magic number or versioning, relied on random(3) never changing to a
different implementation, and were also saving pointers to disk and
reading them back again. It *looks* as if the pointers thus loaded
were reset before being used, but it's not particularly clear as the
main loop of this thing is goto-based FORTRAN translated lightly to C.
I've changed the logic to null these pointers instead of saving and
loading them, and things seem to still work.
The new save files have a header, support versioning, write only sized
types in network byte order, and for the toy encryption to discourage
cheating do something self-contained instead of using random(3) as a
stream cipher.
Because between the original import from 4.4 until earlier today
trying to save would result in SIGSEGV on most platforms, it's
unlikely anyone has a save file, but just in case (since the pointer
issue appears to be nonlethal) I've kept compat code for old save
files.
adventure(6). The handling of this variable is somewhat confusing,
since it is used for two different purposes (controlling the time
required before a saved game can be restored, and controlling various
aspects of dwarf behaviour); in fact, it is also declared twice in
hdr.h. Except possibly when saving a game fails, these uses can never
interfere; when used for controlling dwarf behaviour, we always have
saved == -1. This can be better understood with reference to the
original PDP-10 FORTRAN source (URL in patch, since hdr.h references
the comments of the FORTRAN as still relevant to this version) of
which the C version is a direct translation: the wrong value for
`saved' meant that someone was cheating and had bypassed normal
initialisation. Saving was done by halting and telling the user to
save their core image, so the question of carrying on after saving
failed to open the output file did not arise.
This patch separates the uses of `saved' into uses of two separate
variables.
Closes PR 8005 by Joseph Myers <jsm28@cam.ac.uk>
