If a Key-down event didn't meet the expected criteria,
the event was ignored in the "peek" loop, but it was
never removed from the input queue either, so it
remained at the top of the queue.
It isn't clear if this fixes the dedicated arrow keys
hanging, but if those were resulting in strange
event values too, it just might.
The autocompletion was working, but you couldn't tell
because the text was overwritten with blanks. This was
a recent glitch created during fix for other cursor problems.
This brings things much closer to correct operation (I hope).
- The shift to only moving the cursor on input (<Someone>'s
changes) had a lot of complications, among them was
that sometimes, there is no more input. When the program was
exiting, or bombing the cursor synch never got done, so the
final messages could end up strewn any place the cursor
happened to be dwelling.
- There were two competing output systems in use: the
wintty stuff for the game, and the msmsg and error stuff
used by the sys/share/pcsys, sys/share/pctty, and
sys/share/pcunix routines. Those were meant to mimic
output to stdout, where stuff just got sent to a sequential
display. Over time, there were calls mixed in that depended
on the cursor tracked stuff from the core game, so you
really couldn't be sure where things were going to display.
It wasn't as much of an issue before, because the cursor
really did get moved around as expected. Everything
now ends up in the same output system.
- I even found a use of the real putchar() because
sys/share/pcunix didn't #include wintty.h the same
as the other files, and the macro never got defined.
Who knows where that character was being put -
the game certainly couldn't track it.
While everything I knew to be wrong yesterday is
now working, there may be some other glitches
lurking that I haven't discovered yet.
Please: test, test, test.
<Someone> wrote:
> Using the MCVS IDE I couldn't compile NetHack any longer, due to a
> misplaced #endif and a library not included in the DSP file.
> Here's a patch for both problems for 3.5.0.
> The 3.4.x verison only suffered from the first problem. Patch also
> attached.
There were still some significant startup message problems
with win32tty.
I've spent a lot of time in the debugger tracing through them all.
I think I've got them all worked out now, certainly the ones that
I was aware of. There may be some I haven't discovered.
Testing welcomed of course!
This patch also attempts to diagnose the error where someone tries
to execute NetHack directly out of a zip file, and provide
them with a (hopefully) helpful message similar to what we
might end up telling them if they wrote in. If you want
to test that part, you can comment out the line in the
Makefile that adds "dungeon" to nhdat, and delete the nhdat
in your binary and src directories, and "make install".
Then add the value of your TEMP environment variable as a
DATADIR statement in defaults.nh (here's mine):
DATADIR=C:\DOCUME~1\ALLISO~1\LOCALS~1\Temp
The diagnostic code engages if the game fails to open
dungeon. It then checks to see if it the game dir is the
TEMP directory for your system, and if so it prints the
message.
the win32 cursor restriction stuff messed up any
messages displayed during abnormal start conditions
where the window system never got initialized properly.
among them:
- messages relating to lock files or games in progress
- dungeon errors
- early panic messages
From Newsgroups: rec.games.roguelike.nethack :
> <email deleted>
> Subject: question for windows tty users
>
> I am trying to hunt down a bug, and want to know if I have
> encountered another one of those bizarre "features" that only
> occur on my computer (I seem to get a lot of them).
>
> I can reproduce this bug, or whatever it is, in the official
> Windows binary like this: Start the tty version of NetHack by
> double clicking on the program. You won't see the bug if you
> start it from the command line. When the game asks, "Who are
> you?" press ^C. NetHack will respond with "^C abort. Hit
> <Enter> to end," and then it hangs. Pressing Enter does
> nothing, and the program does not end.
>
> Can anybody else reproduce this behaviour? Thanks in advance.
>
> -- <Someone>
Use fqname buffer 1 for restoring the save file (just like save does when
creating it) so the value won't change out from under the code in unixmain.
- Also moved a tty-specific hack in docompress_file that was causing
the 'y' response to the "keep the save file" prompt to be echoed twice.
fix by <Someone>, updated by <Someone> on r.g.r.n.
changes the colours of the windows tty port so that blue and
bright blue, and cyan and bright cyan are distinguishable. The chief
benefit of this is that floating eyes no longer look like shocking
spheres.
1. Switch to low-level console routines for performance improvements.
2. Instead of moving the cursor around like a real tty, just track the
destination coordinates of where the cursor should be, and
defer the movement until user input is expected.
Credit to <Someone> for #2.
On September 11, 2003 "<Someone>" wrote:
> When we're going to have a different save file format, could
> the last messages in the message history be saved as well, so
> ^P would work the same before and after saving (possibly
> including a few less messages to make room for the startup
> messages?).
This seemed like a reasonable request. This patch:
- adds the core support required.
- adds the tty supporting routines.
Introduce a new set of functions to manage delayed killers in the trunk, used
in addressing the various reports of delayed killer confusion. Since existing
delayed killers are related to player properties, the delayed killers are
keyed by uprop indexes. I did this to avoid adding yet another set of
similar identifiers.
- the new delayed_killer() is used for stoning, sliming, sickness, and
delayed self-genocide while polymorphed. Some other timed events don't
use it (and didn't use the old delayed_killer variable) because they
use a fixed message when the timeout occurs.
- A new data structure, struct kinfo, is used to track both delayed and
immediate killers. This encapsulates all the info involved with
identifying a killer. The structure contains a buffer, which subsumes the
old killer_buf and several other buffers that didn't/couldn't use killer_buf.
- the killer list is saved and restored as part of the game state.
- the special case of usick_cause was removed and a delayed killer list
entry is now used in its place
- common code dealing with (un)sliming is moved to a new make_slimed function
- attempted to update all make dependencies for new end.c -> lev.h
dependency, sorry if I messed any up
Some changes for standard C platforms, to avoid declaring errno explictly.
Such platforms should declare errno in errno.h, which is already included
in the files in question.
Pat Rankin wrote:
> collect them all into some new struct and
> save that separately rather than jamming more non-option stuff
> into struct flags.
This patch:
- collects all context/tracking related fields from flags
into a new structure called "context."
It also adds the following to the new structure:
- stethoscope turn support
- victual support
- tin support
This is an initial round of SAFERHANGUP hangup changes. It introduces
SAFERHANGUP, provides the core framework, and enables it for UNIX.
Window-port changes are provided for win/tty, win/X11 and win/gnome. Qt
changes should be forthcoming after having Warwick look at them.
window.doc is updated so windowport maintainers have an clue what needs to
be done to support SAFERHANGUP.
Although the overlay stuff is destined to be
removed someday, this patch just makes the
use of STATIC_DCL, STATIC_OVL consistent
in the trunk.
[As a side pointless experiment, I was able
to build a working 8086 port of 3.4.2 after
this change that worked correctly in limited
testing right up until it came time to enter
Ft. Ludios., where it couldn't allocated the
required amount of memory.]
> The bug involved using the initalign (and related) indexes into
> the array of alignments as indexes into the respective combo box,
> and these are (apparently) not equivalent. To fix, the combo box
> is queried one by one for the item with the index that produces
> that proper alignment value, and then uses that index found. I
> did not find an API that does this in one step, but this only
> happens once, at dialog initialization.
Changes are: Complete copyright notice, Versions using dot format instead of comma format, and nulls at the end of strings (required, at least for Windows 2000).
Build and install the NetHack.ad file, and change the nethack.sh to append
the HACKDIR to the XUSERFILESEARCHPATH. This means users do not have to
do anything special to use the NetHack X11 resources, nor does it require
teaching nethack where the X11 app-defaults directory is. The change also
updates the X11 and linux doc files in a corresponding way.
> This patch fixes warnings relating to pointers (using int *
> instead of unsigned int * ), provides prototypes for some
> functions, and adds a missing argument to one of the functions
>
> It also changes a bit in the way flex/bison are used in the
> Borland makefile to allow me to test compilation with those
> utilities using a batch file.
Noticed when trying out the "heck patch", where the name of the file
for the sanctum level was changed and I neglected to update my Makefile to
reflect that: nothing was noticed if any files were missing when loading
them into a dlb container. Populating the playground directory for the
non-dlb configuration suffered a similar problem. Now the VMS playground
setup will issue a warning message if this happens (but not abort the
installation; perhaps it ought to do that as well...).
Other ports have the same problem. For example, Unix Makefile.top
relies on file globbing to build the dlb container; if a wildcard pattern
matches no files, dlb won't notice because the shell will have eaten the
offending pattern. Likewise for its non-dlb configuration, where cp won't
notice. However, this isn't a very interesting bug because it won't hit
when the distributed Makefiles correspond with the needed data files.
