I tracked down all the functions that took in an entity's drawframe and
made sure that no matter what value an entity's drawframe was, the game
would never segfault.
To deal with using a different image file for Flip Mode, it looks like
copy-paste was used. This isn't exactly maintainable code. So I'm
replacing it with a reference that changes depending on if the game is
in Flip Mode or not, instead.
Removing the player entity has all sorts of nasty effects, such as
softlocking the game because many inputs require there to be a player
present, such as opening the quit menu.
The most infamous glitch to remove the player entity is the Gravitron
Fling, where the game doesn't see a gravity line at a specific
y-position in the current room, and when it moves the bottom gravity
line it moves the player instead. When the gravity line gets outside the
room, it gets destroyed, so if the player gets dragged outside the room,
they get destroyed, too. (Don't misinterpret this as saying anytime the
player gets dragged outside the room, they get destroyed - it's only the
Gravitron logic that destroys them.)
Also, there are many places in the code that use entity-getting
functions that have a fallback value of 0. If it was possible to remove
the player, then it's possible for this fallback value of 0 to index
obj.entities out-of-bounds, which is not good.
To fix this, entityclass::removeentity() is now a bool that signifies if
the entity was successfully removed or not. If the entity given is the
player (meaning it first checks if it's rule 0, just so in 99% of cases
it'll short-circuit and won't do the next check, which is if
entityclass::getplayer() says the indice to be removed is the player),
then it'll refuse to remove the entity, and return false.
This is a change in behavior where callers might expect
entityclass::removeentity() to always succeed, so I changed the
removeentity_iter() macro to only decrement if removing the entity
succeeded. I also changed entityclass::updateentities() from
'removeentity(i); return true;' to 'return removeentity(i);'.
Apparently it results in Undefined Behavior if the argument given isn't
representable as an unsigned char. This means that (potentially) plain
char and signed chars could be unsafe to use as well.
It's rare that this could happen in practice, though. std::isdigit() is
only used by is_positive_num() which is only used by find_tag(), so
someone would have to deliberately put something crazy after an `&#` in
a custom level file in order for this to happen. Still, better to be
safe than sorry and all.
This fixes a compile error that could happen where the compiler doesn't
know what std::isdigit() is, but I'm puzzled as to why this wasn't
happening earlier, 'cause I've only been reported that it happens by
only one person.
Continuing from #280, another potential source of out-of-bounds indexing
(and thus, Undefined Behavior badness) comes from script commands. A
majority of them don't do any input validation at all, which means the
potential for out-of-bounds indexing and segfaulting in custom levels.
So it's always good to add bounds checks to them.
Interesting note, the only existing command that has bounds checks is
the flag() command. That means you can't turn out-of-bounds flags on or
off. But there's no bounds checks for ifflag(), or customifflag(), which
means you CAN index out-of-bounds with those commands! That's a bit bad
to do, so.
Also, I decided to add the bounds checks for playef() at the
musicclass::playef() level, instead of just the level of the playef()
command. I don't know of any other cases outside of the command where
musicclass::playef() will index out of bounds, but musicclass is the one
containing the indexed vector anyway, I wanted to cover more cases, and
it's better to be safe than sorry.
And this the function with the least amount of cases where its sentinel
value is used unchecked. Thankfully. obj.getplayer() was a bit of a slug
to get through.
obj.getplayer() can return -1, which can cause out-of-bounds indexing of
obj.entities, which is really bad. This was by far the most changes, as
obj.getplayer() is the most used entity-getting function that returns
-1, as well as the most-used function whose sentinel value goes
unchecked.
To deal with the usage of obj.getplayer() in mapclass::warpto(), I just
added general bounds checks inside that function instead of changing all
the callers.
A few months ago, I added ghosts to the VVVVVV: Community Edition editor. I was told recently I should think
about upstreaming it, and with Terry saying go ahead I finally ported them into VVVVVV. There's one slight
difference however--you can choose whether you have them or not in the editor's settings menu. They're off by
default, and this is saved to the save file.
Anyway, when you're playtesting, the game saves the players position, color, room coordinates and sprite every 3
frames. The max is 100, where if it tries to add more, the oldest one gets removed.
When you exit playtesting, the saved positions appear one at a time, and you can use the Z key to speed it up.
[Here's a video of them in action.](https://o.lol-sa.me/4H21zCv.mp4)
2.2 and earlier had this god-awful thing where it put the closing tag of
an edentity onto the next line, and then kept the indentation the same.
This requires parsing the XML in an extremely specific way (i.e.
ignoring the whitespace) so the newline and indentation isn't taken as
part of the actual contents of the tag.
2.3 removed this awful whitespace entirely to make it easier on parsers.
When I tested #270, I tested against a 2.3 re-save of Dimension Open and
diffed the two, because I thought testing against the original version
of the level would result in a bunch of noise I didn't want due to the
whitespace change. Well, I did exactly what I intended, and ended up
ignoring the whitespace change so much that levels saved in this stupid
format ended up getting broken.
Luckily, we can just tell TinyXML-2 to parse a document exactly like how
TinyXML-1 would've parsed it, by supplying the COLLAPSE_WHITESPACE enum
to it (by default it's on PRESERVE_WHITESPACE).
This removes the TinyXML source files, removes it from CMakeLists.txt,
removes all the includes, and removes the functions
FILESYSTEM_saveTiXmlDocument() and FILESYSTEM_loadTiXmlDocument() (use
FILESYSTEM_saveTiXml2Document() and FILESYSTEM_loadTiXml2Document()
instead).
Additionally I've cleaned up the tinyxml2.h include in FileSystemUtils.h
so that it doesn't actually include tinyxml2.h unnecessarily, meaning a
change to TinyXML2 shouldn't rebuild all files that include
FileSystemUtils.h.
Seems a bit wasteful to do the whole "parse the XML document thing"
instead of a simple file check. It doesn't even fail if the XML document
is invalid, but whatever.
Ok, so it was a bit of a struggle at first figuring out the new API, but
honestly it wasn't so bad in the end.
I made a copy of my old unlock.vvv before testing this, and checking
with `diff` the only difference is the new `encoding="UTF-8"` in the XML
declaration, which isn't a bad thing.
Surprisingly, I only had to change some names and stuff around at the
top of the function. The rest of the function could be left untouched
and it worked fine.
Some of the file was indented with two spaces and the rest indented with
tabs. It feels like two different people worked on the file, one more
than the other. Since most of it uses two spaces, I'll just replace the
tabs with two spaces.
This is to respect the fact that the top half of the file is indented
with spaces, while the bottom half is indented with tabs.
Graphics::reloadresources() is on the bottom half.
The previous way manually concatenated the first 7 characters of the
string together (and had an std::min() calculation). The new way instead
does std::string::substr(), which is much more snappy.
The actual unindent is done in a separate commit to minimize noise,
because diffs are terrible at clearly conveying unindents (it should put
all the minus lines together and all the plus lines together, too).
The entirety of the rest of scriptclass::loadcustom() is encased in a
block that first checks if the script with the name even exists. Instead
of indenting the rest of the function, just invert the check and reduce
indentation level.
