#371 ConstantFixedExpressionEvaluator does not check if FIT is possible

[Rainer Müller]

What should happen here?

DCL f FIXED(7) INIT(1024 FIT f);

1024 does not fit into a FIXED(7).
The runtime system would induce a FixedRangeSignal.
Thus, I expect that the compiler emits an error message.
Test is in testsuite/tickets/371_fit.prl

the test should that there are several problems:

1. DCL f FIXED(7) INIT(1024 FIT f); aborts with INIT: could not abort initializer
2. DCL g FIXED(7); DCL f FIXED(7) INIT(1024 FIT g); aborts with constant expected

Reasons:

1. The symbol is not in the symbol table while INIT is evaluated
2. the grammar for constantFixedExpressionFit expects a constantFixedExpression instead of a name or ID?

@2: The German language report also denotes Faktor ::= .... [ FIT konstanter-FIT-Ausdruck ]
This bug seam to be very old!

by the way:

• The possibility of TOFIXED as factor is missing in the grammar
• a negative sign is ignored

[Rainer Müller]

The ignored '-' is now treated

A tests with the Werum compiler showed, that

• FIT in INIT expects a constant FIXED expression and it has no effect on the code except a warning is issued
• FIT in regular expressions just performs a type cast - no range checks occur

RTOS-UH: The language report of RTOS-UH does not contain the word 'FIT'

Next steps:

1. The semantic of FIT on a constant FIXED expression must become defined
2. the constantFixedExpressionEvaluator must be modified towards the defined behavior

[Rainer Müller]

Decision after a short discussion:

1. FIT in constantFixedExpressions are not necessary, since the expressions are evaluated with required precision by the compiler, the result must fit into the target type e.g. in INIT-lists
2. FIT has an fixed operand on rhs. If the lhs has a larger precision as rhs, an error message occurs
3. FIT in constant FIXED expressions must be elaborated in the language report

Fixed with commit 060c8abced6a99360da9a0251cc98a8094a4fc9b