| 1 |
Up to this point, error-recovery hasn't yet been covered i.e., how to continue |
| 2 |
parsing after the parser detects a syntax error. By default tt(parse) returns |
| 3 |
after calling tt(error). This means that an erroneous input line ends the |
| 4 |
calculator. Now we show how to recover from erroneous input. |
| 5 |
|
| 6 |
The b() grammar-specification language includes the reserved symbol tt(error), |
| 7 |
which can be used in production rules. In the example below it has been added |
| 8 |
as a new alternative for recognizing the tt(line) nonterminal: |
| 9 |
verbinsert(//LINE errorcalc/parser/grammar) |
| 10 |
This addition to the grammar allows the calculator to recover from syntax |
| 11 |
errors. If a syntax error is encountered, the third production rule is |
| 12 |
activated, skipping all tokens until a newline token has been encounters. At |
| 13 |
that point tt(line) has been recognized, and parsing continues afresh at the |
| 14 |
next input line (the tt(error) member function is also called, printing its |
| 15 |
message). Different from bison's tt(error) implementation, b() proceeds on the |
| 16 |
assumption that whenever tt(error) is used in a production rule it is the |
| 17 |
grammar constructor's intention to have the parser continue parsing |
| 18 |
(therefore, a statement like `tt(yyerrok)', encountered in bison grammars is |
| 19 |
superfluous in b() grammars). The reserved symbol tt(error) itself causes the |
| 20 |
parsing function to skip all subsequent input until a token that can follow |
| 21 |
tt(error) has been encountered. In the above implementation that token is the |
| 22 |
newline character `tt(\n)' (see chapter ref(RECOVERY)). |
| 23 |
|
| 24 |
This form of error recovery deals with syntax errors. There are other |
| 25 |
kinds of errors; for example, divisions by zero, which raises an exception |
| 26 |
signal that is normally fatal. A real calculator program must handle this |
| 27 |
signal and use whatever it takes to discard the rest of the current line of |
| 28 |
input and resume parsing thereafter. Handling such signals and other forms of |
| 29 |
semantic errors is not discussed here, as it is not specific to b() |
| 30 |
programs. But once a semantic error has been encountered, handling functions |
| 31 |
may call tt(ERROR()), resulting in the same procedure as the one that's used |
| 32 |
for syntax errors: all subsequenct tokens are skipped until a token has been |
| 33 |
encountered that can follow the reserved symbol tt(error). |
| 34 |
|