I've just updated the wiki pages to refer to JavaCC 5. Now I wanted to
add JavaCC 5 to the vendor project, but somehow my svn credentials don't
get accepted:
$ svn-commit
Deleting JavaCC2_1.class
Authentication realm: <http://svn.openlaszlo.org:80> svn.openlaszlo.org
Username: bargull
Password for 'bargull':
Authentication realm: <http://svn.openlaszlo.org:80> svn.openlaszlo.org
Username: bargull
Password for 'bargull':
svn: Commit failed (details follow):
svn: CHECKOUT of '/!svn/ver/16963/vendor': authorization failed: Could not
authenticate to server: rejected Basic challenge (http://svn.openlaszlo.org)
On 2/3/2011 1:34 AM, P T Withington wrote:
We also should send some mail to laszlo-dev warning people about this upcoming
change and telling them how to update their JavaCC before we check in the
change, and then mail once we _do_ check it in to remind them.
I will take care of making sure the laszlo-internal people (who tend to ignore
laszlo-dev) are made aware.
On 2011-02-02, at 19:21, P T Withington wrote:
Comment:
It bugs the heck out of me that they kept the ASI between `return` and an
expression on a separate line. I find it hard to believe that is not always a
bug. But I guess we should obey the standard...
Questions:
1. Did you mean to leave the NoLineTerminator commented out of the #include and
#pragma productions?
2. Should/could the pattern `[LOOKAHEAD({Sc()}) ";"]` be a named production,
for better readability?
3. It appears we have no Java 1.3 or 1.4 code any more. Should we just remove
all mention of those from the build.xml files (and just name the build
directory `build` instead of `build.1.5`)?
Issues:
1. This is surely not your bug, but when I first compiled, I got some
serialization errors about caches building the solo consoles. I am pretty sure
I had done an `ant clean` before starting, which I would have _hoped_ would
nuke any caches. Either I missed that step, or `ant clean` is deficient. We
should try to determine which is the case.
Otherwise, approved!
On 2011-02-02, at 14:17, André Bargull wrote:
I'll update svn.openlaszlo.org/vendor/ and the wiki pages [1,2] later.
[1] http://wiki.openlaszlo.org/SubversionBuildInstructions#Install_JavaCC
[2]
http://wiki.openlaszlo.org/NativeWindows7SubversionBuildInstructions#Install_javacc
Change bargull-20110202-U3G by bargull@Bargull02 on 2011-02-02 17:03:34
in /home/anba/src/svn/openlaszlo/trunk
for http://svn.openlaszlo.org/openlaszlo/trunk
Summary: Upgrade to JavaCC5, fix ASI bugs in parser
New Features: LPP-8788 (We should use a more modern version of JavaCC)
Bugs Fixed: LPP-9714 (Parser does not enfore no-line-terminator in
break/continue/throw/postfix-op statements), LPP-9719 (Semicolon after do-while
creates empty statement), LPP-9720 (Newline in multi-line comment not treated
as newline for ASI), LPP-9721 (TypeIdentifier() should enforce NoLineTerminator
before Nullable()/NonNullable()), LPP-9722 (Multi-line comment not handled when
computing ASI)
Technical Reviewer: ptw, dda
QA Reviewer: (pending)
Overview:
JavaCC 5 has got a bug with javacode non-terminals, somehow the tokenstream
isn't rewinded properly in a (deeply?) nested syntactic lookahead if used in
conjunction with javacode non-terminals (e.g. see lookahead for
ModifiedDefinition() in Directive() and the Sc() non-terminal). Even using a
no-op javacode non-terminal broke the parser (JAVACODE void Nop ()
{/*empty*/}). Because of that reason I had to re-implement the whole optional
semicolon mechanism in the parser. While doing this rework, I've encountered a
couple of bugs (LPP-9714, LPP-9719, LPP-9720, LPP-9721, LPP-9722).
Details:
All Sc() non-terminals for optional semicolons have been rewritten to use a semantic
lookahead: [LOOKAHEAD({Sc()}) ";"]
The new Sc() function simply returns `true` if the next token in the stream is
a semicolon, this semicolon will then simply be consumed. If the next token is
not a semicolon, but an acceptable choice for automatic semicolon insertion
(ASI), Sc() returns `false`. So the lookahead is unsuccessful and by that the
optional block [...] is not processed any further. If the next token is neither
a semicolon nor an acceptable choice for ASI, a parser exception is thrown.
JavaCC cannot infer the content a javacode non-terminal, therefore it couldn't
warn for possible ambiguities in the grammar after Sc() calls. As the old Sc()
non-terminal is gone, I had to add additional lookaheads to resolve these
ambiguities. See CallExpression(), AdditiveExpression(), RelationalExpression()
productions.
LPP-9714 (no-line-terminator rule):
The ECMAScript grammar has got restricted productions for the PostfixExpression,
ContinueStatement, BreakStatement, ThrowStatement and ReturnStatement productions. For
these productions, line terminators are strictly prohibited. This restriction was only
implemented for the ReturnStatement, it's now implemented for the other productions as
well. The pattern used to implement the restriction is simply: [LOOKAHEAD(NonTerminal(),
{NoLineTerminator()}) NonTerminal()]. So a combination of syntactic and semantic
lookahead: if the next token(s) can be parsed as NonTerminal() and there are no
intervening newlines, consume the tokens, otherwise just proceed. I've added an extra
function to produce a nice error message for the ThrowStatement production, so it's
easier for the user to understand what's going wrong when she added a newline after the
`throw` statement. The default error message by JavaCC is pretty useless here:
Encountered "" at line 1, column 61. Was expecting one o
f: (empty list here)
LPP-9719 (do-while loop):
Just added the optional semicolon predicate for the do-while loop in
IterationStatement().
LPP-9720 (newline in multi-line comment):
Scan for newlines in multi-line comments, they count as newlines for the
syntactic grammar, cf. ECMAScript 5, 7.4 Comments
LPP-9721 (restriced production for TypeIdentifier):
TypeIdentifier was not a restricted production, but it should be, otherwise "var x:Boolean \n
!x;", where \n denotes a line break, is not parsed as "var x:Boolean; !x;".
LPP-9722 (multi-line comment for ASI):
The (now removed) optionalSc() function just used the first special token, but
special tokens are chained, e.g. consider multiple single-line and multi-line
comments appearing next to each other. Therefore it's necessary to loop over
the specialToken field, just like it's now done in NoLineTerminator().
Token.java:
JavaCC 5 requires a new interface in Token: Token.newToken(int, String).
TestParserASI.java:
JUnit test case for LPP-9714, LPP-9719, LPP-9720, LPP-9721, LPP-9722
build.xml:
- update 1.4 -> 1.5
- comment out org/openlaszlo/iv/flash/context package, this was recently
discussed on laszlo-dev
- explicitly add EDU/oswego package for build, sometimes my ant doesn't pick it
up, so make it explicit
- add TestParserASI to the "test" target
Tests:
TestParserASI test case, lfc compiles, smokecheck and alldata (swf10,
dhtml)x(Firefox, Opera, Safari, IE)
(Don't forget to download and install JavaCC 5 and update your JAVACC_HOME
environment variable accordingly!)
Files:
M WEB-INF/lib/javacc.jar
M WEB-INF/lps/server/build.xml
A WEB-INF/lps/server/src/org/openlaszlo/test/TestParserASI.java
M WEB-INF/lps/server/sc/src/org/openlaszlo/sc/Parser.jjt
M WEB-INF/lps/server/sc/src/org/openlaszlo/sc/parser/Token.java
Changeset: http://svn.openlaszlo.org/openlaszlo/patches/bargull-20110202-U3G.tar
