I would like to provide an additional reminder / incentive to my colleagues to write tests for new functions that they contribute to a package I have developed.
I am not sure if this is even a good idea, or if there is a better approach, but I would like to know if there is a way to use the automated testing of files in /tests and /inst/tests to encourage developers to write tests for new functions.
For example, is there a test that will throw an error when the package is checked, built, or tested if there are any functions in a package without an associated test?
I recognize that there are many potential problems with this. It would be difficult to determine if the test is any good, and it might be a waste of time to test a trivial function. But it would still serve as a reminder that a test should be written, and perhaps it could enforce one or two very basic rules.
Is there a test that I can write that will prevent the package from compiling if there is a function without a test? I am using the testthat package in R.
It doesn't need to be terribly robust, but the ability to exclude functions by name would be useful.
It sounds like you're looking for "code coverage" functionality. The first step here would be to figure out which code is covered by tests and which isn't. Check out section 2.2 of the RUnit Vignette for information on how to do that.
You may end up having to modify the inspect function in order to throw alerts or handle custom events like that, but I think they have the bulk of what you're trying to do already implemented.
Related
I am new to creating my own packages and I am using roxygen2.
I am creating a package with a lot of internal helper functions and I was wondering if I have to document all of them. I understand the importance of documentation but some functions are fairly simple and are just wrapper around other functions for convenience. I have done a basic search of the web but I don't seem to be able to find a definitive answer.
Any help is appreciated.
It depends what you mean by "have to". One interpretation is, "Do I have to document these functions to pass checks?" The answer to that question is no. As long as the function isn't exported from the package, R CMD check won't require that you document it.
Another interpretation is "Do I have to document it to help myself in maintaining this package?" That question is harder to answer. Some functions are so obvious that they don't really need any documentation beyond their name, e.g. a print method with no extra arguments beyond those of the generic.
Other functions aren't so obvious, or have arguments whose meaning isn't obvious. It's a good idea to document those if you plan to maintain your package for a long time, because you might forget the details between now and whenever a problem arises. And if you are releasing your package to others, you should plan on long term maintenance, because if it is useful, people will use it.
I need to know if the data for training that is passed in the neuralnet call is randomized in the routine or does the routine uses the data in the same order that is given. I really need to know this info for a project that I am working on, and I have not being able to figure it out by looking at the source.
Thnx!
Look into the code - thats one of the most important advantages of FOSS: you can actually check what it is doing (neuralnet is pure R, so you don't even need to fear that you need to dig into FORTRAN or C code, and you can use debug to step through the code with example data to get an overview).
Moreover, if necessary, you can even introduce e.g. a new parameter that allows you to switch off randomization if needed.
Possibly maintainer ("neuralnet") would be willing to help you as well (and able to answer much faster than about everyone else here on SE).
There are many nice things to like about Makefiles, and many pains in the butt.
In the course of doing various project (I'm a research scientist, "data scientist", or whatever) I often find myself starting out with a few data objects on disk, generating various artifacts from those, generating artifacts from those artifacts, and so on.
It would be nice if I could just say "this object depends on these other objects", and "this object is created in the following manner from these objects", and then ask a Make-like framework to handle the details of actually building them, figuring out which objects need to be updated, farming out work to multiple processors (like Make's -j option), and so on. Makefiles can do all this - but the huge problem is that all the actions have to be written as shell commands. This is not convenient if I'm working in R or Perl or another similar environment. Furthermore, a strong assumption in Make is that all targets are files - there are some exceptions and workarounds, but if my targets are e.g. rows in a database, that would be pretty painful.
To be clear, I'm not after a software-build system. I'm interested in something that (more generally?) deals with dependency webs of artifacts.
Anyone know of a framework for these kinds of dependency webs? Seems like it could be a nice tool for doing data science, & visually showing how results were generated, etc.
One extremely interesting example I saw recently was IncPy, but it looks like it hasn't been touched in quite a while, and it's very closely coupled with Python. It's probably also much more ambitious than I'm hoping for, which is why it has to be so closely coupled with Python.
Sorry for the vague question, let me know if some clarification would be helpful.
A new system called "Drake" was announced today that targets this exact situation: http://blog.factual.com/introducing-drake-a-kind-of-make-for-data . Looks very promising, though I haven't actually tried it yet.
This question is several years old, but I thought adding a link to remake here would be relevant.
From the GitHub repository:
The idea here is to re-imagine a set of ideas from make but built for R. Rather than having a series of calls to different instances of R (as happens if you run make on R scripts), the idea is to define pieces of a pipeline within an R session. Rather than being language agnostic (like make must be), remake is unapologetically R focussed.
It is not on CRAN yet, and I haven't tried it, but it looks very interesting.
I would give Bazel a try for this. It is primarily a software build system, but with its genrule type of artifacts it can perform pretty arbitrary file generation, too.
Bazel is very extendable, using its Python-like Starlark language which should be far easier to use for complicated tasks than make. You can start by writing simple genrule steps by hand, then refactor common patterns into macros, and if things become more complicated even write your own rules. So you should be able to express your individual transformations at a high level that models how you think about them, then turn that representation into lower level constructs using something that feels like a proper programming language.
Where make depends on timestamps, Bazel checks fingerprints. So if at any one step produces the same output even though one of its inputs changed, then subsequent steps won't need to get re-computed again. If some of your data processing steps project or filter data, there might be a high probability of this kind of thing happening.
I see your question is tagged for R, even though it doesn't mention it much. Under the hood, R computations would in Bazel still boil down to R CMD invocations on the shell. But you could have complicated muliti-line commands assembled in complicated ways, to read your inputs, process them and store the outputs. If the cost of initialization of the R binary is a concern, Rserve might help although using it would make the setup depend on a locally accessible Rserve instance I believe. Even with that I see nothing that would avoid the cost of storing the data to file, and loading it back from file. If you want something that avoids that cost by keeping things in memory between steps, then you'd be looking into a very R-specific tool, not a generic tool like you requested.
In terms of “visually showing how results were generated”, bazel query --output graph can be used to generate a graphviz dot file of the dependency graph.
Disclaimer: I'm currently working at Google, which internally uses a variant of Bazel called Blaze. Actually Bazel is the open-source released version of Blaze. I'm very familiar with using Blaze, but not with setting up Bazel from scratch.
Red-R has a concept of data flow programming. I have not tried it yet.
Short version
For those that don't want to read through my "case", this is the essence:
What is the recommended way of minimizing the chances of new packages breaking existing code, i.e. of making the code you write as robust as possible?
What is the recommended way of making the best use of the namespace mechanism when
a) just using contributed packages (say in just some R Analysis Project)?
b) with respect to developing own packages?
How best to avoid conflicts with respect to formal classes (mostly Reference Classes in my case) as there isn't even a namespace mechanism comparable to :: for classes (AFAIU)?
The way the R universe works
This is something that's been nagging in the back of my mind for about two years now, yet I don't feel as if I have come to a satisfying solution. Plus I feel it's getting worse.
We see an ever increasing number of packages on CRAN, github, R-Forge and the like, which is simply terrific.
In such a decentralized environment, it is natural that the code base that makes up R (let's say that's base R and contributed R, for simplicity) will deviate from an ideal state with respect to robustness: people follow different conventions, there's S3, S4, S4 Reference Classes, etc. Things can't be as "aligned" as they would be if there were a "central clearing instance" that enforced conventions. That's okay.
The problem
Given the above, it can be very hard to use R to write robust code. Not everything you need will be in base R. For certain projects you will end up loading quite a few contributed packages.
IMHO, the biggest issue in that respect is the way the namespace concept is put to use in R: R allows for simply writing the name of a certain function/method without explicitly requiring it's namespace (i.e. foo vs. namespace::foo).
So for the sake of simplicity, that's what everyone is doing. But that way, name clashes, broken code and the need to rewrite/refactor your code are just a matter of time (or of the number of different packages loaded).
At best, you will know about which existing functions are masked/overloaded by a newly added package. At worst, you will have no clue until your code breaks.
A couple of examples:
try loading RMySQL and RSQLite at the same time, they don't go along very well
also RMongo will overwrite certain functions of RMySQL
forecast masks a lot of stuff with respect to ARIMA-related functions
R.utils even masks the base::parse routine
(I can't recall which functions in particular were causing the problems, but am willing to look it up again if there's interest)
Surprisingly, this doesn't seem to bother a lot of programmers out there. I tried to raise interest a couple of times at r-devel, to no significant avail.
Downsides of using the :: operator
Using the :: operator might significantly hurt efficiency in certain contexts as Dominick Samperi pointed out.
When developing your own package, you can't even use the :: operator throughout your own code as your code is no real package yet and thus there's also no namespace yet. So I would have to initially stick to the foo way, build, test and then go back to changing everything to namespace::foo. Not really.
Possible solutions to avoid these problems
Reassign each function from each package to a variable that follows certain naming conventions, e.g. namespace..foo in order to avoid the inefficiencies associated with namespace::foo (I outlined it once here). Pros: it works. Cons: it's clumsy and you double the memory used.
Simulate a namespace when developing your package. AFAIU, this is not really possible, at least I was told so back then.
Make it mandatory to use namespace::foo. IMHO, that would be the best thing to do. Sure, we would lose some extend of simplicity, but then again the R universe just isn't simple anymore (at least it's not as simple as in the early 00's).
And what about (formal) classes?
Apart from the aspects described above, :: way works quite well for functions/methods. But what about class definitions?
Take package timeDate with it's class timeDate. Say another package comes along which also has a class timeDate. I don't see how I could explicitly state that I would like a new instance of class timeDate from either of the two packages.
Something like this will not work:
new(timeDate::timeDate)
new("timeDate::timeDate")
new("timeDate", ns="timeDate")
That can be a huge problem as more and more people switch to an OOP-style for their R packages, leading to lots of class definitions. If there is a way to explicitly address the namespace of a class definition, I would very much appreciate a pointer!
Conclusion
Even though this was a bit lengthy, I hope I was able to point out the core problem/question and that I can raise more awareness here.
I think devtools and mvbutils do have some approaches that might be worth spreading, but I'm sure there's more to say.
GREAT question.
Validation
Writing robust, stable, and production-ready R code IS hard. You said: "Surprisingly, this doesn't seem to bother a lot of programmers out there". That's because most R programmers are not writing production code. They are performing one-off academic/research tasks. I would seriously question the skillset of any coder that claims that R is easy to put into production. Aside from my post on search/find mechanism which you have already linked to, I also wrote a post about the dangers of warning. The suggestions will help reduce complexity in your production code.
Tips for writing robust/production R code
Avoid packages that use Depends and favor packages that use Imports. A package with dependencies stuffed into Imports only is completely safe to use. If you absolutely must use a package that employs Depends, then email the author immediately after you call install.packages().
Here's what I tell authors: "Hi Author, I'm a fan of the XYZ package. I'd like to make a request. Could you move ABC and DEF from Depends to Imports in the next update? I cannot add your package to my own package's Imports until this happens. With R 2.14 enforcing NAMESPACE for every package, the general message from R Core is that packages should try to be "good citizens". If I have to load a Depends package, it adds a significant burden: I have to check for conflicts every time I take a dependency on a new package. With Imports, the package is free of side-effects. I understand that you might break other people's packages by doing this. I think its the right thing to do to demonstrate a commitment to Imports and in the long-run it will help people produce more robust R code."
Use importFrom. Don't add an entire package to Imports, add only those specific functions that you require. I accomplish this with Roxygen2 function documentation and roxygenize() which automatically generates the NAMESPACE file. In this way, you can import two packages that have conflicts where the conflicts aren't in the functions you actually need to use. Is this tedious? Only until it becomes a habit. The benefit: you can quickly identify all of your 3rd-party dependencies. That helps with...
Don't upgrade packages blindly. Read the changelog line-by-line and consider how the updates will affect the stability of your own package. Most of the time, the updates don't touch the functions you actually use.
Avoid S4 classes. I'm doing some hand-waving here. I find S4 to be complex and it takes enough brain power to deal with the search/find mechanism on the functional side of R. Do you really need these OO feature? Managing state = managing complexity - leave that for Python or Java =)
Write unit tests. Use the testthat package.
Whenever you R CMD build/test your package, parse the output and look for NOTE, INFO, WARNING. Also, physically scan with your own eyes. There's a part of the build step that notes conflicts but doesn't attach a WARN, etc. to it.
Add assertions and invariants right after a call to a 3rd-party package. In other words, don't fully trust what someone else gives you. Probe the result a little bit and stop() if the result is unexpected. You don't have to go crazy - pick one or two assertions that imply valid/high-confidence results.
I think there's more but this has become muscle memory now =) I'll augment if more comes to me.
My take on it :
Summary : Flexibility comes with a price. I'm willing to pay that price.
1) I simply don't use packages that cause that kind of problems. If I really, really need a function from that package in my own packages, I use the importFrom() in my NAMESPACE file. In any case, if I have trouble with a package, I contact the package author. The problem is at their side, not R's.
2) I never use :: inside my own code. By exporting only the functions needed by the user of my package, I can keep my own functions inside the NAMESPACE without running into conflicts. Functions that are not exported won't hide functions with the same name either, so that's a double win.
A good guide on how exactly environments, namespaces and the likes work you find here:
http://blog.obeautifulcode.com/R/How-R-Searches-And-Finds-Stuff/
This definitely is a must-read for everybody writing packages and the likes. After you read this, you'll realize that using :: in your package code is not necessary.
I am working on some code coverage for my applications. Now, I know that code coverage is an activity linked to the type of tests that you create and the language for which you wish to do the code coverage.
My question is: Is there any possible way to do some generic code coverage? Like in, can we have a set of features/test cases, which can be run (along with a lot more specific tests for the application under test) to get the code coverage for say 10% or more of the code?
More like, if I wish to build a framework for code coverage, what is the best possible way to go about making a generic one? Is it possible to have some functionality automated or generalized?
I'm not sure that generic coverage tools are the holy grail, for a couple of reasons:
Coverage is not a goal, it's an instrument. It tells you which parts of the code are not entirely hit by a test. It does not say anything about how good the tests are.
Generated tests can not guess the semantics of your code. Frameworks that generate tests for you only can deduct meaning from reading your code, which in essence could be wrong, because the whole point of unittesting is to see if the code actually behaves like you intended it too.
Because the automated framework will generate artificial coverage, you can never tell wether a piece of code is tested with a proper unittest, or superficially tested by a framework. I'd rather have untested code show up as uncovered, so I fix that.
What you could do (and I've done ;-) ) is write a generic test for testing Java beans. By reflection, you can test a Java bean against the Sun spec of a Java bean. Assert that equals and hashcode are both implemented (or neither of them), see that the getter actually returns the value you pushed in with the setter, check wether all properties have getters and setters.
You can do the same basic trick for anything that implements "comparable" for instance.
It's easy to do, easy to maintain and forces you to have clean beans. As for the rest of the unittests, I try to focus on getting important parts tested first and thouroughly.
Coverage can give a false sense of security. Common sense can not be automated.
This is usually achieved by combining static code analysis (Coverity, Klockwork or their free analogs) with dynamic analysis by running a tests against instrumented application (profiler + memory checker). Unfortunately, this is hard to automate test algorythms, most tools are kind of "recorders" able to record traffic/keys/signals - depending on domain and replay them (with minimal changes/substitutions like session ID/user/etc)