Issue with frontier patterns

%f[set], a frontier pattern; such item matches an empty string at any position such that the next character belongs to set and the previous character does not belong to set. The set set is interpreted as previously described. The beginning and the end of the subject are handled as if they were the character ‘\0’.

– Lua 5.2 Reference Manual, §6.4.1

In the reference interpreter (Lua 5.3.4), the code belows prints a blank line and nil, indicating that the first pattern matched (‘o’ is not in [%.], but ‘.’ is) and that the second one didn’t.

In LuaJ 3.0.1, however, the code fails with

  org.luaj.vm2.LuaError: frontier-pattern-handling.lua:9 vm error: java.lang.ArrayIndexOutOfBoundsException: -1
  stack traceback:
    frontier-pattern-handling.lua:9: in main chunk
    [Java]: in ?
    at org.luaj.vm2.LuaClosure.execute(Unknown Source)
    at org.luaj.vm2.LuaClosure.onInvoke(Unknown Source)
    at org.luaj.vm2.LuaClosure.invoke(Unknown Source)
    at lua.processScript(Unknown Source)
    at lua.main(Unknown Source)
  Caused by: java.lang.ArrayIndexOutOfBoundsException: -1
    at org.luaj.vm2.lib.StringLib$MatchState.match_class(Unknown Source)
    at org.luaj.vm2.lib.StringLib$MatchState.matchbracketclass(Unknown Source)
    at org.luaj.vm2.lib.StringLib$MatchState.match(Unknown Source)
    at org.luaj.vm2.lib.StringLib$MatchState.start_capture(Unknown Source)
    at org.luaj.vm2.lib.StringLib$MatchState.match(Unknown Source)
    at org.luaj.vm2.lib.StringLib.str_find_aux(Unknown Source)
    at org.luaj.vm2.lib.StringLib$match.invoke(Unknown Source)

print(string.match("foo.", "(%f[%.])"))
print(string.match("foo", "(%f[%.])"))

[plamentotev]

I think I found where the issue is. Here is a patch:

--- StringLib.java      2018-03-03 23:06:25.584417500 +0200
+++ luaj-3.0.1/src/core/org/luaj/vm2/lib/StringLib.java 2018-03-03 23:06:53.158652400 +0200
@@ -1066,9 +1066,12 @@
                                                        error("Missing [ after %f in pattern");
                                                }
                                                int ep = classend( poffset );
-                                               int previous = ( soffset == 0 ) ? 0 : s.luaByte( soffset - 1 );
+                                               int previous = ( soffset == 0 ) ? '\0' : s.luaByte( soffset - 1 );
+                                               // The C version does not do this check
+                                               // because in C strings are '\0' terminated
+                                               int next = ( soffset == s.length() ) ? '\0' : s.luaByte( soffset );
                                                if ( matchbracketclass( previous, poffset, ep - 1 ) ||
-                                                        matchbracketclass( s.luaByte( soffset ), poffset, ep - 1 ) )
+                                                        !matchbracketclass( next, poffset, ep - 1 ) )
                                                        return -1;
                                                poffset = ep;
                                                continue;

I would create PR once #4 is resolved.

As for the commits. This could be easily fix once the source forge is back online. I’ve created new issue to track this better #5

[plamentotev]

@Enyby to be honest I forgot the details about this issue. I may take a look at it again but not sure if I would have time soon. Feel free to use the patch and create PR if you want.