Trust, but Verify

Description

Declarative template-based framework for verifying that objects meet structural requirements, and auto-composing error messages when they do not.

Turn S3 Objects Into Templates

Description

Create templates for use by alike. Currently somewhat experimental; behavior may change in future.

Usage

abstract(x, ...)

## S3 method for class 'data.frame'
abstract(x, ...)

## Default S3 method:
abstract(x, ...)

## S3 method for class 'array'
abstract(x, ...)

## S3 method for class 'matrix'
abstract(x, ...)

## S3 method for class 'list'
abstract(x, ...)

## S3 method for class 'lm'
abstract(x, ...)

## S3 method for class 'environment'
abstract(x, ...)

## S3 method for class 'ts'
abstract(x, what = c("start", "end", "frequency"), ...)

Arguments

Details

abstract is intended to create templates for use by alike. The result of abstraction is often a partially specified object. This type of object may not be suited for use in typical R computations and may cause errors (or worse) if you try to use them as normal R objects.

There is no guarantee that the abstracted object is suitable for use as a template to alike as is. You may need to modify it further so that it suits your purposes.

abstract is an S3 generic. The default method will dispatch on implicit classes, so if you attempt to abstract an object without an explicit abstract method, it will get abstracted based on its implicit class. If you define your own abstract method and do not wish further abstraction based on implicit classes do not use NextMethod.

S4 and RC objects are returned unchanged.

Value

abstracted object

Time Series

alike will treat time series parameter components with zero in them as wildcards. This function allows you to create these wild card time series attributes since R does not allow direct creation/modification of ts attributes with zero values.

Make sure you do not try to use the templates you create with this for anything other than as alike templates since the result is likely undefined given R expects non zero values for the ts attribute and attempts to prevent such attributes.

Examples

iris.tpl <- abstract(iris)
alike(iris.tpl, iris[1:10, ])
alike(iris.tpl, transform(iris, Species=as.character(Species)))

abstract(1:10)
abstract(matrix(1:9, nrow=3))
abstract(list(1:9, runif(10)))

Experimental Abstraction Method for GGPlot

Description

Not entirely sure this can ever work well since so much of ggplot is done with proto objects and those do not really use meta data, which makes alike rather useless.

Usage

## S3 method for class 'ggplot'
abstract(x, ...)

Compare Object Structure

Description

Similar to all.equal, but compares object structure rather than value. The target argument defines a template that the current argument must match.

Usage

alike(target, current, env = parent.frame(), settings = NULL)

Arguments

Value

TRUE if target and current are alike, character(1L) describing why they are not if they are not

alikeness

Generally speaking two objects are alike if they are of the same type (as determined by type_alike) and length. Attributes on the objects are required to be recursively alike, though the following attributes are treated specially: class, dim, dimnames, names, row.names, levels, tsp, and srcref.

Exactly what makes two objects alike is complex, but should be intuitive. The best way to understand "alikeness" is to review the examples. For a thorough exposition see the vignette.

Note that the semantics of alikeness for language objects, formulas, and functions may change in the future.

See Also

type_alike, type_of, abstract, vetr_settings for more control of settings

Examples

## Type comparison
alike(1L, 1.0)         # TRUE, because 1.0 is integer-like
alike(1L, 1.1)         # FALSE, 1.1 is not integer-like
alike(1.1, 1L)         # TRUE, by default, integers are always considered real

alike(1:100, 1:100 + 0.0)  # TRUE

## We do not check numerics for integerness if longer than 100
alike(1:101, 1:101 + 0.0)

## Scalarness can now be checked at same time as type
alike(integer(1L), 1)            # integer-like and length 1?
alike(logical(1L), TRUE)         # logical and length 1?
alike(integer(1L), 1:3)
alike(logical(1L), c(TRUE, TRUE))

## Zero length match any length of same type
alike(integer(), 1:10)
alike(1:10, integer())   # but not the other way around

## Recursive objects compared recursively
alike(
  list(integer(), list(character(), logical(1L))),
  list(1:10, list(letters, TRUE))
)
alike(
  list(integer(), list(character(), logical(1L))),
  list(1:10, list(letters, c(TRUE, FALSE)))
)

## `NULL` is a wild card when nested within recursive objects
alike(list(NULL, NULL), list(iris, mtcars))
alike(NULL, mtcars)    # but not at top level

## Since `data.frame` are lists, we can compare them recursively:
iris.fake <- transform(iris, Species=as.character(Species))
alike(iris, iris.fake)

## we even check attributes (factor levels must match)!
iris.fake2 <- iris
levels(iris.fake2$Species) <- c("setosa", "versicolor", "africana")
alike(iris, iris.fake2)

## We can use partially specified objects as templates
iris.tpl <- abstract(iris)
str(iris.tpl)
alike(iris.tpl, iris)
## any row sample of iris matches our iris template
alike(iris.tpl, iris[sample(1:nrow(iris), 10), ])
## but column order matters
alike(iris.tpl, iris[c(2, 1, 3, 4, 5)])

## 3 x 3 integer
alike(matrix(integer(), 3, 3), matrix(1:9, nrow=3))
## 3 x 3, but not integer!
alike(matrix(integer(), 3, 3), matrix(runif(9), nrow=3))
## partial spec, any 3 row integer matrix
alike(matrix(integer(), 3), matrix(1:12, nrow=3))
alike(matrix(integer(), 3), matrix(1:12, nrow=4))
## Any logical matrix (but not arrays)
alike(matrix(logical()), array(rep(TRUE, 8), rep(2, 3)))

## In order for objects to be alike, they must share a family
## tree, not just a common class
obj.tpl <- structure(TRUE, class=letters[1:3])
obj.cur.1 <-  structure(TRUE, class=c("x", letters[1:3]))
obj.cur.2 <-  structure(TRUE, class=c(letters[1:3], "x"))

alike(obj.tpl, obj.cur.1)
alike(obj.tpl, obj.cur.2)

## You can compare language objects; these are alike if they are self
## consistent; we don't care what the symbols are, so long as they are used
## consistently across target and current:

## TRUE, symbols are consistent (adding two different symbols)
alike(quote(x + y), quote(a + b))
## FALSE, different function
alike(quote(x + y), quote(a - b))
## FALSE, inconsistent symbols
alike(quote(x + y), quote(a + a))

Verify Values in Vector are Between Two Others

Description

Similar to isTRUE(all(x >= lo & x <= hi)) with default settings, except that it is substantially faster and returns a string describing the first encountered violation rather than FALSE on failure.

Usage

all_bw(x, lo = -Inf, hi = Inf, na.rm = FALSE, bounds = "[]")

Arguments

Details

You can modify the comparison to be strictly greater/less than via the bounds parameter, and the treatment of NAs with na.rm. Note that NAs are considered to be out of bounds by default. While technically incorrect since we cannot know whether an NA value is in or out of bounds, this assumption is both conservative and convenient. Zero length x will always succeed.

If x and lo/hi are different types, lo/hi will be coerced to the type of x. When lo/hi are numeric and x is integer, if lo/hi values are outside of the integer range then that side will be treated as if you had used -Inf/Inf. -Inf and Inf mean lo and hi will be unbounded for all data types.

Value

TRUE if all values in x conform to the specified bounds, a string describing the first position that fails otherwise

Examples

all_bw(runif(100), 0, 1)
all_bw(runif(100) * 2, 0, 1)
all_bw(NA, 0, 1)              # This is does not return NA
all_bw(NA, 0, 1, na.rm=TRUE)

vec <- c(runif(100, 0, 1e12), Inf, 0)
all_bw(vec, 0)      # All +ve numbers
all_bw(vec, hi=0)   # All -ve numbers
all_bw(vec, 0, bounds="(]") # All strictly +ve nums
all_bw(vec, 0, bounds="[)") # All finite +ve nums

Lightweight Benchmarking Function

Description

Evaluates provided expression in a loop and reports mean evaluation time. This is inferior to microbenchmark and other benchmarking tools in many ways except that it has zero dependencies or suggests which helps with package build and test times. Used in vignettes.

Usage

bench_mark(..., times = 1000L, deparse.width = 40)

Arguments

Details

Runs gc() before each expression is evaluated. Expressions are evaluated in the order provided. Attempts to estimate the overhead of the loop by running a loop that evaluates NULL the times times.

Unfortunately because this computes the average of all iterations it is very susceptible to outliers in small sample runs, particularly with fast running code. For that reason the default number of iterations is one thousand.

Value

NULL, invisibly, reports timings as a side effect as screen output

Examples

bench_mark(runif(1000), Sys.sleep(0.001), times=10)

Set Element to NULL Without Removing It

Description

This function is required because there is no straightforward way to over-write a value in a list with NULL without completely removing the entry from the list as well.

Usage

nullify(obj, index)

## Default S3 method:
nullify(obj, index)

Arguments

Details

This returns a copy of the object modified with null slots; it does not modify the input argument.

Default method will attempt to convert non-list objects to lists with as.list, and then back to whatever they were by using a function with name paste0("as.", class(obj)[[1L]]) if it exists and works. If the object cannot be coerced back to its original type the corresponding list will be returned.

If this is not appropriate for your object type you can write an S3 method for it.

Value

object with selected values NULLified

Note

attributes are copied from original object and re-applied to final object before return, which may not make sense in some circumstances.

Examples

nullify(list(1, 2, 3), 2)
nullify(call("fun", 1, 2, 3), 2)

Compare Types of Objects

Description

By default, checks type_of() objects and two objects are considered type_alike if they have the same type. There is special handling for integers, numerics, and functions.

Usage

type_alike(target, current, settings = NULL)

Arguments

Details

For integers and numerics, if current is integer or integer-like (e.g. 1.0) it will match real or integer target values. Closures, built-ins, and specials are all treated as type function.

Specific behavior can be tuned with the type.mode parameter to the vetr_settings() object passed as the settings parameter to this function.

See Also

type_of, alike, vetr_settings(), in particular the section about the type.mode parameter which affects how this function behaves.

A Fuzzier Version of typeof()

Description

Numerics that are equivalent to integers (e.g x == floor(x)) are classified as integers, and builtin and special functions are reported as closures.

Usage

type_of(object)

Arguments

Value

character(1L) the type of the object

Examples

type_of(1.0001)          # numeric
type_of(1.0)             # integer (`typeof` returns numeric)
type_of(1)               # integer (`typeof` returns numeric)
type_of(sum)             # closure (`typeof` returns builtin)
type_of(`$`)             # closure (`typeof` returns special)

Verify Objects Meet Structural Requirements

Description

Use vetting expressions to enforce structural requirements for objects. tev is a version of vet compatible with magrittr pipes.

Usage

vet(
  target,
  current,
  env = parent.frame(),
  format = "text",
  stop = FALSE,
  settings = NULL
)

tev(
  current,
  target,
  env = parent.frame(),
  format = "text",
  stop = FALSE,
  settings = NULL
)

Arguments

Details

tev just reverses the target and current arguments for better integration with magrittr. There are two major caveats:

• error messages will be less useful since you will get . instead of the deparsed call

• ⁠x \\%>\\% tev(y)⁠ is much slower than vet(y, x) (or even tev(x, y))

Value

TRUE if validation succeeds, otherwise varies according to value chosen with parameter stop

Vetting Expressions

Vetting expressions can be template tokens, standard tokens, or any combination of template and standard tokens combined with && and/or ||. Template tokens are R objects that define the required structure, much like the FUN.VALUE argument to vapply(). Standard tokens are tokens that contain the . symbol and are used to vet values.

If you do use the . symbol in your vetting expressions in your packages, you will need to include utils::globalVariables(".") as a top-level call to avoid the "no visible binding for global variable '.'"' R CMD check NOTE.

See vignette('vetr', package='vetr') and examples for details on how to craft vetting expressions.

See Also

vetr() for a version optimized to vet function arguments, alike() for how templates are used, vet_token() for how to specify custom error messages and also for predefined validation tokens for common use cases, all_bw() for fast bounds checks.

Examples

## template vetting
vet(numeric(2L), runif(2))
vet(numeric(2L), runif(3))
vet(numeric(2L), letters)
try(vet(numeric(2L), letters, stop=TRUE))

## `tev` just reverses target and current for use with maggrittr
## Not run: 
if(require(magrittr)) {
  runif(2) %>% tev(numeric(2L))
  runif(3) %>% tev(numeric(2L))
}

## End(Not run)
## Zero length templates are wild cards
vet(numeric(), runif(2))
vet(numeric(), runif(100))
vet(numeric(), letters)

## This extends to data.frames
iris.tpl <- iris[0,]   # zero row matches any # of rows
iris.1 <- iris[1:10,]
iris.2 <- iris[1:10, c(1,2,3,5,4)]  # change col order
vet(iris.tpl, iris.1)
vet(iris.tpl, iris.2)

## Short (<100 length) integer-like numerics will
## pass for integer
vet(integer(), c(1, 2, 3))
vet(integer(), c(1, 2, 3) + 0.1)

## Nested templates; note, in packages you should consider
## defining templates outside of `vet` or `vetr` so that
## they are computed on load rather that at runtime
tpl <- list(numeric(1L), matrix(integer(), 3))
val.1 <- list(runif(1), rbind(1:10, 1:10, 1:10))
val.2 <- list(runif(1), cbind(1:10, 1:10, 1:10))
vet(tpl, val.1)
vet(tpl, val.2)

## See `example(alike)` for more template examples

## Standard tokens allow you to check values
vet(. > 0, runif(10))
vet(. > 0, -runif(10))

## Zero length token results are considered TRUE,
## as is the case with `all(logical(0))`
vet(. > 0, numeric())

## `all_bw` is like `isTRUE(all(. >= x & . <= y))`, but
## ~10x faster for long vectors:
vet(all_bw(., 0, 1), runif(1e6) + .1)

## You can combine templates and standard tokens with
## `&&` and/or `||`
vet(numeric(2L) && . > 0, runif(2))
vet(numeric(2L) && . > 0, runif(10))
vet(numeric(2L) && . > 0, -runif(2))

## Using pre-defined tokens (see `?vet_token`)
vet(INT.1, 1)
vet(INT.1, 1:2)
vet(INT.1 && . %in% 0:1 || LGL.1, TRUE)
vet(INT.1 && . %in% 0:1 || LGL.1, 1)
vet(INT.1 && . %in% 0:1 || LGL.1, NA)

## Vetting expressions can be assembled from previously
## defined tokens
scalar.num.pos <- quote(numeric(1L) && . > 0)
foo.or.bar <- quote(character(1L) && . %in% c('foo', 'bar'))
vet.exp <- quote(scalar.num.pos || foo.or.bar)

vet(vet.exp, 42)
vet(scalar.num.pos || foo.or.bar, 42)  # equivalently
vet(vet.exp, "foo")
vet(vet.exp, "baz")

Vetting Tokens With Custom Error Messages

Description

Utility function to generate vetting tokens with attached error messages. You should only need to use this if the error message produced naturally by vetr is unclear. Several predefined tokens created by this function are also documented here.

Usage

vet_token(exp, err.msg = "%s")

NO.NA

NO.INF

GTE.0

LTE.0

GT.0

LT.0

INT.1

INT.1.POS

INT.1.NEG

INT.1.POS.STR

INT.1.NEG.STR

INT

INT.POS

INT.NEG

INT.POS.STR

INT.NEG.STR

NUM.1

NUM.1.POS

NUM.1.NEG

NUM

NUM.POS

NUM.NEG

CHR.1

CHR

CPX

CPX.1

LGL

LGL.1

Arguments

Format

An object of class call of length 2.

An object of class call of length 2.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

An object of class call of length 3.

Details

Allows you to supply error messages for vetting to use for each error token. Your token should not contain top level && or ||. If it does your error message will not be reported because vetr looks for error messages attached to atomic tokens. If your token must involve top level && or ||, use I(x && y) to ensure that your error message is used by vet, but beware than in doing so you do not use templates within the I call as everything therein will be interpreted as a vetting expression rather than a template.

Error messages are typically of the form "%s should be XXX".

This package ships with many predefined tokens for common use cases. They are listed in the Usage section of this documentation. The tokens are named in format TYPE[.LENGTH][.OTHER]. For example INT will vet an integer vector, INT.1 will vet a scalar integer vector, and INT.1.POS.STR will vet a strictly positive integer vector. At this time tokens are predefined for the basic types as scalars or any-length vectors. Some additional checks are available (e.g. positive only values).

Every one of the predefined vetting tokens documented here implicitly disallows NAs. Numeric tokens also disallow infinite values. If you wish to allow NAs or infinite values just use a template object (e.g. integer(1L)).

Value

a quoted expressions with err.msg attribute set

Note

This will only work with standard tokens containing .. Anything else will be interpreted as a template token.

See Also

vet()

Examples

## Predefined tokens:
vet(INT.1, 1:2)
vet(INT.1 || LGL, 1:2)
vet(INT.1 || LGL, c(TRUE, FALSE))

## Check squareness
mx <- matrix(1:3)
SQR <- vet_token(nrow(.) == ncol(.), "%s should be square")
vet(SQR, mx)

## Let `vetr` make up error message; note `quote` vs `vet_token`
## Often, `vetr` does fine without explictly specified err msg:
SQR.V2 <- quote(nrow(.) == ncol(.))
vet(SQR.V2, mx)

## Combine some tokens, notice how we use `quote` at the combining
## step:
NUM.MX <- vet_token(matrix(numeric(), 0, 0), "%s should be numeric matrix")
SQR.NUM.MX <- quote(NUM.MX && SQR)
vet(SQR.NUM.MX, mx)

## If instead we used `vet_token` the overall error message
## is not used; instead it falls back to the error message of
## the specific sub-token that fails:
NUM.MX <- vet_token(matrix(numeric(), 0, 0), "%s should be numeric matrix")
SQR.NUM.MX.V2 <-
  vet_token(NUM.MX && SQR, "%s should be a square numeric matrix")
vet(SQR.NUM.MX.V2, mx)

Verify Function Arguments Meet Structural Requirements

Description

Use vetting expressions to enforce structural requirements for function arguments. Works just like vet(), except that the formals of the enclosing function automatically matched to the vetting expressions provided in ....

Usage

vetr(..., .VETR_SETTINGS = NULL)

Arguments

Details

Only named arguments may be vetted; in other words it is not possible to vet arguments passed via ....

Value

TRUE if validation succeeds, otherwise stop with error message detailing nature of failure.

Vetting Expressions

Vetting expressions can be template tokens, standard tokens, or any combination of template and standard tokens combined with && and/or ||. Template tokens are R objects that define the required structure, much like the FUN.VALUE argument to vapply(). Standard tokens are tokens that contain the . symbol and are used to vet values.

If you do use the . symbol in your vetting expressions in your packages, you will need to include utils::globalVariables(".") as a top-level call to avoid the "no visible binding for global variable '.'"' R CMD check NOTE.

See vignette('vetr', package='vetr') and examples for details on how to craft vetting expressions.

Note

vetr will force evaluation of any arguments that are being checked (you may omit arguments that should not be evaluate from vetr)

See Also

vet(), in particular example(vet).

Examples

fun1 <- function(x, y) {
  vetr(integer(), LGL.1)
  TRUE   # do some work
}
fun1(1:10, TRUE)
try(fun1(1:10, 1:10))

## only vet the second argument
fun2 <- function(x, y) {
  vetr(y=LGL.1)
  TRUE   # do some work
}
try(fun2(letters, 1:10))

## Nested templates; note, in packages you should consider
## defining templates outside of `vet` or `vetr` so that
## they are computed on load rather that at runtime
tpl <- list(numeric(1L), matrix(integer(), 3))
val.1 <- list(runif(1), rbind(1:10, 1:10, 1:10))
val.2 <- list(runif(1), cbind(1:10, 1:10, 1:10))
fun3 <- function(x, y) {
  vetr(x=tpl, y=tpl && ncol(.[[2]]) == ncol(x[[2]]))
  TRUE   # do some work
}
fun3(val.1, val.1)
try(fun3(val.1, val.2))
val.1.a <- val.1
val.1.a[[2]] <- val.1.a[[2]][, 1:8]
try(fun3(val.1, val.1.a))

Test Objects

Description

Objects used for testing purposes only.

Generate Control Settings For vetr and alike

Description

Utility function to generate setting values. We strongly recommend that you generate the settings outside of function calls so that setting generation does not become part of the vet/vetr/alike evaluation as that could add noticeable overhead to the function evaluation.

Usage

vetr_settings(
  type.mode = 0L,
  attr.mode = 0L,
  lang.mode = 0L,
  fun.mode = 0L,
  rec.mode = 0L,
  suppress.warnings = FALSE,
  fuzzy.int.max.len = 100L,
  width = -1L,
  env.depth.max = 65535L,
  symb.sub.depth.max = 65535L,
  symb.size.max = 15000L,
  nchar.max = 65535L,
  track.hash.content.size = 63L,
  env = NULL,
  result.list.size.init = 64L,
  result.list.size.max = 1024L
)

Arguments

Details

Settings after fuzzy.int.max.len are fairly low level and exposed mostly for testing purposes. You should generally not need to use them.

Note that a successful evaluation of this function does not guarantee a correct settings list. Those checks are carried out internally by vet/vetr/alike.

Value

list with all the setting values

See Also

type_alike, alike, vetr

Examples

type_alike(1L, 1.0, settings=vetr_settings(type.mode=2))
## better if you are going to re-use settings to reduce overhead
set <- vetr_settings(type.mode=2)
type_alike(1L, 1.0, settings=set)