Default Mode is to Store Tests in
unitizer stores unit tests and their results. By default, it stores them in
rds files in your filesystem. You will be prompted before a file is saved to your filesystem.
rds file is placed in a directory with the same name as your test file, but with "unitizer" appended. For example, if your tests are in "my_file_name.R", then
unitizer will create a folder called "my_file_name.unitizer/" and put an
rds file in it.
?get_unitizer for potential alternatives to saving to your file system.
File Space Considerations
If your tests produce massive objects, the
rds file will be massive. Try designing your tests so they will produce the smallest representative data structures needed for your tests to be useful.
Additionally, note that the
rds files are binary, which needs to be accounted for when using them in version controlled projects.
unitizer does not backup the
rds beyond the single copy in the aforementioned folder. Unit tests are valuable, and without the
unitizer tests become a lot less useful. To the extent you backup your R test files, you should also backup the corresponding ".unitizer/" folder. You could lose / corrupt your
unitizer store in many ways. Some non-exhaustive examples:
- Standard file system SNAFU
- Careless updates to existing
unitizerdeveloper accidentally introduces a bug that destroys your
Alternate Store Locations
unitize stores and loads
unitizers using the
get_unitizer S3 generics . This means you can implement your own S3 methods for those generics to store the
unitizer object off-filesystem (e.g. MySQL databse, etc). See
?get_unitizer for more details, though note this feature is untested.
If you only wish to save your
unitizer to a different location in your filesystem than the default, you do not need to resort to these methods as you can provide the target directory with
Version Control and Unitizer
Committing Binary Files
The main issue with using
unitizer with a version controlled package is that
you have to decide whether you want to include the binary
rds files in the
version control history. Some options:
- do not track the binary files at all (but they are valuable and now not backed up)
- do not track the binary files at all, but implement a secondary back-up system (this sounds really annoying)
- Use a backed-up, non-file system store (see "Alternate Store Locations" above)
- Track the binary files, but manage how often they are committed.
We recommend splitting tests for different functionality into different files. This should mitigate the number of rds files that change with any given source code update, and is good practice anyway. Additionally, we typically only commit the rds files when a feature branch or issue resolution is fully complete.
Additionally a useful
git shortcut to add to your
.gitconfig file that
mitigates how often you commit rds files is:
[alias] ad = !git add -u && git reset -- *.rds
This makes it easy to add all the files you are working on except for the rdses. Once you have stabilized a set of tests you can commit the rds.
All this aside, remember that the rdses are ultimately just as important as the test files, and you should commit them occasionally to ensure you do not use valuable test information.
Collaborating with Unitizer
If you merge in a pull request from a third party you do not fully trust, we recommend you checkout your most recent version of the rdses that are part of the pull request. You can then re-run the tests and review what each of them does.
Modifying an Existing Unitizer
review allows you to review all tests in a unitizer rds with the option of dropping tests from it. See
Warning: this is experimental.
editCalls allows you to modify the calls calls stored in a
unitizer. This is useful when you decide to change the call (e.g. a function name), but otherwise leave the behavior of the call unchanged. You can then upate your test script and the renamed calls will be matched against the correct values in the
unitizer store. Without this you would have to re-review and re-store every test since
unitizer identifies tests by the deparsed call.
There is currently no direct way to split a
unitizer into pieces (see issue #44), but the current work around is to:
- copy the test file and the corresponding
unitizerto a new location
- edit the original test file to remove the tests we want to split off
- run unitizer and agree to drop all removed tests (hint: this is a good time to use
- edit the new test file and remove the tests that are still in the old test file
- run unitizer and agree to drop all removed tests
The net result will be two new
unitizer, each with a portion of the tests from the original
unitizer. Clearly less than ideal, but will work in a pinch.
unitizer Output No Longer Shows on Screen
stderr during test evaluation, so it is possible that in some corner cases
unitizer exits without releasing sinks. We have put substantial effort in trying to avoid this eventuality, but should it occur, here are some things you can do:
sink(type="message")to reset the output stream sinks
- or, restart the R session (type
q()followed by ENTER, then "y" or "n" (without quotes) depending on whether you want to save your workspace or not)
Either way, please contact the maintainer as this should not happen.
unitizer Freezes and Pops up "Selection:"
This is almost certainly a result of an R crash. Unfortunately the normal mechanisms to restore
stderr don't seem to work completely with full R crashes, so when you see things like:
+------------------------------------------------------------------------------+ | unitizer for: tests/unitizer/alike.R | +------------------------------------------------------------------------------+ Running: alike(data.frame(a = integer(), b = factor()), data.frame(a = 1:3, Selection:
what you are not seeing is:
*** caught segfault *** address 0x7fdc20000010, cause 'memory not mapped' Traceback: 1: .Call(ALIKEC_alike, target, current, int.mode, int.tol, attr.mode) 2: alike(data.frame(a = factor(), b = factor()), data.frame(a = 1:3, b = letters[1:3])) Possible actions: 1: abort (with core dump, if enabled) 2: normal R exit 3: exit R without saving workspace 4: exit R saving workspace
The "Selection:" bit is prompting you to type 1-4 as per above. We will investigate to see if there is a way to address this problem, but the solution likely is not simple since the R crash circumvents the
on.exit handlers used to reset the stream redirects. Also, note that in this case the crash is caused by
unitizer (see below).
unitizer Crashes R
Every R crash we have discovered while using
unitizer was eventually traced to a third party package. Some of the crashes were linked to issues attaching/detaching packages. If you think you might be having an issue with this you can always run with
clean.env=FALSE to avoid search path manipulation.
Different Outcomes in Interactive vs. Non Interactive
Watch out for functions that have default arguments of the type:
fun <- function(x, y=getOption('blahblah'))
as those options may be different depending on whether you are running whether you are running R interactively or not. One prime example is
parse(..., keep.source = getOption("keep.source")).
unitize Within Error Handling Blocks
unitize evaluates test expressions within a call to
withCallingHandlers, there are some limitations on successfully running
unitize inside your own error handling calls. In particular,
unitize will not work properly if run inside a
try statement. If test expressions throw conditions, the internal
withCallingHandlers will automatically hand over control to your
try statement without an opportunity to complete
unitize computations. Unfortunately there does not seem to be a way around this since we have to use
withCallingHandlers so that test statements after non-aborting conditions are run.
See this SO Q/A for more details on the problem.
In order to perpetuate the R console prompt illusion,
unitizer needs to override some buit-in functionality, including:
lsis replaced by a special version that can explore the
qare wrappers around the base functions that allow
unitizerto quit gracefully
tracebackwhile not replaced, is supported by a fair bit of voodoo involving among other things assigning to
- History is replaced during
unitizerprompt evaluations with a temporary version of the history file containing only commands evaluated at the
unitizerprompt. The normal history file is restored on exit.