ICU-11245 Minor cleanup of RBBI Line Monkey test.

X-SVN-Rev: 36670
This commit is contained in:
Andy Heninger 2014-10-10 18:18:31 +00:00
parent 1c69528a4e
commit 5757929922

View File

@ -529,6 +529,10 @@ public class RBBITestMonkey extends TestFmwk {
List fSets;
// UnicodeSets for each of the Line Breaking character classes.
// Order matches that of Unicode UAX 14, Table 1, which makes it a little easier
// to verify that they are all accounted for.
UnicodeSet fBK;
UnicodeSet fCR;
UnicodeSet fLF;
@ -538,12 +542,12 @@ public class RBBITestMonkey extends TestFmwk {
UnicodeSet fWJ;
UnicodeSet fZW;
UnicodeSet fGL;
UnicodeSet fCB;
UnicodeSet fSP;
UnicodeSet fB2;
UnicodeSet fBA;
UnicodeSet fBB;
UnicodeSet fHY;
UnicodeSet fCB;
UnicodeSet fCL;
UnicodeSet fCP;
UnicodeSet fEX;
@ -559,15 +563,15 @@ public class RBBITestMonkey extends TestFmwk {
UnicodeSet fAI;
UnicodeSet fAL;
UnicodeSet fCJ;
UnicodeSet fH2;
UnicodeSet fH3;
UnicodeSet fHL;
UnicodeSet fID;
UnicodeSet fSA;
UnicodeSet fJL;
UnicodeSet fJV;
UnicodeSet fJT;
UnicodeSet fH2;
UnicodeSet fH3;
UnicodeSet fRI;
UnicodeSet fSA;
UnicodeSet fXX;
StringBuffer fText;
@ -577,8 +581,6 @@ public class RBBITestMonkey extends TestFmwk {
RBBILineMonkey()
{
// TODO: fiix
fCharProperty = UProperty.LINE_BREAK;
fSets = new ArrayList();
@ -587,15 +589,16 @@ public class RBBITestMonkey extends TestFmwk {
fLF = new UnicodeSet("[\\p{Line_break=LF}]");
fCM = new UnicodeSet("[\\p{Line_break=CM}]");
fNL = new UnicodeSet("[\\p{Line_break=NL}]");
fSG = new UnicodeSet("[\\ud800-\\udfff]");
fWJ = new UnicodeSet("[\\p{Line_break=WJ}]");
fZW = new UnicodeSet("[\\p{Line_break=ZW}]");
fGL = new UnicodeSet("[\\p{Line_break=GL}]");
fCB = new UnicodeSet("[\\p{Line_break=CB}]");
fSP = new UnicodeSet("[\\p{Line_break=SP}]");
fB2 = new UnicodeSet("[\\p{Line_break=B2}]");
fBA = new UnicodeSet("[\\p{Line_break=BA}]");
fBB = new UnicodeSet("[\\p{Line_break=BB}]");
fHY = new UnicodeSet("[\\p{Line_break=HY}]");
fCB = new UnicodeSet("[\\p{Line_break=CB}]");
fCL = new UnicodeSet("[\\p{Line_break=CL}]");
fCP = new UnicodeSet("[\\p{Line_break=CP}]");
fEX = new UnicodeSet("[\\p{Line_break=EX}]");
@ -611,16 +614,15 @@ public class RBBITestMonkey extends TestFmwk {
fAI = new UnicodeSet("[\\p{Line_break=AI}]");
fAL = new UnicodeSet("[\\p{Line_break=AL}]");
fCJ = new UnicodeSet("[\\p{Line_break=CJ}]");
fH2 = new UnicodeSet("[\\p{Line_break=H2}]");
fH3 = new UnicodeSet("[\\p{Line_break=H3}]");
fHL = new UnicodeSet("[\\p{Line_break=HL}]");
fID = new UnicodeSet("[\\p{Line_break=ID}]");
fSA = new UnicodeSet("[\\p{Line_break=SA}]");
fJL = new UnicodeSet("[\\p{Line_break=JL}]");
fJV = new UnicodeSet("[\\p{Line_break=JV}]");
fJT = new UnicodeSet("[\\p{Line_break=JT}]");
fH2 = new UnicodeSet("[\\p{Line_break=H2}]");
fH3 = new UnicodeSet("[\\p{Line_break=H3}]");
fSG = new UnicodeSet("[\\ud800-\\udfff]");
fRI = new UnicodeSet("[\\p{Line_break=RI}]");
fSA = new UnicodeSet("[\\p{Line_break=SA}]");
fXX = new UnicodeSet("[\\p{Line_break=XX}]");
// Remove dictionary characters.
@ -645,14 +647,12 @@ public class RBBITestMonkey extends TestFmwk {
fSets.add(fWJ);
fSets.add(fZW);
fSets.add(fGL);
fSets.add(fCB);
fSets.add(fSP);
fSets.add(fB2);
fSets.add(fBA);
fSets.add(fBB);
fSets.add(fHY);
fSets.add(fH2);
fSets.add(fH3);
fSets.add(fCB);
fSets.add(fCL);
fSets.add(fCP);
fSets.add(fEX);
@ -670,6 +670,8 @@ public class RBBITestMonkey extends TestFmwk {
fSets.add(fSY);
fSets.add(fAI);
fSets.add(fAL);
fSets.add(fH2);
fSets.add(fH3);
fSets.add(fHL);
fSets.add(fID);
fSets.add(fWJ);