better error messages

[ggrothendieck]

There are a number of places where Rcpp can fail because TYPEOF(x) does not equal RTYPE. Since both of these are known at the point that the error is given it would help if these were displayed, e.g. rather than just "some message" use: std:string("some message TYPEOF(x): ") + CHAR(Rf_type2str(TYPEOF(x))) + " RTYPE: " + CHAR(Rf_type2str(RTYPE))
At least that will give some bit of clue as to what the problem might be.

Any other info that might be added to the error message to help locate the problem could be helpful too.

This applies to RcppEigen too.

[eddelbuettel]

Maybe you could try this idea in a pull request?

[kevinushey]

@ggrothendieck is something like this what you imagined? I would welcome you to modify r_cast.h to provide more appropriate error messages and send us a PR for review.

[ggrothendieck]

Yes, that looks good. The important part is to actually modify all points where such an error can occur to ensure that the expanded info is provided.

[kevinushey]

Now that we have a better Rcpp::stop from @romainfrancois it would be sensible to revisit the Rcpp exceptions and use it to provide better error messages.

[eddelbuettel]

Sure. Do you think you may have spare cycles for that any time soon?

[coatless]

As this sort of came up on SO today and its a bit problematic for beginners...

I glanced at the inst/include/Rcpp/r_cast.h. Seems like the commit was regressed. I think @kevinushey wanted to roll out the backend of ::Rcpp::stop() that uses /inst/include/Rcpp/utils/tinyformat.h in place of the exception macro generating ::Rcpp::not_compatible().

To maintain using the same class not_compatible -- but extend it so that it can hold multiple parameters -- I'm thinking just append to the file:

    template <typename T1>
    inline void NORET not_compatible(const char* fmt, const T1& arg1) {
        throw Rcpp::not_compatible( tfm::format(fmt, arg1 ).c_str() );
    }

    template <typename T1, typename T2>
    inline void NORET not_compatible(const char* fmt, const T1& arg1, const T2& arg2) {
        throw Rcpp::not_compatible( tfm::format(fmt, arg1, arg2 ).c_str() );
    }

    template <typename T1, typename T2, typename T3>
    inline void NORET not_compatible(const char* fmt, const T1& arg1, const T2& arg2, const T3& arg3) {
        throw Rcpp::not_compatible( tfm::format(fmt, arg1, arg2, arg3 ).c_str() );
    }

...

Label as "Enhancement" + "API"?

[coatless]

@eddelbuettel @kevinushey Thoughts on this? I would like to hopefully roll something out before next week.

[kevinushey]

I'm definitely on board for commits / PRs that improve the error messages reported by Rcpp.

[eddelbuettel]

Sure. If it works in that case and doesn't create trouble otherwise...

[coatless]

Proposal

• Create a new RCPP_ADVANCED_EXCEPTION_CLASS exception generation macro to provide support for tfm::format(...) that mimics present implementation of Rcpp::stop() and Rcpp::warning().
• Improve existing error throws by adding details on object structure (see list below)
• Remove error codes not used

Enable Parameter Support / tfm::format(...)

Exception Class	Current	Proposed	Rationale
no_such_slot	SIMPLE	EXCEPTION	Provide the name of the slot being accessed.
not_a_closure	SIMPLE	EXCEPTION	Provide the object type
index_out_of_bounds	SIMPLE	ADVANCED	Provide the index requested that fell out of bounds alongside of the length of the vector.
not_compatible	EXCEPTION	ADVANCED	Provide the starting class and requested end class.

Table Macro Use Key:

• SIMPLE = RCPP_SIMPLE_EXCEPTION_CLASS
• EXCEPTION = RCPP_EXCEPTION_CLASS
• ADVANCED = RCPP_ADVANCED_EXCEPTION_CLASS (new mimics stop()/warning() )

Note: The difference between SIMPLE and EXCEPTION is the ability to pass a std::string to EXCEPTION that can dynamically change output whereas SIMPLE error codes are static.

Error Codes to Remove

The following error codes are not used (or are not detectable via grep / GitHub search):

• no_such_field
• reference_creation_class: Not used in Reference.h

Misc List of Error Code Class Usage

Below is a list of the current exception classes in Source (via GitHub)

RCPP_SIMPLE_EXCEPTION_CLASS

• not_a_matrix
• index_out_of_bounds
• parse_error
• not_s4
• not_reference
• not_initialized
• no_such_slot
• no_such_field
• not_a_closure
• no_such_function
• unevaluated_promise

RCPP_EXCEPTION_CLASS

• not_compatible
• S4_creation_error
• reference_creation_error
• no_such_binding
• binding_not_found
• binding_is_locked
• no_such_namespace
• function_not_exported
• eval_error

[coatless]

By the way... I completely forgot to detail how the new exception message should be written.

Per feedback in #667, the goal here will be to state the error like before (in a similar syntax) and then follow it up with additional diagnostic information in the form of [key1=value1; key2=value2...]. You can see this take form in #667's ba4403d commit.

For simplicity, here is a list of the original exception messages contrasted with the new exception messages:

Old	New
"cannot convert to environment"	"Cannot convert object to an environment: [type:%s; target=ENVSXP]."
"cannot convert to function"	"Cannot convert object to a function:
--	[type=%s; target=CLOSXP, SPECIALSXP, or BUILTINSXP]."
"not a promise"	"Not a promise: [type = %s]."
"expecting a single value"	"Expecting a single string value: [type=%s; extent=%i]."
"cannot convert to symbol (SYMSXP)"	"Cannot convert object to a symbol:
--	[type=%s; target=SYMSXP]."
"expecting an external pointer"	"Expecting an external pointer: [type=%s]."
"expecting a single value"	"Expecting a single value: [extent=%i]."
"expecting a string"	"Expecting a string: [type=%s]."
"expecting a string vector"	"Expecting a string vector: [type=%s; required=STRSXP]."
std::string("could not convert using R function : ") + fun	"Could not convert using R function: %s."
"not compatible"	"Not compatible conversion to target: [type=%s; target=%s]."
"not compatible with requested type"	"Not compatible with requested type: [type=%s; target=%s]."
"not compatible with STRSXP"	"Not compatible with STRSXP: [type=%s]."
"index out of bounds"	"Index is out of bounds: [index=%i; extent=%i]."
"index out of bounds"	"Column index is out of bounds: [index=%i; column extent=%i]."
"index out of bounds"	"Row index is out of bounds: [index=%i; row extent=%i]."
"index out of bounds"	"Location index is out of bounds: [row index=%i; row extent=%i;
--	"column index=%i; column extent=%i]."
"index out of bounds"	"Index is out of bounds: Object is missing names."
"index out of bounds"	"Index is out of bounds: [index='%i']."
"index out of bounds"	"Iterator index is out of bounds: [iterator index=%s; iterator extent=%i]"
"index out of bounds"	"Iterator index is out of bounds: [iterator=%s; index=%i; extent=%i]"

Any thoughts would be welcome as I'd like to have this change ready to go after #674 is merged.

[coatless]

@eddelbuettel:

Would you prefer it if I staged the exception message changes in two parts?

e.g.

• PR 1: Modification to exception.h to split into cpp98/cpp11 and adding a local mod in tinyformat to enable variadic templates?
• PR 2: Exception message changes to the proposed above?

Or just do it all together?

[eddelbuettel]

Two sounds like a reasonable split.

"Bite-sized". Some folks of course swallow warthogs whole so it all depends...