Consistently consider a nation bankrupt when its treasury is
negative. Some places considered $0 as bankrupt, some didn't. Fix
the ones that did:
(repo_list): report command misreported countries with $0 as broke.
(init_nats): If you had $0, logging out and back in bankrupted you.
(produce_sect, upd_ship): Failed to build sectors and produce stuff
for countries with $0.
(status): Round money before checking solvency, and consistently treat
$0 as solvent. Before, you became bankrupt / solvent when your money
went below / above zero before rounding, but not when you hit zero
exactly.
(check_trade): Fix bug that made your money evaporate when you didn't
have enough to pay. This was aggravated by the removal of automatic
loans in rev. 1.15. Clean up after that revision, belatedly.
Remove claim that ships and units on missions will not fire support.
It's always been wrong for ships, and land units were changed to match
ships in lndsub.c rev. 1.49.
Reject unexpected redirections:
(redir_authorized): New parameter expected, reject when zero.
(doredir): Pass !redir_fp, do not close it. Before, an unexpected
redirection silently replaced the existing one.
(dopipe): Pass !redir_fp. Before, an unexpected redirection silently
replaced the existing one, leaking its FILE.
(doexecute): Pass 1.
Refactor redirection data structure, no functional change:
(redir_is_pipe): New.
(doredir, dopipe): Set it.
(pipe_fp): Remove. Users changed to use redir_fp and redir_is_pipe.
Flaws in the Empire protocol make redirections within execute next to
impossible to implement correctly, and nested execute practically
useless (#116377). Catch and refuse those:
(executing): New.
(doexecute, prompt): Set and clear it.
(redir_authorized): Reject if set.
(doexecute, play): Don't signal doexecute() failure through input_fd,
because that screws up up when we're executing already. Increment
send_eof in doexecute() instead.
(send_eof): External linkage.
(play): Increment send_eof instead of setting it, decrement it instead
of clearing it. Fixes a race condition: when we got EOF on input_fd
and a failed doexecute() in the same iteration of the loop, only one
EOF cookie was sent, leaving the client hung.
(flash, wall): Use player->comtail[] to find the raw message. The old
hack to find worked only when there were no spaces in flash's first
argument. Closes #941740.
(execute): Use player->comtail[1] instead of player->argp[1]. This
makes execute send everything after the command back to the client
verbatim, not just the first argument (with quotes and funny
characters stripped). The client always expected that, and got
confused when the server sent something else.
Record raw arguments, to be used in the next changesets:
(player): New member comtail.
(parse): New parameter tail. Reorder parameter list.
(command, execute): Pass player->comtail.
(player_login, emp_config, do_unit_move): Pass NULL. No functional
change.
(getilist): Use pln_airbase_ok(). This fixes a number of bugs: planes
stuck in foreign bases, on inefficient ships or land units, or land
units loaded on ships or land units could still intercept.
(can_fly): Remove.
(getilist): Remove bogus fuel check. It was inconsistent with the one
in mission_pln_equip(), which is called from ac_intercept(). The
inconsistency appeared in Empire3, and had no ill effect then. But
rev. 1.42 screwed up the fuel test here: it made interceptors that
don't use fuel (SAMs) require petrol to be present to fly.
(carrier_planes): New parameter msl.
(carriersatxy, pln_onewaymission): Pass zero. No functional change.
(pln_airbase_ok): Pass whether the plane is a missile. This fixes
non-x-light missiles on ships without capability M_FLY. Broken in
rev. 1.76. Launch and interdict weren't affected. Air defense was,
but the stock game has only x-light SAMs.
Ron Koenderink [Thu, 22 Nov 2007 12:57:00 +0000 (12:57 +0000)]
(nav_ship): Move the special autofeed code to a separate
function nav_load_ship_at_sea().
(nav_load_ship_at_sea): New code. Generalize the autofeed to
support ships with M_OIL.
(play): Continue after input read error, treating it like EOF. This
fixes error handling when a script can be opened but not read: it
switches back to standard input. And it handles errors on standard
input more gracefully: orderly shutdown, buffered output is still
written out.
Fix redirections with execute. Redirection consumed the remembered
input, and execute couldn't find it and mistakenly raised the
tampering deity alarm. Closes #804644:
(saved_bytes): New.
(save_input): Set it.
(forget_input): New.
(seen_input): Don't discard, return a value for forget_input().
(save_input): Return a value for forget_input(), just because it makes
sense.
(input_to_forget, redir_authorized, prompt): Save value of
seen_input() to new input_to_forget in redir_authorized(), pass it to
forget_input() in prompt().
Rewrite much of client's playing phase code:
(EOF_COOKIE, INTR_COOKIE, input_fd, send_intr, recv_output)
(recv_input, intr, play): New playing phase code. No native Windows
support yet. Sends just one EOF cookie on EOF on standard input
instead of up to three. Old servers (before recvclient.c rev. 1.16)
fail to terminate the session when they receive an EOF cookie at an
argument prompt. The session then hangs; use SIGINT to get out. No
longer blocks on sending input, which could deadlock the session.
Closes #827090. Fixes error handling for select(). Fixes race
condition that could cause server output to be discarded on EOF on
standard input.
(main): Replace old playing phase code by a call to play().
(intr, sock, interrupt, handleintr): Replaced by play(), remove.
[_WIN32] (hStdIn): Ditto.
(auxfp): New.
(servercmd, prompt, doexecute): Remove parameter auxfi, use auxfp.
(eight_bit_clean): Move to servcmd.c.
(servercmd): Work on a single non-C_DATA line instead of getting lines
from an ioqueue.
(servercmd, output, screen, outch): Deal with all ids in servercmd()
rather than some there and some in output(). Don't treat C_NOECHO,
C_ABORT, C_CMDERR, C_BADCMD specially. Fix C_FLASH and C_EXIT to
ignore redirections; they used to ignore them only for some parts.
Replace output() by outch(), fold screen into outch().
(servercmd): Truncate long prompts and telegram infos to prevent
buffer overflow.
(prompt): Use new parameters code, prompt, teles instead of global
variables mode, the_prompt, num_teles.
(num_teles, the_prompt, mode, nbtu, nmin): Remove.
(prompt): Don't write an empty line before argument prompts to auxfp.
(servercmd): Don't strip newline from redirections and execute,
doredir(), dopipe() and doexecute() need it now.
(doredir, dopipe, doexecute): Use new seen_input() instead of gettag().
(doexecute): Set input_fd and leave reading the script file to play().
(serverio, termio, sendeof): Replaced by play(), remove.
(LBUF_LEN_MAX, lbuf, lbuf_init, lbuf_len, lbuf_full, lbuf_line)
(lbuf_putc): New.
(RING_SIZE, ring, ring_init, ring_len, ring_space, ring_peek)
(ring_getc, ring_putc, ring_putm, ring_discard, ring_search)
(ring_from_file, ring_to_file): New.
(clear_recent_input, save_input, seen_input): New.
(MAX): New.
(ioqueue, io, ioq_init, ioq_dequeue, ioq_read, ioq_write, ioq_qsize)
(ioq_drain, ioq_gets, ioqtobuf, enqueuecc, dequeuecc): Unused, remove.
(QEMPTY, qelem, insque, remque, initque): Unused, remove.
(tagstruct, taglist, io_init, gettag): Unused, remove.
(main) [_WIN32]: Don't call WSAStartup() until after option
processing, so that we can version and help even when it fails. Fix
program exit status on failure.
Belatedly update standalone build for last night's changeset:
(main): Rewrite argument parsing to use getopt(). New options -h and
-v.
(print_usage): New.
(saveargv): Lame attempt at protecting users who foolishly specify
passwords on the command line. Antisocially zaps the complete command
line, not just the password. Remove, at least for now.
(main): Update accordingly.
(load_plane_ship, load_land_ship, load_plane_land, load_land_land):
isdigit() can return anything, not just 0 or 1, and you can't combine
its values with &=! The bug could make loading operations fail
noisily instead of silently, depending on the system's implementation
of isdigit().
Shut down session on receipt of "ctld\n" even when reading arguments:
(player): New member eof.
(recvclient): Return -1 without receiving input when it is set. Set
it on receipt of "ctld\n".
(execute): Clear it after receiving the script.
(execute): Prompted for a missing argument, but then passed a null
pointer to prexec() instead. Some systems (GNU, Windows) deal
gracefully with printing null strings, others crash.
Ron Koenderink [Mon, 12 Nov 2007 20:08:15 +0000 (20:08 +0000)]
Fix the test scripts for the players for the new world -R 1 for fairland
and for -R 1 for the server. Cleanup the shell selection for the current builds
and the push the shell fixes to the config files. Add PATCH and
CVSROOT environment variables to add flexibility. Add the changes for
the new autoconf build and remove build.conf and makedepend.
Swiitch All patch to use the source code All patches.
Fix the spelling of sandbox.
(carriersatxy): Remove parameters wantflags, nowantflags, and use
carrier_planes() instead, ignoring carriers that can only operate
x-light planes. No functional change.
(carrier_planes): External linkage.
(carrier_planes): New.
(pln_onewaymission): Use it. No functional change.
(pln_airbase_ok): Use it to enforce capabilities. Before, a plane
could fly once it got on the ship. Because ships can load only planes
that can fly from them, this has no effect other than make that sanity
check redundant; remove it.
(pln_airbase_ok): Update land unit case to keep it similar to the ship
case.
(pln_airbase_ok): New parameter noisy to allow suppressing messages.
Check player->cnum only when noisy. Update existing caller to pass 1.
No functional change.
(mission_pln_airbase_ok): Remove, use pln_airbase_ok() instead. No
functional change.
(pln_airbase_ok): Plug information leak: carrier owner was checked
last, so when it was (no longer) allied, attempting to fly off it let
you test the conditions checked before that. Check owner first.
(mission_pln_airbase_ok): Matching change, to avoid diverging from
pln_airbase_ok().
(pln_airbase_ok, mission_pln_airbase_ok): Oops when plane's carrier is
bad. Before, the plane was destroyed, and the player got a message.
(pln_airbase_ok, mission_pln_airbase_ok): Oops when plane's base
sector is bad.
(pln_airbase_ok): Oops when the player doesn't own the plane. Before,
this was checked only for planes on carriers, and the plane was
destroyed, and the player got a misleading message.
(mission_pln_wanted): Broken like its cousin pln_wanted() was, only
more so: additionally, any of P_E, P_K, P_L satisfied any subset of
P_E, P_K, P_L. Chainsaw fixed this one in plnsub.c, but not here.
Remove, use the fixed pln_wanted() instead.
(pln_wanted): External linkage.
(pln_wanted): Rewrite. Old version was incomprehensible, slow and
buggy: any P_E, P_L, P_K in wantflags were ignored when the plane
lacked P_ESC. The bug bit only when non-escort interceptors escorted
a one-way mission to a carrier. pln_oneway_to_carrier_ok() then could
not fit the plane on the carrier, and the command failed complaining
about lack of room on the carrier. Broken since Chainsaw added
escorts, abusable before 4.2.17 catched it. Also change behavior when
only one of P_F and P_ESC is in wantflags: new version requires that
flag, while old version treats it as if both flags were in wantflags:
a plane having either is okay. Current code never passes such
wantflags.
(fit_plane_on_ship, fit_plane_off_ship): Change how planes are fit on
ships, so that order of loading no longer matters: choppers can use
chopper or plane slots, x-lights can use x-light or plane slots, light
planes can use plane slots. However, plane slots require M_FLY or
M_MSL to hold missiles or x-lights, and M_FLY to hold anything else.
Choppers and x-lights can now use plane slots even when not light.
X-light choppers, which don't exist in the stock game, can no longer
use x-light slots.
(fit_plane_on_ship, fit_plane_off_ship): Don't require M_CHOPPER for
use of chopper slot. Same for M_XLIGHT and x-light slot. No
functional change, because presence of slots implies the flags (see
init_mchr()).
Collect the test whether a plane fits on a carrier in one place:
(fit_plane_on_ship, fit_plane_on_land): New.
(pln_oneway_to_carrier_ok, put_plane_on_ship, count_planes)
(put_plane_on_land, count_land_planes): Use them. No functional
change.
(fit_plane_off_ship, fit_plane_off_land): New.
(take_plane_off_ship, take_plane_off_land): Use them. This oopses
when the carriers plane counter underflows.
(take_plane_off_ship): Document that carrier's plane counters may be
screwed up.
(take_plane_off_ship, take_plane_off_land): Oops when the plane isn't
on the carrier.
(take_plane_off_ship, take_plane_off_land): Don't fail when the plane
to be taken off can't go on this type of carrier, just take it off.
No error condition left, so return void. Callers couldn't do anything
useful with the status anyway, and most didn't bother. Change those
that did.