I think this assertion must have been put under '#if 0' during early
development, and accidentally never taken out once the game started
actually working. Putting it back in doesn't cause the self-tests to
fail, so I'm reinstating it - if it does fail, I'd like to know about
it!
Some people don't bother to use the right-click UI action that marks a
line as 'definitely not' rather than the initial default yellow
'unknown'. Previously, Loopy gave those people a UI annoyance for some
classes of mistake they made during solving: it would reliably
highlight a clue square with too _many_ edges around it, but not one
with too few - because in normal right-click-ful play, a clue with too
few LINE_YES only becomes an error when there aren't enough
LINE_UNKNOWN around it to potentially become the remaining YESes in
future.
This change arranges that once the player closes the loop, _then_ we
light up underfilled clues, on the basis that adding any further edge
would be an obvious error, so it's no longer sensible to assume that
the user might be planning to come back and do so.
(It's not a very timely notification of errors - it's easy to imagine
someone making a mistake like this very near the start of play and
only finding out about it when they close the loop at the very end. I
discuss possible improvements in a comment, but I don't think any
improvement avoids that problem completely, so I think it may just be
a form of annoyance that right-click-less players have to live with.)
Loopy differs from the other recently-reworked puzzles in that it
doesn't disallow loops completely in the solution - indeed, one is
actually required! But not all loops are what you wanted, so you have
to be a bit more subtle in what you highlight as an error. And the new
findloop system doesn't make that easy, because it only answers the
question 'is this edge part of a loop?' and doesn't talk about loops
as a whole, or enumerate them.
But since I was working in this area anyway, I thought I might as well
have a think about it; and I've come up with a strategy that seems
quite sensible to me, which I describe in a big comment added in
loopy.c. In particular, the new strategy should make a more sensible
decision about whether to highlight the loop or the non-loop edges, in
cases where the user has managed to enter a loop plus some extra
stuff.
If you had a single connected path linking the source to the
destination but _also_ had a spurious edge elsewhere in the grid, then
the spurious edge would be highlighted as an error, but it wouldn't
inhibit declaring the game complete and showing the victory flash.
Tracks's previous loop detector was very basic, and only bothered to
highlight one loop, even if the player managed to create more than one
at a time. Now we highlight all of them.
The old face-dsf based loop detector is gone, and now we just call
findloop instead.
This is just a code cleanup: it doesn't fix any bugs that I know of.
In principle, it provides the same futureproofing we gained by making
the same change in Net, but Slant as a puzzle is less adaptable to
topologically interesting surfaces - in particular, you _can't_ play
it on any edgeless surface like a torus or Klein bottle, because no
filled grid can be loop-free in the first place. (The only way a
connected component can avoid having a loop surrounding it is if it
connects to the grid edge, so there has to _be_ a grid edge.) But you
could play Slant on a Mobius strip, I think, so perhaps one day...
I've removed the old algorithm (the one described as 'footpath dsf' in
the findloop.c appendix comment, though I hadn't thought of that name
at the time), and replaced it with calls to the new API.
This should have no functional effect: there weren't any known bugs in
the previous loop-finder that affected currently supported play modes.
But this generality improvement means that non-orientable playing
surfaces could be supported in the future, which would have confused
the old algorithm. And Net, being the only puzzle as yet that's played
on a torus, is perhaps the one most likely to want to generalise
further at some point.
Bridges only needs a loop detector for its non-default 'don't allow
loops' mode. But the one it had was using the graph-pruning strategy,
which means it had the dumb-bell bug - two loops joined by a path
would highlight the path as well as the loops. Switching to the new
findloop system fixes that bug.
A side effect is that I've been able to remove the 'scratch' array
from the game_state, which was only used by the old loop finder, so
that should save memory.
In the course of another recent project I had occasion to read up on
Tarjan's bridge-finding algorithm. This analyses an arbitrary graph
and finds 'bridges', i.e. edges whose removal would increase the
number of connected components. This is precisely the dual problem to
error-highlighting loops in games like Slant that don't permit them,
because an edge is part of some loop if and only if it is not a
bridge.
Having understood Tarjan's algorithm, it seemed like a good idea to
actually implement it for use in these puzzles, because we've got a
long and dishonourable history of messing up the loop detection in
assorted ways and I thought it would be nice to have an actually
reliable approach without any lurking time bombs. (That history is
chronicled in a long comment at the bottom of the new source file, if
anyone is interested.)
So, findloop.c is a new piece of reusable library code. You run it
over a graph, which you provide in the form of a vertex count and a
callback function to iterate over the neighbours of each vertex, and
it fills in a data structure which you can then query to find out
whether any given edge is part of a loop in the graph or not.
Where possible (mostly with the Nikoli links), they've been updated to their
modern locations. At least one link had to become a Wayback Machine link;
I didn't bother making the floodit.appspot.com link a Wayback one because
there's no content there without the backing of Google App Engine. There
are other implementations online nowadays, of course, but I didn't want to
change the meaning of the text if at all possible. In addition, I added
Flash warnings for the Nikoli pages that now use Flash for instructions.
These are necessary because the argument to a ctype function cannot be
a negative value unless it's EOF. Thanks to Cygwin gcc for pointing
out the mistake, and to Patrick Shaughnessy for this patch.
The web page currently assumes it's called 'rect' rather than
'rectangles', because the web-page building script uses the first
field of each line of gamedesc.txt, same as the Unix binary name.
Rather than add another confusingly-almost-identical field to that
file, it's easier to just rename this one docs section to make the
assumption of equality hold.
The solver's array ctx->iscratch[] is used for a completely different
purpose in the DIFF_HARD code, compared to what it's used for in
DIFF_EASY and DIFF_NORMAL. In particular, a different number of
elements of the array are used - but the code which sets up the array
was not correctly initialising all of them.
I was also unable to find any clear comment that even explained _that_
the purpose of the array was entirely different between the two cases,
let alone explaining _what_ the two purposes were. So I've written
some comments as part of this commit, which should make things a bit
less confusing next time. (Though not much, I admit.)
error: 'r2.x' may be used uninitialized in this function
Its happening when using gcc 5.3 with musl C library. its considering
the case when case falls into default and immediately after exiting
this there is a check if (r1.h > 0 && r1.w > 0) where r1 element is
used but not assigned anything.
GCC is not noticing the control flow where the initilization will
always work due to assertion call can be a function call from libc
Signed-off-by: Khem Raj <raj.khem@gmail.com>
The only situation in which it actually can't find a solution is if
the puzzle is already solved, in which case it can at least fill in
*error to say so before it returns NULL.
This utility works basically the same as galaxiespicture: you feed it
a .xbm bitmap on standard input, and it constructs a game ID which
solves to exactly that image. It will pre-fill some squares if that's
necessary to resolve ambiguity, or leave the grid completely blank if
it can.
The game previously only supported numeric clues round the edge; but
if for some reason you really want a puzzle with a specific solution
bitmap and that bitmap doesn't happen to be uniquely soluble from only
its row and column counts, then this gives you a fallback approach of
pre-filling a few grid squares to resolve the ambiguities.
(This also applies if the puzzle is uniquely soluble *in principle*
but not by Pattern's limited solver - for example, Pattern has never
been able to solve 4x4:2/1/2/1/1.1/2/1/1 and still can't, but now it
can solve 4x4:2/1/2/1/1.1/2/1/1,Ap which has the hard part done for
it.)
Immutable squares are protected from modification during play, and
used as initial information by the solver.
The algorithm for deducing how many squares in a row could be filled
in just from the initial clue set was focusing solely on _black_
squares, and forgot that if a row has a totally empty clue square then
everything in it can be filled in as white!
Now the solver can cope with puzzles such as 3x3:/1///1/ , where it
would previously have spuriously considered that it had no idea where
to start.
The game_state now includes a pointer to a game_state_common
containing all the row and column clues, which is reference-counted
and therefore doesn't have to be physically copied in every dup_game.
Where facilities exist, that is. Like the approach I took with PuTTY
yesterday, Buildscr will now run a code-signing script over the binary
if you specify one in the bob config, and otherwise should fall back
to just leaving that step out.
gtk_misc_set_alignment was deprecated in GTK 3.14. But my replacement
code using gtk_label_set_{x,y}align doesn't work there, because that
function wasn't introduced until GTK 3.16, so there are two minor
versions in the middle where a third strategy is needed.
(That middle strategy doesn't permit arbitrary float alignments, but
that's OK, bceause we only actually use multiples of 0.5.)
Previously moving 1 pixel would be treated as a failed drag and not an unlock.
Now you only have to release the button somewhere on the island you started on.
A user pointed out that if you construct a 'solution' in which no clue
square has too _many_ borders but at least one has too few, and then
bring those clues up to their count by adding extra stray border lines
_inside_ a connected component (avoiding actually dividing any
component completely into two), then the game checker treats that as
solved for victory-flash purposes, on the grounds that (a) the grid is
divided into components of the right size and (b) all clues are
satisfied.
A small example is 4x4n4:22a2b2c33, with the non-solution of dividing
the grid into four 2x2 square blocks and then adding a spurious extra
edge between the two 3 clues. The old Palisade completion check would
flash for victory _at the same time_ as highlighting the spurious edge
in COL_ERROR.
Fixed by enforcing in is_solved() that every border line must separate
two distinct connected components.
The solve move stored in 'aux' by new_game_desc consists of printable
characters in the range '@' to 'O', each representing a 4-bit bitmap
of edges around a cell. But the one generated on the fly by
solve_game() was missing out the 0x40 bit and just returning
characters in the range ^@ to ^O - which would not only have been
horrible if you found such a string in a save file, but also meant
that a game with any completely borderless square would have a
solution move string terminating early due to the ^@, causing
execute_move() to reject it.
Example: ./palisade --test-solve --generate 1 5x5n5#12345-37 now
succeeds, where previously it failed an assertion.
Commit 44e2690ab loosened check_completion's idea of what made a
square count as 'having track in it' for purposes of checking
violations of the row/column counts. Unfortunately, that loosened
notion also applied to the check for the game being complete - so the
game would announce a win as soon as you had every square shaded, even
if you hadn't actually laid all the exact track positions down.
Now we separately count up the number of track-ish squares and the
number of fully completed ones, and use the former for error checking
and the latter for completion checking.
- Count any square as having a track either if the square is marked
as such (rendered as a different background), or if at least one
adjacent edge is marked as containing a segment of train track
(rendered as train tracks if they're placed, else as an '=').
- Do the same counting in rows and columns.
- Mark a clue as an error if too many monsters are seen, even if
some squares are empty.
- Mark a clue as an error if too few monsters are seen, taking into
account how many more sightings are possible given the number of
empty squares and how many times each of them are visited.
Probably because I wrote a couple of loops up to the maximum cell
value using the non-idiomatic <= for their termination test, I also
managed to use <= inappropriately for iterating over every cell of the
grid, leading to a couple of references just off the end of arrays.
Amusingly, it was the Emscripten front end which pointed this out to
me by actually crashing as a result! Though valgrind found it just
fine too, once I thought to run that. But it comes to something when
running your C program in Javascript detects your memory errors :-)
The previous solver could cope with inferring a '1' in an empty
square, but had no deductions that would enable it to infer the
existence of a '4'-sized region in 5x3:52d5b1a5b3. The new solver can
handle that, and I've made a companion change to the clue-stripping
code so that it aims to erase whole regions where possible so as to
actually present this situation to the player.
Current testing suggests that at the smallest preset a nontrivial
ghost region comes up in about 1/3 of games, and at the largest, more
like 1/2 of games. I may yet decide to introduce a difficulty level at
which it's skewed to happen more often still and one at which it
doesn't happen at all; but for the moment, this at least gets the
basic functionality into the code.
Jonas Kölker points out that commit a800ff16b (which fixed a bug in
the previous attempt) left in another bug: if the puzzle size was
changed while the window was maximised, the system would fail to
recompute the tile size and would redraw completely wrongly.
So it's not optional after all to run midend_size(), even if the
drawing area size hasn't changed. I've reverted that code to be
unconditional, and now only the refresh of the Cairo-side backing
store system is conditionalised - and that's conditionalised on
changes to either the size of the actual window _or_ the size of the
contained pixmap. (The latter should defend against redraw failure in
the case where the puzzle aspect ratio changes, so that neither the
window size nor the tile size changes but a recentre is still needed.)
I _think_ this now fixes all the cases of resizing: this time I've
tested making an unmaximised puzzle window bigger or smaller, and
within a maximised window, forcing the puzzle to scale up, scale down,
or change its aspect ratio without changing its tile size. All work,
on GTK2 and GTK3, and yet we still don't get the visible flicker on
status line updates that was the reason I started fiddling with this
code in the first place.
(We _do_ still call configure_area on every update of the status line,
at least under GTK3; that's going to cause a forced full redraw on
every timer tick in Mines, which is wasteful of CPU, so it would still
be nice to find a better way of identifying the cases in which no
resizing at all was necessary and we could avoid renewing the game
drawstate. But the current code at least doesn't have any display
_errors_ that I know of, which is an improvement on its previous
state.)
- Lay bridges (crosess) with Control-arrow (Shift-arrow)
- Jump (non-orthogonally) to nearby islands with number keys, a..f
- Mark islands as done with a single tap on the space bar
- The file selector for loading and saving gets a g_free().
- The handling of saving (menu_save_event) gets an sfree().
- It's also slightly restructured to prevent future errors.
- menu_load_event was already structured to prevent this error.
- The OLD_FILESEL code seems to not need a g_free().
Position the cursor in the top (bottom) row, press enter and press up
(down). The game acts as if you had pressed right, both with Enter-
and Space-based dragging.
Pressing Ctrl-arrow or Shift-arrow on a tile now moves the row or
column under the tile. With Ctrl, the cursor moves as well so you can
keep making moves that affect the same tile; with Shift, the cursor
stays put so you can keep making moves that affect the same location.
These button codes are generated by the back/forward button pair on
the sides of some mice, and web browsers treat these as the back and
forward actions in the page history.
Pressing H now suggests the lexicographically first row consistent
with all previous feedback.
The previous function of the H key to toggle a hold marker on the
current peg is now performed by Space / CURSOR_SELECT2, which is more
in line with other puzzles anyway.
Recall that the hint feature is really an incremental solver. Apply
it repeatedly until the board is solved. Grade puzzles as solvable
or unsolvable by checking their parity.
This is really an incremental solver. It alternates between solving
rows and solving columns. Each row and column is solved one piece at
a time. Except for some temporary trickery with the last two pieces
in a row or column, once a piece is solved it is never moved again.
(On non-square grids it first solves some rows or some columns until
the unsolved part is a square, then starts alternating.)
