Here is a rundown of what typical error messages look like. I'll concentrate on gcc, since I assume this is the primary target. One typical problem is trying to construct an object from the wrong arguments. Here is an excerpt from the error message report:
{
const fmpzxx a;
fmpzxx_ref ar(a);
}
Compiler error output is 37 line(s), 8535 characters
------START COMPILER ERROR OUTPUT-------
t-compiler-errors.cc: In function ‘int main()’:
t-compiler-errors.cc:71:24: error: no matching function for call to ‘flint::fmpzxx_expression<flint::operations::immediate, flint::flint_classes::ref_data<flint::fmpzxx_expression<flint::operations::immediate, flint::detail::fmpz_data>, long int> >::fmpzxx_expression(const fmpzxx&)’
t-compiler-errors.cc:71:24: note: candidates are:
In file included from t-compiler-errors.cc:31:0:
/home/ness/src/flint/fmpzxx.h:51:5: note: template<class T, class U> flint::fmpzxx_expression::fmpzxx_expression(const T&, const U&, typename flint::mp::enable_if<flint::fmpzxx_expression<Operation, Data>::enablec2<T, U> >::type*)
.....
/home/ness/src/flint/fmpzxx.h:43:7: note: no known conversion for argument 1 from ‘const fmpzxx {aka const flint::fmpzxx_expression<flint::operations::immediate, flint::detail::fmpz_data>}’ to ‘const flint::fmpzxx_expression<flint::operations::immediate, flint::flint_classes::ref_data<flint::fmpzxx_expression<flint::operations::immediate, flint::detail::fmpz_data>, long int> >&’
------END COMPILER ERROR OUTPUT---------
The "...." denotes a lot more similar messages I have skipped. This is one of the error messages which got longer, but much more descriptive. Basically what is happening is that none of the constructors is applicable (as it states in the second line!), and then the compiler lists all five constructors (which is not particularly helpful in our case, but I suppose usually a good strategy). In the old error message one of the constructors was invoked, which then failed way down in the implementation details.
One thing to observe is that printing out all the type names in full is really distracting. Maybe I should write a sed script to shorten them.
An example of similar nature is trying to assign to something for which there simply is no rule to do the assignment (e.g. to a constant reference). Here is an excerpt:
{
fmpzxx a;
fmpzxx_srcref b(a);
b = a;
}
Compiler error output is 7 line(s), 3014 characters
------START COMPILER ERROR OUTPUT-------
In file included from /home/ness/src/flint/fmpzxx.h:32:0,
from t-compiler-errors.cc:31:
/home/ness/src/flint/flintxx/expression.h: In instantiation .......
/home/ness/src/flint/fmpzxx.h:50:5: required from ..........
t-compiler-errors.cc:85:13: required from here
/home/ness/src/flint/flintxx/expression.h:255:1: error: invalid application of ‘sizeof’ to incomplete type ‘flint::detail::STATIC_ASSERT_FAILED<false>’
/home/ness/src/flint/flintxx/expression.h:256:9: error: ‘doit’ is not a member of ‘rule_t {aka flint::rules::assignment<flint::fmpzxx_expression<flint::operations::immediate, flint::flint_classes::srcref_data<flint::fmpzxx_expression<flint::operations::immediate, flint::detail::fmpz_data>, flint::fmpzxx_expression<flint::operations::immediate, flint::flint_classes::ref_data<flint::fmpzxx_expression<flint::operations::immediate, flint::detail::fmpz_data>, long int> >, long int> >, flint::fmpzxx_expression<flint::operations::immediate, flint::detail::fmpz_data>, void>}’
------END COMPILER ERROR OUTPUT---------
This time the "........." denote really really really long lines I chopped off. We see a simulated STATIC_ASSERT here (and this line is the only thing new compared with yesterday's results). It's not as great as one might like, but at least the STATIC_ASSERT assures us that the following errors are most likely garbage. There is also a message associated with the STATIC_ASSERT, which can be found by looking in the source. Alternatively, when compiling in C++11 mode, the error message is displayed directly:
/home/ness/src/flint/flintxx/expression.h:254:9: error: static assertion failed: no rule to assign expression T to derived_t (after evaluating)
As a final example, consider trying to do unimplemented arithmetic operations. Here is an excerpt:
TEST_FMPZXX_ARITH_WRONG_DEEP
{
fmpzxx a;
newtype n;
a + (a*a + (a / n) + a)*a;
}
Compiler error output is 10 line(s), 1387 characters
------START COMPILER ERROR OUTPUT-------
t-compiler-errors.cc: In function ‘int main()’:
t-compiler-errors.cc:99:25: error: no match for ‘operator/’ in ‘a / n’
t-compiler-errors.cc:99:25: note: candidate is:
In file included from /home/ness/src/flint/fmpzxx.h:32:0,
from t-compiler-errors.cc:31:
/home/ness/src/flint/flintxx/expression.h:490:1: note: template<class Expr1, class Expr2> typename flint::detail::binary_op_helper<Expr1, flint::operations::divided_by, Expr2>::enable::type flint::operator/(const Expr1&, const Expr2&)
/home/ness/src/flint/flintxx/expression.h:490:1: note: template argument deduction/substitution failed:
/home/ness/src/flint/flintxx/expression.h: In substitution of ‘template<class Expr1, class Expr2> typename flint::detail::binary_op_helper<Expr1, flint::operations::divided_by, Expr2>::enable::type flint::operator/(const Expr1&, const Expr2&) [with Expr1 = flint::fmpzxx_expression<flint::operations::immediate, flint::detail::fmpz_data>; Expr2 = newtype]’:
t-compiler-errors.cc:99:25: required from here
/home/ness/src/flint/flintxx/expression.h:490:1: error: no type named ‘type’ in ‘flint::detail::binary_op_helper<flint::fmpzxx_expression<flint::operations::immediate, flint::detail::fmpz_data>, flint::operations::divided_by, newtype>::enable {aka struct flint::mp::enable_if<flint::traits::is_implemented<flint::rules::UNIMPLEMENTED>, int>}’
------END COMPILER ERROR OUTPUT---------
This is unchanged compared to yesterday, because it is actually pretty good.
No comments:
Post a Comment