NAME

Marpa::R2::Semantics::Phases - SLIF recognizer phases

About this document

This document describes in detail the sequence of events in the SLIF recognizer. It serves two purposes.

  • It describes the sequence of events in the SLIF recognizer in fuller detail than the other documents do.

  • It describes the use of more than one parse series.

Full detail of the sequence of events inside the SLIF recognizer is not usually necessary. Applications usually find that things "just work". But some application do depend on the details. These include

  • applications which make unusual use of side effects in the semantics; and

  • applications which alter their symbol tables at runtime.

The semantics has two different modes, one of which is deprecated. If the SLIF recognizer is created from a grammar which uses the SLIF grammar's action_object setting, the SLIF's semantics will be in "legacy" mode. Otherwise, the SLIF's semantics will be in "standard" mode.

Legacy mode is deprecated and should not be used in new applications. It is documented in its own section. Unless otherwise specified, the other sections of this document assume that "standard" mode is in use.

Parse trees and parse series

Because Marpa allows ambiguous parsing, each parse can produce a a series of zero or more parse trees, -- a parse series. Each parse tree has its own parse result. Because the parse may fail, there may be zero parse trees in a parse series. Because a parse may be ambiguous, there may be more than one parse tree in a parse series.

The SLIF allows the recognizer to be run several times on the same virtual input stream. Each of these runs is a separate parse series. Most applications use only one parse series.

The call to the SLIF recognizer's read() method is the start of the first parse series. The first parse series continues until there is a call to the series_restart() method or until the recognizer is destroyed. Each call of the series_restart() method starts a new parse series.

Few applications need to use more than a single parse series. But they are useful in some specialized applications. Starting a new parse series allows the applications to change some settings and to extend the input. Specifically,

  • Once input is read into a SLIF recognizer's virtual input stream, it is there for the life of the recognizer. It cannot be "unread". But during the Reading Phase of a new parse series, a recognizer can extend the virtual input stream, using the recognizer's resume() method.

  • Within a parse series, the semantics package cannot change. But you can specify a new semantics package after starting a new parse series.

  • Certain other SLIF recognizer settings can be changed after starting a new parse series. These include the SLIF recognizer's end setting. For details, see "Recognizer settings" in Marpa::R2::Scanless::R.

SLIF recognizer phases

A SLIF recognizer is always in one of three phases:

  • The Initial Phase;

  • a Reading Phase; or

  • an Evaluation Phase.

In these documents, phase and subphase names are capitalized for clarity.

Reading and Evaluation Phases are always part of a parse series. The Initial Phase is never part of any parse series.

The Initial Phase

The Initial Phase of a SLIF recognizer begins when the recognizer is created with the a call of the new() method. It ends when the read() method is called. It will also end, of course, if the recognizer is destroyed, but most applications will want to continue into the next phases.

There is always exactly one Initial Phase for every recognizer. The Initial Phase is not part of any parse series. Very little can happen in the Initial Phase, but it is possible to change some recognizer settings using the set() method.

The Reading Phase

The Reading Phase starts when a parse series starts, that is, with either a call to the read() method or a call to the series_restart() method. The Reading Phase ends when the first of the following occurs:

  • The recognizer is destroyed.

  • A new parse series is begun by calling the series_restart() method. This and the previous case are unusual. Applications almost always want to continue the parse series into the Evaluation Phase.

  • The value() method is called for this SLIF recognizer. This is the most common case.

The read() method can be called only once for a SLIF recognizer. But, while a Reading Phase continues, the resume() method may be called multiple times to continue reading input. The resume() method should only be called except during the Reading Phase of a parse series.

A Reading Phase is always part of a parse series. There is always exactly one Reading Phase for every parse series.

The Evaluation Phase

The Evaluation Phase begins with a recognizer's first call to the SLIF's value() method. It ends when the recognizer is destroyed, or when the SLIF's series_restart() method is called for that recognizer. If a parse series ends before the value() method is called, there is no Evaluation Phase.

The value() method may be called more than once during the Evaluation Phase. The second and later calls of the value() method will return the result of the next parse tree, if there is one. When there are no more parse trees, the value() method will return a Perl undef.

If you call the resume() method during an Evaluation Phase, the behavior of Marpa::R2 is not defined. You should not do this.

When it occurs, an Evaluation Phase is always part of a parse series. But the Evaluation Phase is optional -- a parse series may end without entering into an Evalution Phase.

The Evaluation Phase has two subphases:

  • The Parse Tree Setup Subphase.

  • The Parse Tree Traveral Subphase.

Node Evaluation Time is the Tree Traversal Subphase, as seen from the point of view of each rule node. It is not a separate phase.

Parse Tree Setup Subphase

The Parse Tree Setup Subphase occurs during during the first call to the value() method of every series. In the Parse Tree Setup Subphase, the following happens:

  • The per-parse argument is created. If a per-parse constructor was found, it is run at this point, and the per-parse argument is its return value.

  • All action names are resolved to actions -- Perl scalars or rule evaluation closures. The rule evaluation closures are not called in the Parse Tree Setup Subphase. They will be called later, in the Parse Tree Traversal Subphase.

Exactly one Parse Tree Setup Subphase occurs for each parse tree.

Parse Tree Traversal Subphase

During the Parse Tree Traversal Subphase, the rule evaluation closures are called.

Legacy mode

If the SLIF recognizer is created from a grammar which uses the action_object SLIF grammar setting, the SLIF recognizer will be in "legacy" mode. Legacy mode is deprecated. It is documented here only for the support of legacy applications.

In legacy mode, there is an additional evaluation subphase, the Parse Series Setup Subphase, and tasks performed in standard mode by the Parse Tree Setup Subphase are split between the two subphases.

The Parse Series Setup Subphase occurs just before the Parse Tree Setup Subphase, in the first value method call of the first parse series. It is important to emphasize that the Parse Series Setup Subphase occurs only in the first parse series.

In legacy mode, the only task performed by the Parse Tree Setup Phase is creation of the per-parse argument. Action resolution is performed in the the Parse Series Setup Phase.

One practical implication of this is that, in legacy mode, the semantics package cannot be changed by starting a new parse series. Both are decided once and for all in the Parse Series Setup Subphase, and the Parse Series Setup Subphase occurs only once during the life cycle of a SLIF recognizer.

The precise timing of semantics operations is affected by the changes described above. These changed timings, in turn, may affect applications that rely on side effects of the semantics.

Copyright and License

  Copyright 2022 Jeffrey Kegler
  This file is part of Marpa::R2.  Marpa::R2 is free software: you can
  redistribute it and/or modify it under the terms of the GNU Lesser
  General Public License as published by the Free Software Foundation,
  either version 3 of the License, or (at your option) any later version.

  Marpa::R2 is distributed in the hope that it will be useful,
  but WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  Lesser General Public License for more details.

  You should have received a copy of the GNU Lesser
  General Public License along with Marpa::R2.  If not, see
  http://www.gnu.org/licenses/.