inventory option is set, displayed to cw instead of msgw. This caused
permanent corruption of the message line.
Some sections of code that prompt the user for a string of input were
calling get_str() with cw (the player-visible screen containing the
map), which caused whatever the player typed to get printed starting at
cw's idea of the cursor position, which was usually the Rogue's @-sign.
This corrupted the map.
The problem has been fixed by passing msgw (the message line at the top
of the screen) to get_str(), so the player's typing appears where msgw
thinks the cursor should be, which is in the sensible place right after
the prompt. Some other get_str() invocations which used hw or stdscr
have been left unmodified.
A buffer called curpurch, which stores a description of an item in a
trading post which the player might be interested in, was only 15
bytes. It was overflowing into oldrp, a room pointer, leading to
segfaults. The size of curpurch has been increased to LINELEN*2,
which matches the size of prbuf, which is returned by inv_name and
then strcpy()'d to curpurch. As long as nothing overflows prbuf it
should be safe now.
NOTE that this breaks savefile compatibility.