Recent changes to bugs

Phrase error recovery always throws away unexpected token

Carl Reinke — Wed, 11 Jul 2012 20:06:22 -0000

When the parser reaches phrase-level error recovery, it never retries the unexpected token after shifting the error token. This is problematic when that token begins a valid phrase -- the result is that a valid phrase following an error phrase is also thrown away as a part of the error phrase.

The attached patch addresses this issue, as well as improving the accuracy of the source range of the error token. (I also changed the termination condition for the simulator since reinserting the unexpected token can make the stream buffer full.)

Error recovery throws NullPointerException

Carl Reinke — Fri, 27 Jan 2012 17:42:40 -0000

During error recovery the parser throws NullPointerException (buffer is null during call to Parser.remove(int i)).

It is possible that I am simply misusing the "error" non-terminal, in which case it would be helpful for beaver-cc to emit a warning.

The attached example (a modification of the MLton grammar for Standard ML), results in the following output:

:0,0-0,0: Syntax Error: unexpected token "else"
:0,0-0,0: Recovered: replaced unexpected token with #21
:0,0-0,0: Syntax Error: unexpected token ""fail:\n""
:0,0-0,0: Recovered: removed unexpected token ""fail:\n""
:0,0-0,0: Syntax Error: unexpected token ";"
:0,0-0,0: Recovered: replaced unexpected token with #28
:0,0-0,0: Syntax Error: unexpected token "end"
:0,0-0,0: Recovered: removed unexpected token "end"
:0,0-0,0: Syntax Error: unexpected token "fun"
Exception in thread "main" java.lang.NullPointerException
at beaver.Parser$TokenStream.remove(Parser.java:240)
at beaver.Parser.recoverFromError(Parser.java:667)
at beaver.Parser.parse(Parser.java:468)
at beaver.Parser.parse(Parser.java:409)
at Main.main(Main.java:15)

unused nonterminal cause IllegalStateException

Carl Reinke — Fri, 20 Jan 2012 03:40:44 -0000

Instead of warning about unused nonterminal (as with unused terminals), an exception is thrown:

java.lang.IllegalStateException: found less nonterminals than previously counted
at beaver.spec.GrammarBuilder.getNonTerminals(GrammarBuilder.java:602)
at beaver.spec.GrammarBuilder.visit(GrammarBuilder.java:530)
at beaver.spec.ast.GrammarTreeRoot.accept(GrammarTreeRoot.java:27)
at beaver.comp.ParserGenerator.parseGrammar(ParserGenerator.java:837)
at beaver.comp.ParserGenerator.compile(ParserGenerator.java:787)
at beaver.comp.run.Make.compile(Make.java:128)
at beaver.comp.run.Make.main(Make.java:165)

Out-of-index-exception in GrammarBuilder.java:trim()

Anonymous — Sat, 31 Oct 2009 00:50:31 -0000

In some cases the method GrammarBuilder.trim() throw an out-of-index exception, e. g. if the passed parameter is already empty (but not null). Then the statement "code.charAt(--i)" try to access the element at position -1. The same problem can occur when \r and \n are cutted at the end of the string. The attached patch fix this problem by adding check-code for index i.

exit code is 0 on error

Anonymous — Sat, 15 Aug 2009 02:33:42 -0000

The exit code is always 0, also if there is an error. This cause problems if beaver is used in a build environment. The attached patch set the exit code to 1 if the grammar contain an error or when an illegal argument exception occur.

missing Ant task parameters

Blaine Simpson — Sat, 22 Oct 2005 17:42:21 -0000

According to the Beaver how2run.html page:

"The task attributes have one to one correspondence
with command line switches."

(Ant calls task attributes "parameters"). However,
there are 10 command line
switches and only 9 Ant task parameters. I
particularly want to use the .stat
file option from Ant, and it's not available.

I'm adding this issue as a bug instead of a feature
request only because
the Ant target does not behave according to the Beaver
docs without this
feature. No problem if you want to close this tracker
and open a new
feature request instead.

(Even if all command line switches were implemented for
Ant, there would still
not be a 1:1 correspondence, since there is also a
target parameter for
"file", which is a command line argument but not a
command line switch.

Type is not propagated on shortcuts

Anonymous — Thu, 17 Feb 2005 01:27:34 -0000

Hello,

beaver is great work!

1) Is it possible to make beaver support mutiple goals?
Somtimes i need only to parse sub parts.

2 less important) If i assign to an nonterminal 'x' a java
type 'X'. The list detection generates for the 'xlist' the
type 'X[]'. That is very goot for effectiv work. But this
works only for 'x.javavarname' and not
for 'x.javavarname?'. For 'x.javavarname' the type is
Symbol. It woul be better if it is 'X[]' too.

Question) Could it be that beaver or flex changes hash
values of the parsed java objects? In my debugger i
have seen Strings same content and with diffrent hash.
This causes problems with HashMaps.

Thanks,

Michael Wagner

Sources ought to be in sf.net CVS

Jochen Wiedmann — Thu, 07 Oct 2004 13:00:04 -0000

For integration within gump (see
http://gump.apache.org), it would
be nice, if the sources were available in the sf.net CVS.

The nice thing is, that this would create automatic
error messages,
whenever a dependent Apache project (Jaxme in that
case) would
be in conflict with the CVS head.

Jochen

documentation: meaning of symbol()

Uli Bubenheimer — Fri, 17 Sep 2004 15:08:43 -0000

From one of the code samples on the spec.html
documentation page:

return symbol(lst); // "symbol" is a shortcut for "new
beaver.Symbol"

I understand this to mean: Use "symbol(lst)" and Beaver
will take care of expanding it to "new
beaver.Symbol(lst)" somehow.

But apparently that is not what happens. Is it just
there for confusion?

Unhelpful class name lookup error message

Uli Bubenheimer — Thu, 16 Sep 2004 16:12:43 -0000

I tried to compile a grammar specification with
0.9.3.4, which I had written for 0.9.1.4. When running
my Ant build script, I got the following error:

BUILD FAILED: C:\KOP\build.xml:63: Failed to lookup
class name in
C:\KOP\src\com\certusoft\koprt\ADLParser.beaver

This error message does not give an indication what the
problem is. Several issues:
1. The message does not indicate that this is an error
which occurred while trying to find the name of the
target class as specified by the %class directive in
the ADLParser.beaver specification file. "Class name"
could mean anything, some beaver class, some ant class,
some application class, whatever.

2. There is no indication what the actual problem is.
At least the message part of the original exception
should be printed.

3. I assume the actual reason for this error is that my
ADLParser.beaver file still uses plain curly braces
rather than {: and }:. Beaver should say something like
"syntax error" with a line number, etc., to indicate
what the problem is, since it might be tricky to find
otherwise. I am attaching the actual exception (which I
obtained by adding a printStackTrace in the code).