A-code command parsing

While all command examples in this document are given in the upper case for ease of visual identification, A-code parser is case-insensitive: all commands are automatically forced into lower case.

Where and when parsing takes place

There is no separate A-code directive for player command parsing. Parsing takes place as an integral part of acquiring player input. There are two separate minor directives which do so: INPUT and QUERY. It is INPUT that does the heavy lifting. The much simpler QUERY directive is there merely as a convenience for asking yes/no questions. For reasons explained elsewhere it has to be avoided in A-code games, which are expected to be built in single-turn or library modes.

The QUERY directive

The QUERY directive is in fact one of the set of A-code conditionals. It takes as its single argument a text, which gets displayed as a prompt and expects yes or no in response. For example:

QUERY "Do you really mean it?"
   SAY "It shall be done, boss!"    # Player said yes
   SAY "Oh, OK. You had me worried there for a moment."   # Player said no

Parsing is extremely primitive: any response starting with 'n' (or 'N') is taken to mean "no", and any response starting with 'y' (or 'Y') is taken to mean "yes". A null response is interpreted as "yes". Any other response is rejected with "Eh? Do me a favour and answer yes or no! Which will it be?" (this text is at present hard-wired into the kernel). If the player still responds with something that cannot be interpreted as yes or no, the parser says "(OK, smartass... I'll assume you mean YES - so there!)" (also hard-wired) and returns true.

The INPUT directive

The INPUT directive is the one that acquires and parses player commands. It takes as its optional argument a text (or a variable pointing to a text), which is to be used as a prompt, instead of the default question mark followed by a space. The rest of this document is devoted to parsing done by INPUT.

While the basic form of a player command has the simple "<verb> <noun>" format, the parser can handle compound commands, reducible to a series of simple command in the basic format. When a compound command is given, successive executions of the INPUT directive process the implied simple commands, one at a time.

The result of parsing a simple command is delivered in three variables: ARG1, ARG2 and STATUS. ARG1 and ARG2 contain respectively refnos of the verb and the noun of the simple command parsed. The STATUS variable is set to the number of words in the command: one or two, or to the automatically defined constant BADSYNTAX a syntax error has been identified. For more detail see below, in the section on identifying vocabulary words.

Core parser operation

Conceptually, A-code command parsing has three separate components:

  • Parsing that is performed by the INPUT directive regardless of the game code,
  • Additional parsing features which are enabled by some particular entities being declared in the game code, and
  • Optional parsing features made available by the kernel to A-code games.

This section deals with core parsing features, which are not influenced by the game code.

Simple commands

A command consists of one or more words (a.k.a. tokens), delimited by blanks (spaces) and/or commas (,) and/or dots (.) and/or semicolons (;). All leading and trailing blanks are ignored and multiple blanks are treated as a single blank.

A simple player command has the verb or verb/noun structure. E.g.


Compound commands

Compound commands can be constructed by joining simple commands by command delimiters. The parser recognises three such delimits, which are completely synonymous: a semicolon (;), a dot (.) or ' then ' (note that unlike the colon or the dot, the ' then ' delimiter must be surrounded by blanks, to make it into a separate token.

Compound command may also feature object iteration: a verb followed by a list of nouns, separated by commas (,) or ' and '. Again, the surrounding blanks are mandatory for the latter form.

Thus for example,

is equivalent to

and is parsed as meaning


In order to be maximally forgiving, the parser will also understand some incorrect but unambiguous variants on this syntax. E.g. in ' and then ' (equivalent to ',.') the iteration delimiter is ignored. So


will have the obviously intended effect.

The special English words AND and THEN are default equivalents of a comma and a semicolon respectively, but they can be replaced by others by a game's code. See the A-code UTF8 section.

Identifying command words

Individual command tokens are matched against the game's vocabulary. At game's discretion, the matching process may be

  • restricted to exact matches, or
  • permit minimal (automatically derived) abbreviations, or
  • permit single typo correction – see the vocabulary description for more details.

See the A-code vocabulary description for details of command words matching against game's vocabulary.

If a command word is not found in the game's vacabulary and the ENTNAME variable is defined, the list of entity names is also searched, looking for an exact match.

If a simple command is parsed successfully, values of ARG1 and ARG2 are set to the refno values corresponding to the first and the second command token respectively. However, a further enhancement, if ARG1 value turns out to be in the range of object or location refnos, and ARG2 in the range of verb refnos, the two command words are swapped around. Thus, for example BIRD GET gets parsed as GET BIRD.

An additional parsing feature is present for games of style 11 or higher. If no match is found for the second command word, the kernel still does not give up. It is possible that the player was trying to reference something mentioned in an object or place description, or in some response recently given to an earlier command. Every time some text is output, all of its words longer than 3 characters and not ending in "ing", are stored in a separate, temporary vocabulary. This vocabulary, which is re-initialised whenever the player changes location, is scanned for an exact match (no abbreviations or typos). If a match is found, this is still treated as a matching failure, but of a different kind, so that if desired, it can be treated differently by A-code source.

If the parser fails to match a vocabulary word, the corresponding ARG1 or ARG2 variable is set to one of special pre-defined values:

  • BADWORD – no match of any kind.
  • AMBIGWORD – abbreviation matching gives a non-unique result – more than one vocabulary word could be meant.
  • AMBIGTYPO – single typo matching gives a non-unique result.
  • SCENEWORD – the match is not against the vocabulary by against the list of words used by the guide since the last change of location.
  • BADSYNTAX – any other parsing failure. (Note that in this case the STATUS variable is also set to BADSYNTAX.)

Finally if a command consists of just a single word (the STATUS variable is set to 1), the ARG2 variable is set to the predefined constant NOWORD.

Treating blanks as list or command separators

The verb SAY is treated by the parser as a special case in that spaces can be used instead of commas as in SAY FEE FIE FOE, which is treated as SAY FEE, FIE, FOE (which is equivalent to SAY FEE; SAY FIE; SAY FOE).

Inverting verb/noun order

By default, in two word commands, the first command token is considered to be the verb and the second one as the noun, e.g. TAKE BOTTLE. However, if the first token is not likely to be a verb (being e.g. a place or an object) and the the second is identifiably a verb, the parser will automatically swap them around, making BOTTLE TAKE also a legitimate command.

Additional parsing features enabled by game code

Some functionality of the A-code kernel is only present if game source defines particular entities.

Repeating commands

If game source declares the word AGAIN (possibly with some synonyms), commands can be repeated by using AGAIN as a verb. If used within a compound command it will repeat the last sub-command delimited by THEN (or a dot or a semicolon). If used on its own, it will repeat the whole of the player's last input, which may be a compound command (and may itself contain AGAIN on order to repeat its sub-command).

The special English word AGAIN is merely the default "repeater" word.

So, for example:


is equivalent to




is equivalent to


Non-vocabulary "nouns"

Sometimes it is not desirable to match player input against the games vocabulary at all. Saving and restoring games is an obvious example of this – players cannot be restricted to game's vocabulary for naming saved games. One could, of course, insist that SAVE and RESTORE do not take a save name as the second word of the command, but invoke instead a separate input routine, which takes the desired save name without any reference to the vocabulary. However, the same applies to any command which takes a numerical argument, e.g. specifying screen width or height in the console mode.

Rather than having special code for handling such (and similar) exceptions, A-code's approach is to tag relevant verbs as "special". In the absence of flag settings for vocabulary words (possibly to be rectified in the future), the solution is to groups declaration of such verbs between declarations of two pseudo-verbs: first.special and last.special. This works because (a) A-code allocates refnos (in a given entity category) in the order of declarations and (b) if prefixed with '-' these pseudo-verbs are themselves allocated refno, but are not added to the game's vocabulary. Here's an example based on Adv770:

# The next block are specials, not requiring validation of ARG2.
verb -first.special        # Mark the first one
verb again, repeat, =r
verb save, suspend, pause
verb restore, load
verb rest, wait            # In case players type REST MYGAME
verb !length, =!line, =!width
verb !scroll, =!screen, =!depth
verb !margin, =!offset
verb restart, initialise
verb why
verb please
verb -last.special         # Mark end of special verbs
# End of verbs not requiring validation of ARG2.

The kernel is aware of the special significance of first.special and last.special and will automatically suppress validation of the second command word when parsing a command with any verbs defined as special in this manner. However, a similar mechanism can be profitably used by A-code source. Here's another example from Adv770:

verb -first.direction
verb north, =n
verb northeast, =ne
verb east, =e
verb southeast, =se
verb south, =s
verb southwest, =sw
verb west, =w
verb northwest, =nw
verb -last.compass.point
verb up, =u, upward, ascend
verb down, =d, downwards, descend
verb -last.direction

This makes it easy to check for a command word being a direction


or a compass point:


Handling IT

Handling the indexical noun IT is fairly straightforward in A-code. One declares a dummy object of that name (with whatever synonyms may be deemed appropriate) and sets its value to be a pointer to an appropriate object. The value of IT should be set to a pointer to an object in the following situations:

  • A player command explicitly names an object.
  • Player's inventory gets listed and the list consists of a single object.
  • Objects in the current location get listed and the list consists of a single object.

OTOH the value of IT should be probably cleared (set to zero) if either of the two kind of object list contain more than one item.

If a player's command contains the word IT and the IT object has a non-zero value FAKECOM can be used to set ARG! or ARG2 (as appropriate) to pretend that the command explicitly named the object pointed at by IT.

None of the above involves any special kernel functionality, so why is the matter even mentioned in this description of command parsing by the kernel? The reason is that there is one other situation which requires the value of IT to be modified, and that situation is handled by the kernel.

If IT is declared as an object and the game source successfully uses the DEFAULT directive to select a default object, then IT is automatically set by the kernel to point at the object defaulted to.

Optional parsing features available to game code

Other features of the A-code kernel, which are also dependent on some entities being or not being defined in game source, do not take effect automatically, but have to be explicitly triggered by game code as and when appropriate.

Orphan processing

If a player issues a single word command, which can be assumed to imply a a target object on an action (e.g. PUSH on its own, or KEYS ditto), the game can do better than just complain that not enough information has been given. If the A-code source declares PLS.CLARIFY as a flag, setting this flag on the automatic variable STATUS affect the way the next command gets parsed.

If the next command also consists of a single word, the "orphan processing" mechanism comes into play. It combines this command with the preceding one-word one, so that the player does not have to repeat the previously typed word. On the other hand, if the next command consists of two words, it is parsed in the usual manner and the clarification request is ignored. The PLS.CLARIFY flag gets automatically unset in either case, so that it does not have to unset explicitly by the game's code.

Here's an example of orphan processing in action:

? get
What do you want me to get?
? rod
You get the rod.
? rod
What do you want me to do with the rod?
? drop
You drop the rod.

The underlying code dealing with generic GET requests could look like this:

action get
   ifeq status, 1
      default portable            # Find default object flagged as portable
      ifeq arg2, ambigword        # If more than one possible target
         flag status, pls.clarify # Activate orphan processing
         quip "What do you want me to {arg1}?"

The code for handling the drop command in the above example would have to come into play, once the game fails to find an action associated with the word ROD.

   call arg1                   # If arg1 has an associated action, execute it
   ifflag arg1, object         # We fell through, so presumably no such action
      flag status, pls.clarify # Activate orphan processing
      quip "What do you want me to do with the {arg1}?"

Handling collective nouns

The A-code language makes it easy for a game to permit use of collective nouns such as, for example, ALL or TREASURE. This is done by using the A-code directive DOALL before handling the verb to be applied to a collective noun. Here, for example is a very simple code to handle GET ALL:

action get
# Check for the command being GET ALL.
  ifkey all               
    doall here, portable  # Sets up the "do all" loop
# The rest is ordinary handling of GET
  ifeq status, 2          # Do we know what to get?
  ifhere arg2             # Is the the object in the same place as the player?
  ifflag arg2, portable   # Is the object portable?
    apport arg2, inhand   # If so, relocate the object to player's possessions
    quip "You {arg1} the {arg2}  # Report the action.

The DOALL directive in the above example sets up the do all loop of command processing, with the effect of setting ARG2 to the first object matching the specified criteria (being both co-located with the player and flagged as portable), and sets the value of STATUS to 2. The effect of this is that the rest of ACTION GET code results in that object being picked up and the REPEAT code loop restarted (due to the use of QUIP instead of SAY when reporting the action).

Because of the do loop being active, instead of prompting the player for the next command, the kernel constructs that command by taking the same verb (GET in our case) and combining it with the next matching object, if any. That command is again processed in the normal way. This continues until such time as there are no matching objects left, at which point the do all loop is terminated and the player prompted for a command.

This is obviously a very simplistic implementation and various checks may be necessary, such as e.g. checking that the player has a spare carrying capacity for the next object to be picked up. Such checks cab be performed as appropriate using the IFDOALL directive, which executes its associated block of code if the do all loop is active. If needs be, the loop can be aborted by the FLUSH directive.

TREASURE could be handled in a similar manner, assuming that in addition to being flagged as being portable (via the game-defined flag PORTABLE in the above example), it is also flagged as valuable by some other game defined flag.

Using EXCEPT with collective nouns

As a further enhancement of handling collective nouns, if EXCEPT is declared as a vocabulary word, the parser accepts an extension of the simple verb/noun command structure. Where a collective noun (e.g. ALL) is allowable as a command noun, it can be followed by EXCEPT (or a synonym thereof) and a list of entities to be exempt from the requested action. For example:


will be understood in a natural manner.

This expanded syntax has to, of course, cater for the possibility of unrecognised words being given by the player in the list of exceptions. To do so, this syntax enhancement brings into existence the automatic variable ARG3. If a word in the exception list is not recognised, the do-all loop is aborted and ARG3 is set to an appropriate error code, and its associated word is the unrecognised word of the player's command. Here is an example of this kind of error handling in Adv770:

   ifkey all
      call shadow.shutup
      ifeq arg3, badsyntax
         quip no.except, arg3
      ifeq arg3, ambigword
         quip tell.me.more, arg3
      ifeq arg3, badword
         quip nocomprende.object, arg3
      ifeq arg3, ambigtypo
         quip is.it.a.typo? arg3

Back to the documentation index
To the Mipmip home page
Feel free to leave a comment!
Mike Arnautov (27 March 2023)