|AACC compiler generator firstname.lastname@example.org (1994-04-04)|
|From:||email@example.com (Joseph H Allen)|
|Date:||Mon, 4 Apr 1994 14:11:02 GMT|
I've posted an experimental LALR(1) parser generator called AACC (Ack!
Another Compiler Compiler!) to alt.sources
AACC is the result of an effort to replace the hand-rolled parser in my
interpreted language Ivy with a table driven parser. Ivy's syntax
contains a number of features for which YACC generated parsers are not
adequate. Namely, definable operators, lack of semicolons and blocks
indicated through indentation (although this particular feature could be
solved by the lexer as in Python).
One minor feature of AACC is that it provides an event-driven parser,
instead of a blocking parser. This is important for MUDs (one of IVY's
intended applications) where multiple input sources must be handled
asynchronously. YACC's parsing function, yyparse() is called once to
compile an entire source file. Whenever it needs input, it calls yylex(),
which may block if input is not yet available. AACC, on the other hand,
generates a parsing function which is called every time a new token
becomes available, and is thus friendly to event-driven applications.
The primary advantage of AACC however, is full control over conflict
resolution. YACC allows some conflicts to be resolved through the use of
precedence rules and gives a default handling for unresolved conflicts.
AACC, on the other hand, identifies each conflict in detail and allows the
user to either select the action for the conflict or to provide a function
for resolving the conflict during compiler run-time.
Here's a simple operator grammar for demonstrating AACC:
pgm expr : result # Final result
expr constant : pass0
expr expr op expr : doinfix
expr op expr : doprefix
expr lparen expr rparen : pass1
Each production (syntax rule) is on a separate line. The first word on
the line is the non-terminal symbol on the left hand side of the
production. The remaining words (up to the ':') are terminals (tokens)
and non-terminals on the right side of the production. The word following
the ':' is a function to call when the production is reduced (I.E., when
the right side is matched). Pound-signs introduce comments.
This grammar is ambiguous and when fed to AACC, this is printed:
Conflict: shift reduce(expr ==> op expr) : op
Conflict: shift reduce(expr ==> expr op expr) : op
Each conflict message indicates whether a possible action is a shift and
lists the possible reductions which could occur, followed by the next
input token (after the ':'). The first conflict message above indicates
that there's a shift/reduce conflict when "op expr" has been recognized
and another "op" is on the input. The second message indicates that
there's a shift/reduce conflict when "expr op expr" has been recognized
and another "op" is on the input.
Once the conflicts are identified, conflict resolution commands can be
appended to the grammar:
pgm expr : result
expr constant : pass0
:infix expr expr op expr : doinfix
:prefix expr op expr : doprefix
expr lparen expr rparen : pass2
(shift prefix : op) prefix # Resolve -A +B
(shift infix : op) : prec # Resolve A+B*C
The contents of the the parenthesis identify the conflict. They contain
'shift' if the conflict involves a shift, a list of one or more labels
which refer to productions (possible reductions), and the next input token
(following the ':').
The resolution of the conflict follows the conflict identifier. In the
first command above, the word 'prefix' indicates that the reduction should
occur, not the shift (this would be the case if we assume that all prefix
operators have higher precedence than infix operators). For the second
command, the ': prec' indicates that a function prec() will be called when
this conflict is recognized by the parser. Prec() should evaluate the
conflict using data not available to the parser and return the action to
be taken in the form of an index into the conflict identifier. If prec()
returns 0, the first action listed will be taken, the shift. If prec()
returns 1, the second action listed will be taken, the reduce.
If this grammar is fed into AACC, no conflict messages are printed and two
files are created: parsetab.c (containing the parsing tables) and parse.h
(containing function declarations). The parser is in a separate file
(parse.c) which is included with AACC.
The line "typedef int VAL;" in parse.h defines the value associated with
each terminal and non-terminal. When a reduction function is called (when
a production has been recognized), it should retrieve the VALs from the
top of the stack which are associated with the right side of the
production, operate on them, and return a new VAL for the non-terminal on
the left side of the production.
VAL is typically redefined to a parse-tree node pointer, and the reduction
functions do tree node CONSes.
Currently, the parser does no error recovery. I will add this in the next
version. Also I'll write a man page.
Return to the
Search the comp.compilers archives again.