NEWS.md
l0 has been removed. Now vector targets receiving something 0-length in input will not fail (in the previous version, l0 was required). Vector length can still be checked with the len(a,b) mechanism.New set of hook error and warning functions for reporting in complex internal functions. These are set_hook to set a function as the hook and stop_hook/warn_hook error and warning functions used to report as if they were called from the hooked function.
warni and stopi for warning and stop functions with character interpolation using stringmagic
check_arg_plus and check_value_plus become check_set_arg and check_set_value to improve clarity.fsignif now displays trailing 0s appropriately and is neutral to character vectors (instead of throwing an error).
the comparison types now can evaluate values from the calling frame:
z = 32
x = 77
try(check_value(x, "numeric scalar LE{z}"))
#> Error: in check_value(x, "numeric scalar LE{z}"):
#> Value 'x' must be a numeric scalar lower than, or equal to, 32. Problem: it is strictly greater than 32.stop_up now accepts the argument msg which is an extra message displayed right after the error message.NA tolerance for the main classes vector and matrix (opposite to the previous behavior). The new keywords NO NA replace the former keywords NA OK for these classes.test = function(x){
check_arg(x, "numeric vector")
}
# Before (version <= 1.1.0)
test(c(5, NA)) # => Error
# Now (version >= 1.2.0)
test(c(5, NA)) # => is OKlogical type is requested.test = function(x){
check_arg(x, "logical scalar")
}
# Before (version <= 1.1.0)
test(0) # => is OK
# Now (version >= 1.2.0)
test(0) # => Error, must be logicalThe new behavior is equivalent to the previous behavior when the type was strict logical. Now strict logical has been removed and loose logical has been introduced which accepts 0 and 1. Why this change? I remarked that when you allow arguments of very different types, one of which was a logical scalar, it was really useful to distinguish between them safely using isTRUE or is.logical. With the previous default behavior you could have a 0 or 1 that would prevent you from doing that. Of course you could use strict logical to trigger that behavior but it was non-intuitive.
In check_value: you can use the new argument .prefix to design the error message.
Function check_value_plus now returns the value of the element it has checked and possibly modified.
x = "t"
y = check_value_plus(x, "match(This, Is, Raw)")
y # => "This"
fml = ~head(Petal.Length)
z = check_value_plus(fml[[2]], "evalset numeric vector", .data = iris)
zsfill now accepts NAs.Error messages in check_value have been improved.
enumerate_items now returns the empty string when 0-length vectors are in input.
The main class match could lead the argument to lose its attributes (like names), now corrected.
Vectors containing only NAs when NAs are tolerated could lead to a wrong error message, now corrected.
Bug when check_value_plus was called from the global environment and an error was triggered.
Bug when check_value was called and .arg_name was provided.
Bug when the “...” argument contained NULL values.
The developer mode now never applies to functions from packages using dreamerr
Behavior of argument nmax in enumerate_items is fixed.
Type conversion of matrices returned a vector, now corrected.
sfill: formatting function for strings. Fills a character string up to the required length. Helps to form nice messages.
set_up: sets the argument up semi-globally (i.e. throughout all calls within a function).
set_check: turns on/off argument checking semi-globally (i.e. throughout all calls within a function).
.message), you can use the special character __ARG__. If found, it will be replaced by the appropriate argument name.Message informing no NA tolerance in the types is prompted only when the error comes from NA values.
The developer mode catches more errors and provide more suggestions.
check_value now doesn’t throw an error for missing .x.
Small bug that could occur for the types "integer scalar na ok" with the argument equal to NA.
Important bug when using features requiring evaluations in the environment. Now the default environment is set up appropriately.
Bug when using var(data,env) in a formula class with missing data.
Very sneaky bug when badly forming a call to check_arg to check ... in nested functions is now caught.
In enumerate_items, when there were many elements, quote = TRUE led to quote the XX others that shouldn’t be quoted. Now corrected.
In the formula class, if a variable was passed as a Formula (notice the capital F), this caused an error if one-sidedness or two-sidedness were to be checked.
Bug when in n_letter when negative numbers are in. This bug propagated into check_arg.
This package is the outcome of over a decade of coding and developing packages in R. As a package developer I always wanted my code to be “safe”: so that if a user provides arguments of the good type the functions always work, and if there’s an argument of the wrong type then an informative error message is shown.
The big problem is that error handling is extremely time consuming, especially if you want to specify informative error messages stating clearly where the problem comes from. This problem is compounded when you want to offer the user flexible arguments that can be of many types. To do error handling properly in such situations is a nightmare: you end up with big if/else trees with a lot of different messages. And it doesn’t improve the speed of your code! So you have little incentives to invest time in proper error handling.
This is why I developed this function. It’s one line of code, and it does all the work for you. From the developer side it’s super simple, and from the user side, s/he ends up with extremely informative error messages therefore saving a lot of time in debugging.