Quick C++/SFML Tips

While I’m writing a series on working with SFML and C++, I thought I’d share some quick dirty tips for working with SFML that I’ve experienced lately. Some of these emerged while branching out to other development contexts that I’m normally not entrenched in – so you’ll forgive me if they seem axiomatic to you – and others simply failed to make the transition from mind to paper (or screen in this case).

Getting Started with SFML and Visual Studio

It’s evident after seeing some posts on the SFML forums that people don’t RTFM. TL;DR isn’t a thing to worry about here, so be sure to check this page (linked below) out. Visual Studio doesn’t require counter-intuitive thought concerning environment configurations – a compiler is a compiler – but the way one configures the compiler is measurably convoluted, especially if you’re used to programming in the Linux world. These steps are also valid if you’re considering creating a DLL to leverage shared code.

SFML on Visual Studio – https://www.sfml-dev.org/tutorials/2.5/start-vc.php

DLL Woes

Speaking of creating DLLs, there’s a nasty little caveat with the default Windows header file. Evidently the versions of the min and max functions implemented in it are grossly incompatible with the ones in STL. While not a SFML issue per-se, it’s important to be aware of because it’ll likely creep in when you least expect it, and trying to determine what the root cause is from the output of the compiler is going to require several witchdoctors and an irrefutable, globally-accepted proof of String Theory. The red herring for this typically comes in the form of error C2589: ‘(‘ illegal token on right side of ‘::’ (a.k.a. The go f-yourself error).

The fix for this is the NOMINMAX preprocessor directive. You can either add it as a file-level define at the head of the file, or you can use the Project Properties dialog and add it to All Configurations and All Platforms by navigating to C/C++->Preprocessor, and appending the NOMINMAX option to the Preprocessor Definitions field. If ever you come back to this dialog to ensure that the value was set, you’ll need to drill-down into each configuration and platform to see that the value was applied.

Deleted Copy Constructors, sf::NonCopyable

A core component of a game engine that I wrote has an Asset Manager that’s very similar to the one used in MonoGame, except that it doesn’t use the Pipeline concept. Assets are loaded into memory via PhysicsFS, and they’re translated into SFML Asset Constructs that are stored in a STL Container, specifically std::unordered_map. Some SFML Asset Constructs, specifically sf::Music, inherit from classes that leverage sf::Thread, and, of crucial note, sf::Thread inherits from sf::NonCopyable. While this utility class doesn’t explicitly delete the copy constructor and assignment operators, it marks them as private. Children of this class will likely, if you’re using C++11 or greater, have these functions implicitly deleted since they’re not valid. In the absence of STL Containers, this isn’t too much of an issue, especially since attempts at copies or assignments would result from explicit statements that you yourself wrote. When STL Containers are around and you encounter an error from implicitly deleted function calls, we’ve traipsed into another arena where compiler output is infamously horrid to the degree of being near useless.

To give some concrete to my exposition, the offending statement was this:

...
typedef sf::Music sfmusic;
typedef std::unordered_map < std::string, sfmusic > ab_bgm;
...

std::unordered_map leverages std::pair to join the key to the value, and while I haven’t been able to dissect the issue deeper than this, it would appear that std::pair is likely subsuming the lifecycle of the objects it contains. Because there is no copy constructor or assignment operator for an object that inherits from sf::Thread, and because std::pair is attempting to leverage either one of those functions in some way, the latter is going to throw up in the most flamboyant of ways.

Although what’s next is likely not a representation of a clean or efficient way to mitigate this, I’ve found that it works. For starters, the declaration changes slightly:

...
typedef sf::Music sfmusic;
typedef std::unordered_map < std::string, sfmusic* > ab_bgm;
...

Next, the member function of the Asset Manager that is responsible for copying the asset data from raw bytes into live SFML Asset Constructs takes an extra step of manually allocating the memory for it before using the sf::Music copyFromMemory function:

...
case targetloader::bgm:
    bgmb [file] = new sfmusic ();
    bgmb [file]->openFromMemory (d, f.length ());
...

Of course, because we’re now wandering down the path of explicit memory allocations, we’ve got to be responsible for cleaning it up, so the intermediate destructor does some work to delete allocations in this bank, if there were any, before removing the bank itself.

 

Monogame – Working with Touch

*NOTE – This may also work on other platforms with touch as well, but this hasn’t been tested on anything other than Android at the current moment.

One fundamental aspect to understand about touch and gestures is that code needs structured around the idea of continuous input. Thus, any touch recognition code should be encapsulated somewhere within the main game loop; where this is would be dependent upon how one elects to parse input. The reason for this is that the nature of all touch processing is a sequence of touch events that get fired in succession; there are only a few exceptions to this rule.

The other aspect to understand clearly is nomenclature mapping. What you think you want to achieve in your design may, unless you’re already familiar with either the Monogame or Xamarin frameworks, not necessarily be what either of those aforementioned frameworks are calling it through its API parlance. So be prepared to step outside your normal wheelhouse and, perhaps, discard some long-standing assertions about what truly a touch/gesture really is.

First, the consideration needs made as to if you’re going to handle touches or gestures – don’t get too caught up on thinking that a touch is a single press, because this will cause you some unnecessary grief later. If your program ebbs toward the former, there’s usually no requisite configuration needed, other than to check for the existence of a touch-capable input device and adjust your engine accordingly (the only other exception here would be if one wishes to perform touch-to-cursor mappings, but this is outside the immediate scope of this article). Conversely, gesture handling requires some configurations be made before they can be used, and this is where the Monogame documentation, as well as a majority of the information available on various forums, falls fatally short. We will focus on this once we start addressing gesture handling, which follows a look at touch handling.

Checking for the existence of hardware with touch capabilities can be queried through the TouchPanelCapabilities object, triaged through the TouchPanel static object as such:

TouchPanel.GetCapabilities ().IsConnected;

The IsConnected property is a boolean that indicates the presence of a touch-capable device. Obviously, the absence of one suggests either that the device is intermittently available, or that other sources of input are necessary. In this case, neither the touch nor gesture APIs would be valid.

Concerning touches, the current state of the touch panel holds a collection, called TouchCollection, of TouchLocation objects. Each of these correspond to a single touch event generated by the user when a press is determined to have occurred on the sensitive surface. It’s fair to suggest that multiple points of contact yield multiple TouchLocation instances within the TouchCollection (i.e. as a practical example, one finger would generate one instance, two fingers would generate two instances, etc. This can best be observed while debugging a program using this technique, and paying attention to the Id property of the TouchLocation instance(s).). In order to properly ascertain the most accurate state of the touch panel, we’ll need to do the following:

  1. Obtain the current state of the touch panel
  2. Determine if there are any TouchLocation instances in the TouchCollection collection
  3. If there are, take relative action based on any of the properties of the current TouchLocation instance

This can be accomplished using the following general template

TouchCollection tc = TouchPanel.GetState ();
foreach (TouchLocation tl in tc) {
 // Do something here
}

Fortunately, TouchLocation instances are very simple to use. There are only four properties that are of any practical significance, and one utility function for trying to determine the previous location (which, at least during my testing, wasn’t useful for anything). They are as follows:

  • Id – An Integer that uniquely identifies a single TouchLocation instance within a TouchCollection. It’s unknown how its potential values are determined, or if orphaned values are subject to recycling after the host TouchLocation has fallen out of scope.
  • Position – A Vector2 that provides the X and Y coordinate serving as the source for the event. Effectively, this is the location on the sensitive surface where contact was made.
  • Pressure – A floating-point number indicating the pressure of the touch. I’m unclear on how exactly this works, since my tests always reported a zero in this property. The only conclusions I can come up with here are either that my device doesn’t support touch sensitivity of this kind, or I missed a configuration step to enable this functionality.
  • State – An instance of TouchLocationState that determines what kind of touch we’re dealing with. This property can have one of four values:
    • Invalid – The press has been somehow deemed invalid; I’ve never seen this occur in testing, and am left to think that either I failed to satisfy conditions to make this occur, or that it’s simply a dump case for an exceptional condition.
    • Moved – Seen when a press has either remained in the same position or has moved. This is a very important state as it has a great deal of practical application, so do try to keep it in mind.
    • Pressed – Observed when a point of contact on the sensitive surface is newly minted. This is only fired once, and will not be seen again, regardless if the contact remains afterward. This could potentially be a source of confusion for a number of programmers.
    • Released – Likely the last state that a TouchLocation would be in before falling out of scope, it will be fired when the source of contact which generated this particular instance is no longer available.

Having said all of that, you now have enough information to properly utilise touches within your programs. As was stated before, simply ensure that your code is contained within the main game loop somewhere, since the data will likely change per frame, or at least as often as it can. An example of how to implement code based off this logic will be illustrated at the end of this discussion.

Gestures, insofar as Monogame is concerned, are a bit of a hassle to work with, mostly due to the lack of upstream documentation. We will seek to correct this imbalance, if not for the official documentation, then at least for programmers who wish to learn more about them.

As was stated previously, although Monogame performs a considerable amount of work to make available the touch and gesture API for programs deployed on capable hardware, gestures are left a bit open-ended by comparison to their touch counterparts. What this means is that as a programmer, you’re required to provide some configuration before the gesture API can be utilised. Here, we assume that you’re only interested in gestures, and not touch-to-cursor mappings, hence will only discuss the former.

Before proceeding, some basic information should be given as to the nature of gestures, and how they’re procedurally handled.

Although there is some very minor overlap with how a touch and gesture are expressed, they are two discrete entities. Gestures can be composed of one or more points of contact, and it’s expected that the locations of these contacts will change, in a particular way, over an indeterminate amount of time (ergonomic measurements would likely dispute this generalisation, but it is for this conversation a generalisation, and not a scientific claim). The ways in which these contacts change, or at least the resultant shape the change yields, as well as the number of contacts involved in the measurement, hints at the particular kind of gesture. In other words, a single point of contact that has a gradual change in its X-axis, be it positive or negative, which would yield a ray (or a series of small rays), is generally considered to be a horizontal drag gesture. When considering the same principles but looking to the Y-axis instead, we now find ourselves dealing with a vertical drag gesture. Pinch and zoom gestures typically involve two or more points of contact that move near concertedly away from or toward each other along the same logical axis. Perhaps paradoxically, at least when considering a contrast between touches and gestures, taps, double taps, and long-presses are registered as gestures as well; these are more concerned with the sustainment of a single point of contact relative to the time from when it was first recognised.

From a stock perspective, Monogame provides eleven types of gestures, referred to as GestureTypes. These types will effectively determine how gesture detection is performed (it’s unclear if the GestureType framework can be extended to facilitate custom gesture types, but this is a considerably advanced topic which will not be discussed here). However, Monogame will not automatically read the touch panel for gestural input. Instead, it needs instructed on which kinds of gestures to detect, and this is provided by the programmer. In any non-looping block of code, preferably during the initialisation routines, you’ll need to specify what are called EnabledGestures, which is a property of the TouchPanel static object. Multiple gestures can be configured by OR’ing one or more of the types together in the assignment statement. For example, if I wanted to parse for both a HorizontalDrag and a DragComplete gesture, I would write the following statement:

TouchPanel.EnabledGestures = GestureType.HorizontalDrag | GestureType.DragComplete;

Once this is complete, you’ll have done enough to get Monogame to start playing nice with at least these two kinds.

Parsing gestural input is, in essence, no different than parsing touch input, but there are some minor differences to the process. To start, we must first determine if there are any gestures with which to read data from. If we do not do this, attempts to read directly from the gesture store will generate fatal exceptions. Fortunately, the TouchPanel static object provides a boolean property called IsGestureAvailable, which will inform clients of the availability of queued gesture data. If we have data, we must convert the data into a sample, which is packaged into the GestureSample class. As with the TouchLocation object, the GestureSample object contains several properties that are of practical interest to the programmer, especially when making contextual decisions that respond to this kind of input. GestureSamples include the following properties:

  • Delta – A Vector2 instance which provides the delta, or difference, data for the first touch point in the gesture. This will change over time, and will always be relative to the coordinates where the touch was first recognised.
  • Position – A Vector2 instance which contains the current coordinates of the first touch point in the gesture.
  • Timestamp – A TimeSpan instance that indicates the time when the gesture was first recognised.
  • GestureType – A GestureType instance that indicates what type of gesture was determined based off several criteria.

Additionally, GestureSample contains properties called Delta2 and Position2, which are used to track a second point of contact that’s being measured as part of the current gesture. What this implies is that insofar as the stock gestures are concerned, Monogame will only be able to handle gestures where no more than two points of contact are involved.

My advice here is to experiment with the data through debugging until you’re comfortable with how these gestures are read, because there are some nuances with how the data continuously polls respective to different gesture kinds. For example, a HorizontalDrag gesture will, while the drag is occurring, constantly emit the HorizontalDrag signal until the contact source is released, terminating the gesture. At this point, if one is checking for the DragComplete signal as well, releasing the contact source will cause the touch panel to emit the DragComplete signal.

Examples:

To determine if a single press has been made:

TouchCollection tc = TouchPanel.GetState ();
foreach (TouchLocation tl in tc) {
 if (TouchLocationState.Pressed == tl.State) {
  // Execute your domain-specific code here
 }
}

To determine if a press was made, and has been held in the same position for an arbitrary period of time:

TouchCollection tc = TouchPanel.GetState ();
foreach (TouchLocation tl in tc) {
 if (TouchLocationState.Moved == tl.State) {
  // Execute your domain-specific code here
 }
}

To track the position of a horizontal drag:

(1) During game initialisation:

TouchPanel.EnabledGestures = GestureType.HorizontalDrag;

(2) During game loop:

while (TouchPanel.IsGestureAvailable) {
 GestureSample gs = TouchPanel.ReadGesture ();
 if (GestureType.HorizontalDrag == gs.GestureType) {
  // Execute your domain-specific code here
 }
}

PhysicsFS/PhysFS++ Tutorial

This is a follow-up from a promise that I made in my tutorial video on designing an asset manager using SFML. That video can be found here.

This tutorial is centred around PhysFS++, a C++ wrapper for PhysicsFS. For the sake of the tutorial, it’s assumed that the reader is familiar with PhysicsFS and what it’s capable of. Described here is a workflow for simple use of the library. Anything more discrete is beyond the scope and will have to be ascertained by the reader on their own accord.

Lastly of note, PhysFS++ encapsulates PhysicsFS calls in a PhysFS namespace. The functions are global within this namespace and there are only a few classes that are provided. PhysFS++ further encapsulates key PhysicsFS constructs, notably those corresponding to archived files, into said classes that are derivatives of STL stream classes (a huge boon).

Like some other libraries, PhysicsFS requires explicit initialisation before it can be used. This is facilitated by a function named init. It takes one argument of type const char* and is semantically directed at a terminal invocation argument triaged through the much loved argv. However, this can be an empty string especially if you’re not expecting to handle terminal invocations. Thus, a very general call to init can be performed as such:

PhysFS::init (nullptr);

Right of the bat we have to mention a caveat here. From PhysFS++, there’s no way to assert the initialisation process of PhysicsFS. This is counter-intuitive to the upstream library which relies on the tried-and-true zero on failure, non-zero on success return values as sanity checks. Furthermore, while PhysFS++ implements C++ exceptions, the init function doesn’t throw any at all. Ostensibly, what one is left with is an unchecked call to init. Because one needs to compile PhysFS++ from the source, it is possible to modify the code, as I have done, to add a check to this call.

Once init has successfully completed, the next step is to mount an archive file into the virtual filesystem created by init. This is performed with the mount function. mount expects three arguments: the archive file on disk, a string specifying a mount point in the virtual filesystem, and a boolean which appends the mount point to the search path. One should place the archive file in the working directory for the binary file of their executable so PhysFS++ can see it. The second argument can be an empty string which would force root mounting, and the third can be true. Thus, for our example, if we assume we have an archive file named funphotos.zip on our disk in the binary’s working directory, we can issue a call to mount as such:

PhysFS::mount (“funphotos.zip”, “”, 1);

Unfortunately, as was the case with init, mount wraps an upstream function that adheres to the zero-or-nonzero return value paradigm that is outright dropped by PhysFS++ with no exception catering otherwise. Thus, if mount fails to mount the archive for whatever reason, it’ll be a little difficult to ascertain why; plan accordingly.

Assuming mount has returned properly, one can start to work with the files that are contained in the archive. At this point, you should begin working in the mentality of filesystem calls. It’s possible to have archive files with complex directory layouts which would require one to perform recursive searches. That being said, everything should be considered a file – even a directory. An analysis of the PhysicsFS and PhysFS++ APIs will provide for you the full breadth of your available capabilities so for the sake of brevity, only a select number of those will be touched here.

Enumeration of files in a directory can be performed with the enumerateFiles function. It takes as an argument a string which indicates the directory to use as a root for the enumeration. As a return it provides the caller with a vector of strings indicating the name of the file. For ease of use, a call to this to enumerate the files in root can be performed as such:

auto rootfiles = PhysFs::enumerateFiles (“/”);

To actually work with a file in the filesystem, one needs to use the PHYSFS_file ADT. This upstream ADT is encapsulated by one of three classes: PhysFS::ifstream, PhysFS::ostream, or PhysFS::fstream. Each of these is fantastic in that they encapsulate a PHYSFS_file ADT in a standard stream which means that one can now perform established stream operations on the file itself. Each class has its own use: fstream for reading files, ostream for writing to files, and ifstream for bidirectional file handling. In any case, any of these classes can be instantiated with a string argument that contains a fully qualified pathname to a file in the virtual filesystem that one wants to work with. For example, given that our virtual filesystem has a file named goofycats.png located at the path /textures/cats, we could instantiate a read file as such:

PhysFS::fstream gc (“/textures/cats/goofycats.png”);

Again, the underlying upstream call to open the file goes unchecked.

What you do from here is largely subjective relative to your program’s context. Say, for example (and this is actually pretty specific to my own use case), that you have a series of files of various formats in an archive that are being used in a video game program. The multimedia library you’re using should have ADTs that represent types of assets such as textures, fonts, etc. One could do some work to transpose the raw data of these streams into the aforementioned ADTs for use in the multimedia library. The following snippet of code, borrowed from one of my own projects, illustrates this. A PhysFS::fstream instance named f is created with a certain file. Following up is a char array named d is instantiated with the size of f. The read function of f is called to transpose the bytes from f into d. Then, d is used to instantiate an instance of sf::Music from SFML contained in a std::unordered_map:

code

This covers most the basic and general needs of users of PhysicsFS. Should you require anything else from it, you’d do well to read up on the Doxygen file of PhysicsFS or the source for PhysFS++. Lastly, when one is done with the library, you’ll need to call a deinitialization function called deinit:

PhysFS::deinit ();

Vagabond 1.0 – Nearly Here

Yes that’s right. After much toil, and missing my self-imposed deadline for a release, I’m nearly complete with the program and can give it a 1.0 which means that it’ll go into General Availability.

I’m not posting Vagabond News on its website because I want that strictly to be a place to get the program from and a reference for information that developers would be interested in. Just wanted to get that out of the way as well. 🙂

Right now the program is about 98% done. All that remains is to test two additional features, push to the repository (assuming they work, which they will) and then build Debian, RPM and Source Taballs for deliverables. I haven’t built Debian packages for quite some time so I’m having to go through several refreshers to bring myself back up to speed on the workflow.

After this release, I have some additional features I want to work in so development will continue alongside general maintenance coding. I want to be able to add support for choosing to use either the official Android SDK or to use the Replicant SDK. This is more of an ethical decision as I feel like Replicant is a great choice for developing in an open-source manner but might not be the best idea if you’re writing Android to be published on Play. But the option should be there. I’m also considering the idea of implementing SDK Administrative Task Shortcuts. These are commands that can be passed to Vagabond that will inflate to more robust tasks passed to the VM that maintain the SDK.

If all goes as planned, and work and life leave me alone long enough, I can potentially have this ready to go by the end of the week.

Another Vagabond Update

I wanted to take some time out from working on it to provide you with an update to the Vagabond project.

Recently I had some piqued interest on Twitter for the project and it’s lit a little more of a fire under my arse to get it going. I’d continued to work on it but I was putting in on the back-burner in favor of other projects that I’ve been working on. It’s safe to say that I’ve got enough interest now to get it done.

If you need another synopsis on it, please check out my previous post about what Vagabond is.

Changes

My PoC code was a shell script that did the whole thing for me. I’ve since migrated the project to C++ and am going to build both Debian and RedHat Packages and also provide a source tarball. After some consideration, I may actually take this and eventually migrate to a web-interface in Perl but I’m still not too sure yet. I’ve got too much work going into the C++ conversion that I need to stick with that and see where that takes me. Additionally, I’ve put the development branch up on my GitHub. The project is under the BSD 3-Clause License so you can do what you want with it so long as it carries the license and you don’t use my name on derived works without pinging me first.

Right now the project is sitting at about 3011 lines of code (tallying both insertions and deletions but the graph is laid out on the GitHub). This may seem inflated but I’m adhering to the 80-line gutter since I’m editing exclusively in vim and I hate the word wrap on there. Rightly this could be consolidated and I’d probably manage to shave off a hundred or so lines. A compilation is possible as it is now and the help feature is 100% complete. The create option is responsive to tests but doesn’t do anything useful at this point and the Vagrant Propagation command isn’t coded yet. Most of the code might seem redundant compared to the PoC but I wanted to use some of the more robust features of C++ like Exceptions and Namespacing for proper encapsulation.

I’m still looking to be able to have this done by the end of next week. At least a rev one. If anything comes up that keeps that from happening, I’ll change the date but I’m pouring nearly all of my free time into the project so I should be able to hit it.

Ohio Linux Fest – The Aftermath

I’m back home from the Ohio Linux Fest and I’m very, very tired. As much as I want to write more, I think I’m going to keep this short and sweet for the time being and just give a minor overview and save most of the major content for some of the upcoming podcasts that I’m going to be co-hosting to go over the event.

First of all, I know that some of you reading this most likely attended the event and even came to my talk to which I want to extend an infinite amount of gratitude to you for. Little fact – this was my first talk ever at a technology conference. That being said, the turnout far exceeded any expectations I had and the audience was amazing. I got so much great feedback from all of you and you all graciously put up with my horrible jokes. Even more humbling to me was the swath of people who came up after I was done. That made all the stress and shitting bricks worth it.

That being said, the slides that were uploaded to the OLF website on my session profile are not the ones that I used for the presentation. The ones I used were modified in terms of order and containing a few slides that detailed my previous work. Some of you had asked about getting a copy of the slides and I want to make sure that you get the most recent ones so I’m going to upload them and you can download them at your leisure. Of course, I won’t be doing this until tomorrow since I’m seriously going straight to bed right after this. You all can wait until then.

The other thing that I didn’t have that I got asked for was business cards. Thankfully my contact information is “federated” in the sense that all the names are similar. My website is gregfmartin.com, my email is greg@gregfmartin.com and my phone number is 1-800-gregfmartin (not really but it sounds good). Had I known that there was going to be this clamoring for my contact information, I would have better prepared for it.

Finally, I wanted to give honorable mention to Brian Wagner for being not only a tremendous help and resource but a great friend. Also to Stephen McLaughlin a.k.a. DoorToDoorGeek for being a wealth of audio production knowledge and for help with the audio recording equipment (even though I screwed up due to technical difficulties).

Development Journal – Smsr Update 001

I’d like to think that I started working on a rough draft with Smsr and then it evolved into something else.

This “rough draft” had a bit of code in it encapsulated in an entity called ConversationThread. A function of this class in particular was really nasty since what it did was take all of the SMS messages on the device and effectively organize them in a cohesive manner that didn’t rely on querying sub-providers like Sms.Inbox or Sms.Sent. The reasoning behind this was that when I would query some of the more specific Content Providers, like Sms.Conversations, the Cursor returned from that query would only contain three fields regardless of the applied projection. In other words, despite the fact that the contract class inherited from base interfaces that gave it the additional fields you’d want, this inheritance doesn’t apply to what’s returned from queries against the provider because these columns are missing from the Cursor. So it wouldn’t contain things like ADDRESS or _ID. And because these Content Providers weren’t designed with the ability to perform SQL Joins via queries (despite the fact that the documentation would lead you to believe so), the only other alternative, really, is to perform several separate queries against several different Content Providers and join them using either MatrixCursor or CursorJoiner. The issue there is (A) how to do this without blocking on the UI thread and (B) because subsequent queries would rely specifically on information that was contained in previous queries, how then to make sure that the queries had completed before attempting to use them?

I really tried to avoid the solution where multiple queries were involved because it just seemed so… lackluster. Not to mention that the idea of tossing back and forth between five or six Cursors seemed like an explosion in the works. But there was a major issue with the way I was doing it originally – real-time updates. Because the backing data was effectively abstracted into this seemingly convoluted construct that didn’t directly bind to the backing data, if something changed in the data (i.e. received a new text message), the structure would effectively have to be completely rebuilt. One pass on building that thing was expensive enough since it relied heavily on Collections. Doing it over and over again could lead to some really janky issues.

So I went back to the drawing board. And what I came out with, after struggling for a bit to understand MatrixCursor, was a solution that does in fact query multiple Content Providers which will yield several Cursors and then combines them into a MatrixCursor which is then given to my custom CursorAdapter. The result here also adds to a modified layout for the Conversation Stubs which uses a LayerDrawable to get the contact’s image (default if none) and apply a gradient over top of it that leads to the other half of the list.

Unfortunately, this change has broken the Activity that will show the Conversation Thread because the data that’s now contained in the stubs is different than what that Activity was looking for in the first place. But the stub viewer looks nice! 😀

If you want to follow along, the branch that this work is being done on is the “rewrite” branch.

https://github.com/gregfmartin/smsdemo/tree/rewrite