ICU-20618 Regex nested lookaround expressions, clean up active match region handling.
diff --git a/icu4c/source/i18n/regexcmp.cpp b/icu4c/source/i18n/regexcmp.cpp
index 5d1eb62..cacc069 100644
--- a/icu4c/source/i18n/regexcmp.cpp
+++ b/icu4c/source/i18n/regexcmp.cpp
@@ -561,7 +561,7 @@
// sequence; don't change without making updates there too.
//
// Compiles to
- // 1 START_LA dataLoc Saves SP, Input Pos
+ // 1 LA_START dataLoc Saves SP, Input Pos, Active input region.
// 2. STATE_SAVE 4 on failure of lookahead, goto 4
// 3 JMP 6 continue ...
//
@@ -575,10 +575,14 @@
// 8. code for parenthesized stuff.
// 9. LA_END
//
- // Two data slots are reserved, for saving the stack ptr and the input position.
+ // Four data slots are reserved, for saving state on entry to the look-around
+ // 0: stack pointer on entry.
+ // 1: input position on entry.
+ // 2: fActiveStart, the active bounds start on entry.
+ // 3: fActiveLimit, the active bounds limit on entry.
{
fixLiterals();
- int32_t dataLoc = allocateData(2);
+ int32_t dataLoc = allocateData(4);
appendOp(URX_LA_START, dataLoc);
appendOp(URX_STATE_SAVE, fRXPat->fCompiledPat->size()+ 2);
appendOp(URX_JMP, fRXPat->fCompiledPat->size()+ 3);
@@ -599,18 +603,23 @@
case doOpenLookAheadNeg:
// Negated Lookahead. (?! stuff )
// Compiles to
- // 1. START_LA dataloc
+ // 1. LA_START dataloc
// 2. SAVE_STATE 7 // Fail within look-ahead block restores to this state,
// // which continues with the match.
// 3. NOP // Std. Open Paren sequence, for possible '|'
// 4. code for parenthesized stuff.
- // 5. END_LA // Cut back stack, remove saved state from step 2.
+ // 5. LA_END // Cut back stack, remove saved state from step 2.
// 6. BACKTRACK // code in block succeeded, so neg. lookahead fails.
// 7. END_LA // Restore match region, in case look-ahead was using
// an alternate (transparent) region.
+ // Four data slots are reserved, for saving state on entry to the look-around
+ // 0: stack pointer on entry.
+ // 1: input position on entry.
+ // 2: fActiveStart, the active bounds start on entry.
+ // 3: fActiveLimit, the active bounds limit on entry.
{
fixLiterals();
- int32_t dataLoc = allocateData(2);
+ int32_t dataLoc = allocateData(4);
appendOp(URX_LA_START, dataLoc);
appendOp(URX_STATE_SAVE, 0); // dest address will be patched later.
appendOp(URX_NOP, 0);
@@ -644,14 +653,16 @@
// Allocate a block of matcher data, to contain (when running a match)
// 0: Stack ptr on entry
// 1: Input Index on entry
- // 2: Start index of match current match attempt.
- // 3: Original Input String len.
+ // 2: fActiveStart, the active bounds start on entry.
+ // 3: fActiveLimit, the active bounds limit on entry.
+ // 4: Start index of match current match attempt.
+ // The first four items must match the layout of data for LA_START / LA_END
// Generate match code for any pending literals.
fixLiterals();
// Allocate data space
- int32_t dataLoc = allocateData(4);
+ int32_t dataLoc = allocateData(5);
// Emit URX_LB_START
appendOp(URX_LB_START, dataLoc);
@@ -696,14 +707,16 @@
// Allocate a block of matcher data, to contain (when running a match)
// 0: Stack ptr on entry
// 1: Input Index on entry
- // 2: Start index of match current match attempt.
- // 3: Original Input String len.
+ // 2: fActiveStart, the active bounds start on entry.
+ // 3: fActiveLimit, the active bounds limit on entry.
+ // 4: Start index of match current match attempt.
+ // The first four items must match the layout of data for LA_START / LA_END
// Generate match code for any pending literals.
fixLiterals();
// Allocate data space
- int32_t dataLoc = allocateData(4);
+ int32_t dataLoc = allocateData(5);
// Emit URX_LB_START
appendOp(URX_LB_START, dataLoc);
diff --git a/icu4c/source/i18n/regeximp.h b/icu4c/source/i18n/regeximp.h
index 51db882..590d216 100644
--- a/icu4c/source/i18n/regeximp.h
+++ b/icu4c/source/i18n/regeximp.h
@@ -123,7 +123,7 @@
// saved input position, FAIL rather than taking
// the JMP
URX_LA_START = 37, // Starting a LookAround expression.
- // Save InputPos and SP in static data.
+ // Save InputPos, SP and active region in static data.
// Operand: Static data offset for the save
URX_LA_END = 38, // Ending a Lookaround expression.
// Restore InputPos and Stack to saved values.
diff --git a/icu4c/source/i18n/rematch.cpp b/icu4c/source/i18n/rematch.cpp
index ef06c21..6d6ea0f 100644
--- a/icu4c/source/i18n/rematch.cpp
+++ b/icu4c/source/i18n/rematch.cpp
@@ -3805,11 +3805,13 @@
case URX_LA_START:
{
- // Entering a lookahead block.
+ // Entering a look around block.
// Save Stack Ptr, Input Pos.
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
+ U_ASSERT(opValue>=0 && opValue+3<fPattern->fDataSize);
fData[opValue] = fStack->size();
fData[opValue+1] = fp->fInputIdx;
+ fData[opValue+2] = fActiveStart;
+ fData[opValue+3] = fActiveLimit;
fActiveStart = fLookStart; // Set the match region change for
fActiveLimit = fLookLimit; // transparent bounds.
}
@@ -3819,7 +3821,7 @@
{
// Leaving a look-ahead block.
// restore Stack Ptr, Input Pos to positions they had on entry to block.
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
+ U_ASSERT(opValue>=0 && opValue+3<fPattern->fDataSize);
int32_t stackSize = fStack->size();
int32_t newStackSize =(int32_t)fData[opValue];
U_ASSERT(stackSize >= newStackSize);
@@ -3839,8 +3841,10 @@
// Restore the active region bounds in the input string; they may have
// been changed because of transparent bounds on a Region.
- fActiveStart = fRegionStart;
- fActiveLimit = fRegionLimit;
+ fActiveStart = fData[opValue+2];
+ fActiveLimit = fData[opValue+3];
+ U_ASSERT(fActiveStart >= 0);
+ U_ASSERT(fActiveLimit <= fInputLength);
}
break;
@@ -3916,17 +3920,19 @@
case URX_LB_START:
{
// Entering a look-behind block.
- // Save Stack Ptr, Input Pos.
+ // Save Stack Ptr, Input Pos and active input region.
// TODO: implement transparent bounds. Ticket #6067
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
+ U_ASSERT(opValue>=0 && opValue+4<fPattern->fDataSize);
fData[opValue] = fStack->size();
fData[opValue+1] = fp->fInputIdx;
- // Init the variable containing the start index for attempted matches.
- fData[opValue+2] = -1;
// Save input string length, then reset to pin any matches to end at
// the current position.
+ fData[opValue+2] = fActiveStart;
fData[opValue+3] = fActiveLimit;
+ fActiveStart = fRegionStart;
fActiveLimit = fp->fInputIdx;
+ // Init the variable containing the start index for attempted matches.
+ fData[opValue+4] = -1;
}
break;
@@ -3949,8 +3955,8 @@
U_ASSERT(minML >= 0);
// Fetch (from data) the last input index where a match was attempted.
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
- int64_t &lbStartIdx = fData[opValue+2];
+ U_ASSERT(opValue>=0 && opValue+4<fPattern->fDataSize);
+ int64_t &lbStartIdx = fData[opValue+4];
if (lbStartIdx < 0) {
// First time through loop.
lbStartIdx = fp->fInputIdx - minML;
@@ -3976,10 +3982,10 @@
// getting a match. Backtrack out, and out of the
// Look Behind altogether.
fp = (REStackFrame *)fStack->popFrame(fFrameSize);
- int64_t restoreInputLen = fData[opValue+3];
- U_ASSERT(restoreInputLen >= fActiveLimit);
- U_ASSERT(restoreInputLen <= fInputLength);
- fActiveLimit = restoreInputLen;
+ fActiveStart = fData[opValue+2];
+ fActiveLimit = fData[opValue+3];
+ U_ASSERT(fActiveStart >= 0);
+ U_ASSERT(fActiveLimit <= fInputLength);
break;
}
@@ -3993,7 +3999,7 @@
case URX_LB_END:
// End of a look-behind block, after a successful match.
{
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
+ U_ASSERT(opValue>=0 && opValue+4<fPattern->fDataSize);
if (fp->fInputIdx != fActiveLimit) {
// The look-behind expression matched, but the match did not
// extend all the way to the point that we are looking behind from.
@@ -4004,13 +4010,13 @@
break;
}
- // Look-behind match is good. Restore the orignal input string length,
+ // Look-behind match is good. Restore the orignal input string region,
// which had been truncated to pin the end of the lookbehind match to the
// position being looked-behind.
- int64_t originalInputLen = fData[opValue+3];
- U_ASSERT(originalInputLen >= fActiveLimit);
- U_ASSERT(originalInputLen <= fInputLength);
- fActiveLimit = originalInputLen;
+ fActiveStart = fData[opValue+2];
+ fActiveLimit = fData[opValue+3];
+ U_ASSERT(fActiveStart >= 0);
+ U_ASSERT(fActiveLimit <= fInputLength);
}
break;
@@ -4035,8 +4041,8 @@
U_ASSERT(continueLoc > fp->fPatIdx);
// Fetch (from data) the last input index where a match was attempted.
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
- int64_t &lbStartIdx = fData[opValue+2];
+ U_ASSERT(opValue>=0 && opValue+4<fPattern->fDataSize);
+ int64_t &lbStartIdx = fData[opValue+4];
if (lbStartIdx < 0) {
// First time through loop.
lbStartIdx = fp->fInputIdx - minML;
@@ -4061,10 +4067,10 @@
// We have tried all potential match starting points without
// getting a match, which means that the negative lookbehind as
// a whole has succeeded. Jump forward to the continue location
- int64_t restoreInputLen = fData[opValue+3];
- U_ASSERT(restoreInputLen >= fActiveLimit);
- U_ASSERT(restoreInputLen <= fInputLength);
- fActiveLimit = restoreInputLen;
+ fActiveStart = fData[opValue+2];
+ fActiveLimit = fData[opValue+3];
+ U_ASSERT(fActiveStart >= 0);
+ U_ASSERT(fActiveLimit <= fInputLength);
fp->fPatIdx = continueLoc;
break;
}
@@ -4079,7 +4085,7 @@
case URX_LBN_END:
// End of a negative look-behind block, after a successful match.
{
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
+ U_ASSERT(opValue>=0 && opValue+4<fPattern->fDataSize);
if (fp->fInputIdx != fActiveLimit) {
// The look-behind expression matched, but the match did not
// extend all the way to the point that we are looking behind from.
@@ -4096,10 +4102,10 @@
// Restore the orignal input string length, which had been truncated
// inorder to pin the end of the lookbehind match
// to the position being looked-behind.
- int64_t originalInputLen = fData[opValue+3];
- U_ASSERT(originalInputLen >= fActiveLimit);
- U_ASSERT(originalInputLen <= fInputLength);
- fActiveLimit = originalInputLen;
+ fActiveStart = fData[opValue+2];
+ fActiveLimit = fData[opValue+3];
+ U_ASSERT(fActiveStart >= 0);
+ U_ASSERT(fActiveLimit <= fInputLength);
// Restore original stack position, discarding any state saved
// by the successful pattern match.
@@ -5336,11 +5342,13 @@
case URX_LA_START:
{
- // Entering a lookahead block.
+ // Entering a look around block.
// Save Stack Ptr, Input Pos.
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
+ U_ASSERT(opValue>=0 && opValue+3<fPattern->fDataSize);
fData[opValue] = fStack->size();
fData[opValue+1] = fp->fInputIdx;
+ fData[opValue+2] = fActiveStart;
+ fData[opValue+3] = fActiveLimit;
fActiveStart = fLookStart; // Set the match region change for
fActiveLimit = fLookLimit; // transparent bounds.
}
@@ -5348,9 +5356,9 @@
case URX_LA_END:
{
- // Leaving a look-ahead block.
+ // Leaving a look around block.
// restore Stack Ptr, Input Pos to positions they had on entry to block.
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
+ U_ASSERT(opValue>=0 && opValue+3<fPattern->fDataSize);
int32_t stackSize = fStack->size();
int32_t newStackSize = (int32_t)fData[opValue];
U_ASSERT(stackSize >= newStackSize);
@@ -5370,8 +5378,10 @@
// Restore the active region bounds in the input string; they may have
// been changed because of transparent bounds on a Region.
- fActiveStart = fRegionStart;
- fActiveLimit = fRegionLimit;
+ fActiveStart = fData[opValue+2];
+ fActiveLimit = fData[opValue+3];
+ U_ASSERT(fActiveStart >= 0);
+ U_ASSERT(fActiveLimit <= fInputLength);
}
break;
@@ -5434,17 +5444,19 @@
case URX_LB_START:
{
// Entering a look-behind block.
- // Save Stack Ptr, Input Pos.
+ // Save Stack Ptr, Input Pos and active input region.
// TODO: implement transparent bounds. Ticket #6067
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
+ U_ASSERT(opValue>=0 && opValue+4<fPattern->fDataSize);
fData[opValue] = fStack->size();
fData[opValue+1] = fp->fInputIdx;
- // Init the variable containing the start index for attempted matches.
- fData[opValue+2] = -1;
// Save input string length, then reset to pin any matches to end at
// the current position.
+ fData[opValue+2] = fActiveStart;
fData[opValue+3] = fActiveLimit;
+ fActiveStart = fRegionStart;
fActiveLimit = fp->fInputIdx;
+ // Init the variable containing the start index for attempted matches.
+ fData[opValue+4] = -1;
}
break;
@@ -5462,8 +5474,8 @@
U_ASSERT(minML >= 0);
// Fetch (from data) the last input index where a match was attempted.
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
- int64_t &lbStartIdx = fData[opValue+2];
+ U_ASSERT(opValue>=0 && opValue+4<fPattern->fDataSize);
+ int64_t &lbStartIdx = fData[opValue+4];
if (lbStartIdx < 0) {
// First time through loop.
lbStartIdx = fp->fInputIdx - minML;
@@ -5485,10 +5497,10 @@
// getting a match. Backtrack out, and out of the
// Look Behind altogether.
fp = (REStackFrame *)fStack->popFrame(fFrameSize);
- int64_t restoreInputLen = fData[opValue+3];
- U_ASSERT(restoreInputLen >= fActiveLimit);
- U_ASSERT(restoreInputLen <= fInputLength);
- fActiveLimit = restoreInputLen;
+ fActiveStart = fData[opValue+2];
+ fActiveLimit = fData[opValue+3];
+ U_ASSERT(fActiveStart >= 0);
+ U_ASSERT(fActiveLimit <= fInputLength);
break;
}
@@ -5502,7 +5514,7 @@
case URX_LB_END:
// End of a look-behind block, after a successful match.
{
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
+ U_ASSERT(opValue>=0 && opValue+4<fPattern->fDataSize);
if (fp->fInputIdx != fActiveLimit) {
// The look-behind expression matched, but the match did not
// extend all the way to the point that we are looking behind from.
@@ -5513,13 +5525,13 @@
break;
}
- // Look-behind match is good. Restore the orignal input string length,
+ // Look-behind match is good. Restore the orignal input string region,
// which had been truncated to pin the end of the lookbehind match to the
// position being looked-behind.
- int64_t originalInputLen = fData[opValue+3];
- U_ASSERT(originalInputLen >= fActiveLimit);
- U_ASSERT(originalInputLen <= fInputLength);
- fActiveLimit = originalInputLen;
+ fActiveStart = fData[opValue+2];
+ fActiveLimit = fData[opValue+3];
+ U_ASSERT(fActiveStart >= 0);
+ U_ASSERT(fActiveLimit <= fInputLength);
}
break;
@@ -5539,8 +5551,8 @@
U_ASSERT(continueLoc > fp->fPatIdx);
// Fetch (from data) the last input index where a match was attempted.
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
- int64_t &lbStartIdx = fData[opValue+2];
+ U_ASSERT(opValue>=0 && opValue+4<fPattern->fDataSize);
+ int64_t &lbStartIdx = fData[opValue+4];
if (lbStartIdx < 0) {
// First time through loop.
lbStartIdx = fp->fInputIdx - minML;
@@ -5561,10 +5573,10 @@
// We have tried all potential match starting points without
// getting a match, which means that the negative lookbehind as
// a whole has succeeded. Jump forward to the continue location
- int64_t restoreInputLen = fData[opValue+3];
- U_ASSERT(restoreInputLen >= fActiveLimit);
- U_ASSERT(restoreInputLen <= fInputLength);
- fActiveLimit = restoreInputLen;
+ fActiveStart = fData[opValue+2];
+ fActiveLimit = fData[opValue+3];
+ U_ASSERT(fActiveStart >= 0);
+ U_ASSERT(fActiveLimit <= fInputLength);
fp->fPatIdx = continueLoc;
break;
}
@@ -5579,7 +5591,7 @@
case URX_LBN_END:
// End of a negative look-behind block, after a successful match.
{
- U_ASSERT(opValue>=0 && opValue+1<fPattern->fDataSize);
+ U_ASSERT(opValue>=0 && opValue+4<fPattern->fDataSize);
if (fp->fInputIdx != fActiveLimit) {
// The look-behind expression matched, but the match did not
// extend all the way to the point that we are looking behind from.
@@ -5596,10 +5608,10 @@
// Restore the orignal input string length, which had been truncated
// inorder to pin the end of the lookbehind match
// to the position being looked-behind.
- int64_t originalInputLen = fData[opValue+3];
- U_ASSERT(originalInputLen >= fActiveLimit);
- U_ASSERT(originalInputLen <= fInputLength);
- fActiveLimit = originalInputLen;
+ fActiveStart = fData[opValue+2];
+ fActiveLimit = fData[opValue+3];
+ U_ASSERT(fActiveStart >= 0);
+ U_ASSERT(fActiveLimit <= fInputLength);
// Restore original stack position, discarding any state saved
// by the successful pattern match.
diff --git a/icu4c/source/test/intltest/regextst.cpp b/icu4c/source/test/intltest/regextst.cpp
index db66048..b6391d2 100644
--- a/icu4c/source/test/intltest/regextst.cpp
+++ b/icu4c/source/test/intltest/regextst.cpp
@@ -3525,11 +3525,16 @@
}
}
parseMatcher->appendTail(deTaggedInput);
- REGEX_ASSERT_L(groupStarts.size() == groupEnds.size(), line);
+
+ if (groupStarts.size() != groupEnds.size()) {
+ errln("Error at line %d: mismatched <n> group tags in expected results.", line);
+ failed = true;
+ goto cleanupAndReturn;
+ }
if ((regionStart>=0 || regionEnd>=0) && (regionStart<0 || regionStart>regionEnd)) {
- errln("mismatched <r> tags");
- failed = TRUE;
- goto cleanupAndReturn;
+ errln("mismatched <r> tags");
+ failed = TRUE;
+ goto cleanupAndReturn;
}
//
diff --git a/icu4c/source/test/testdata/regextst.txt b/icu4c/source/test/testdata/regextst.txt
index 759a196..9135612 100644
--- a/icu4c/source/test/testdata/regextst.txt
+++ b/icu4c/source/test/testdata/regextst.txt
@@ -89,6 +89,28 @@
"abc(?!def)" b "<r><0>abc</0></r>xyz"
#
+# Nested Lookahead / Behind
+#
+"one(?=(?:(?!<out>).)*</out>)" "<out><0>one</0> stuff</out>"
+"one(?=(?:(?!<out>).)*</out>)" "<out>one <out></out>"
+
+# More nesting lookaround: pattern matches "qq" when not preceded by 'a' and followed by 'z'
+"(?<!a(?!...z))qq" "<0>qq</0>c"
+"(?<!a(?!...z))qq" "f<0>qq</0>c"
+"(?<!a(?!...z))qq" "aqqz"
+
+# More nested lookaround: match any two chars preceded and followed by an upper case letter.
+# With gratuitious nesting of look-arounds and capture from the look-arounds.
+
+"(?=(?<=(\p{Lu})(?=..(\p{Lu})))).." "<1>A</1><0>jk</0><2>B</2>"
+"(?=(?<=(\p{Lu})(?=..(\p{Lu})))).." "ajkB"
+"(?=(?<=(\p{Lu})(?=..(\p{Lu})))).." "Ajkb"
+
+# Nested lookaround cases from bug ICU-20564
+"(?<=(?<=((?=)){0}+))" "<0></0>abc"
+"(?<=c(?<=c((?=c)){1}+))" "c<0><1></1></0>cc"
+
+#
# Anchoring Bounds
#
"^def$" "abc<r><0>def</0></r>ghi" # anchoring (default) bounds
@@ -1456,11 +1478,14 @@
"abc(?=de(?=f))...g" "<0>abcdefg</0>"
"abc(?=de(?=f))...g" "abcdxfg"
+# Bug ICU-20618 Assertion failure with nested look-around expressions.
+#
+"(?<=(?<=b?(?=a)))" "hello, world."
+
# Random debugging, Temporary
#
-
#
# Regexps from http://www.regexlib.com
#