Header lexy/dsl/recover.hpp

The try_, recover and find rules for error recovery.

Unless otherwise specified, lexy will cancel parsing on the first error. This cancellation propagates outward until either the entire parsing fails, or a try_ rule is reached. This is similar to how an exception will perform stack unwinding until the entire program is terminated or a try block reached. try_ allows recovering from the error so parsing can continue after the first error.

Rule lexy::dsl::try_

lexy/dsl/recover.hpp
namespace lexy::dsl
{
    constexpr rule auto try_(rule auto rule);
    constexpr rule auto try_(rule auto rule, rule auto recovery_rule);

    constexpr branch-rule auto try_(branch-rule auto rule);
    constexpr branch-rule auto try_(branch-rule auto rule, rule auto recovery_rule);
}

try is a rule that catches any errors raised during parsing of rule and recovers from them, so parsing can continue.

Parsing

Parses rule.

Branch parsing

Tries to parse rule and succeeds if that succeeds. If rule backtracks, this is not considered a situation where recovery needs to happen, and it also backtracks. Only recovers if rule fails, failure of recovery (see below) do not lead to a backtrack; at this point the branch has been committed.

Errors

All errors raised by rule. If rule has not already recovered, recovers for it:

  • The single argument overload does it by consuming nothing, parsing then simply continues as if nothing happened.

  • The two argument overload does it by parsing recovery_rule. If that succeeds, recovery is finished. Otherwise, raises all errors from it unchanged and cancels.

Values

If rule succeeded, all values produced by rule. Otherwise, all values produced by recovery_rule. In particular, the values of rule are not produced in that case.

The single argument overload useful if rule matches something that is not technically necessary for parsing, but just there for redundancy. The two argument overload accepts a recovery_rule whose purpose is to bring the parser into a known state. This is useful to skip bad input until a well-known synchronization point is reached, like a statement separator. If recovery_rule fails, it requires a try_ rule at a higher level.

The behavior of try_ is analogous to a C++ try block. If an error is raised during rule, it is caught by the recovery_rule, which corresponds to a catch (…​) block. If recovery_rule succeeds, this corresponds to a catch block that is executed without error. If recovery_rule raises an error, this corresponds to a catch block whose execution raises another exception. recovery_rule can also fail without raising an error; this corresponds to a throw; statement that re-throws the exception.

Example 1. Recover from missing version numbers
struct version
{
    std::optional<int> major, minor, patch;
};

struct production
{
    static constexpr auto rule = [] {
        // If we don't have an integer, recover by producing nullopt.
        auto number = dsl::try_(dsl::integer<int>(dsl::digits<>), dsl::nullopt);
        auto dot    = dsl::try_(dsl::period);
        return number + dot + number + dot + number;
    }();

    static constexpr auto value = lexy::construct<version>;
};
Note
See tutorial.cpp for a more complete version number parser.

Rule lexy::dsl::find

lexy/dsl/recover.hpp
namespace lexy::dsl
{
    struct find-dsl // models rule
    {
        rule auto limit(token-rule auto ... limits);
    };

    constexpr find-dsl find(token-rule auto ... tokens);
}

find is a recovery rule that skips input until a synchronization token rule is reached.

Parsing

Tries to match any of the tokens anywhere in the remaining input; consumes everything until but not including the first of tokens. Fails if EOF is reached first or one of limits matches, if they were specified.

Errors

None. It can fail, but does so without raising errors.

Values

None.

Parse tree

A single token node whose range covers everything consumed before the token. Its lexy::predefined_token_kind is lexy::error_token_kind.

If recovery is successful, the reader has been advanced to the position just before the first of the tokens. Parsing can then continue from this known state, e.g. from a statement separator. If recovery fails, no additional error is raised but parsing is canceled, to try recovery at a higher level.

Example 2. Parse a list of declarations with error recovery using find()
constexpr auto id          = dsl::identifier(dsl::ascii::alpha);
constexpr auto kw_function = LEXY_KEYWORD("function", id);
constexpr auto kw_type     = LEXY_KEYWORD("type", id);

struct function_decl
{
    static constexpr auto rule = [] {
        auto arguments = dsl::parenthesized(LEXY_LIT("..."));
        auto body      = dsl::curly_bracketed(LEXY_LIT("..."));

        return kw_function >> id + arguments + body;
    }();
};

struct type_decl
{
    static constexpr auto rule //
        = kw_type >> id + dsl::lit_c<'='> + id + dsl::semicolon;
};

struct production
{
    static constexpr auto whitespace = dsl::ascii::space;
    static constexpr auto rule       = [] {
        auto decl = dsl::p<function_decl> | dsl::p<type_decl>;

        // We recover from any errors by skipping until the next decl.
        auto decl_recover = dsl::find(kw_function, kw_type);
        auto try_decl     = dsl::try_(decl, decl_recover);

        return dsl::list(try_decl);
    }();
};

Rule lexy::dsl::recover

lexy/dsl/recover.hpp
namespace lexy::dsl
{
    struct recover-dsl // models rule
    {
        rule auto limit(token-rule auto ... limits);
    };

    constexpr recover-dsl find(branch-rule auto ... branches);
}

recover is a recovery rule that skip input until a follow-up branch rule can be parsed.

Parsing

Tries to parse any of branches anywhere in the remaining input; consumes everything before the selected branch and then the branch itself. Fails if EOF is reached first or one of limits matches, if they were specified.

Errors
  • All errors raised by parsing the selected branch. The rule fails if the selected branch fails.

  • A failed recovery does not raise an error.

Values

All values produced by the selected branch.

Parse tree

A single token node whose range covers everything consumed before the token. Its lexy::predefined_token_kind is lexy::error_token_kind.

Unlike find, recover directly continues with one rule. If recovery has been successful, it has parsed the selected rule. Parsing can then continue as it would normally do a after that rule. If recovery fails, no additional error is raised but parsing is canceled, to try recovery at a higher level.

Example 3. Parse a list of declarations with error recovery using recover()
constexpr auto id          = dsl::identifier(dsl::ascii::alpha);
constexpr auto kw_function = LEXY_KEYWORD("function", id);
constexpr auto kw_type     = LEXY_KEYWORD("type", id);

struct function_decl
{
    static constexpr auto rule = [] {
        auto arguments = dsl::parenthesized(LEXY_LIT("..."));
        auto body      = dsl::curly_bracketed(LEXY_LIT("..."));

        return kw_function >> id + arguments + body;
    }();
};

struct type_decl
{
    static constexpr auto rule //
        = kw_type >> id + dsl::lit_c<'='> + id + dsl::semicolon;
};

struct production
{
    static constexpr auto whitespace = dsl::ascii::space;
    static constexpr auto rule       = [] {
        auto decl = dsl::p<function_decl> | dsl::p<type_decl>;

        // We recover from any errors by continuing with the next declaration.
        auto decl_recover = dsl::recover(dsl::p<function_decl>, dsl::p<type_decl>);
        auto try_decl     = dsl::try_(decl, decl_recover);

        return dsl::list(try_decl);
    }();
};

See also