Commit 73da478c authored by Iain Ireland's avatar Iain Ireland Committed by Commit Bot

[regexp] Remove function pointer in TextEmitPass

TextEmitPass uses a function pointer to determine which pass to
call. This function pointer is only assigned inside TextEmitPass, and
can easily be eliminated by moving the calls to each possible target
inside the switch statement that assigns the function pointer.

I made this change because SpiderMonkey uses a static analysis pass to
verify that everything is rooted properly across calls that might GC,
and that analysis is conservative when calling function pointers. We
can white-list function pointers that are known to be safe, but the
code being called through this function pointer is complex enough
(and the function pointer is unnecessary enough) that it seemed best
to just remove the function pointer entirely.

R=jgruber@chromium.org

Bug: v8:10303
Change-Id: I5fbb0df290a2288c4d3db6d43a563385337162ea
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2091398
Commit-Queue: Jakob Gruber <jgruber@chromium.org>
Reviewed-by: 's avatarJakob Gruber <jgruber@chromium.org>
Cr-Commit-Position: refs/heads/master@{#66672}
parent 025319bf
......@@ -862,10 +862,6 @@ static bool ShortCutEmitCharacterPair(RegExpMacroAssembler* macro_assembler,
return false;
}
using EmitCharacterFunction = bool(Isolate* isolate, RegExpCompiler* compiler,
uc16 c, Label* on_failure, int cp_offset,
bool check, bool preloaded);
// Only emits letters (things that have case). Only used for case independent
// matches.
static inline bool EmitAtomLetter(Isolate* isolate, RegExpCompiler* compiler,
......@@ -2314,7 +2310,6 @@ void TextNode::TextEmitPass(RegExpCompiler* compiler, TextEmitPassType pass,
for (int j = preloaded ? 0 : quarks.length() - 1; j >= 0; j--) {
if (first_element_checked && i == 0 && j == 0) continue;
if (DeterminedAlready(quick_check, elm.cp_offset() + j)) continue;
EmitCharacterFunction* emit_function = nullptr;
uc16 quark = quarks[j];
if (elm.atom()->ignore_case()) {
// Everywhere else we assume that a non-Latin-1 character cannot match
......@@ -2322,6 +2317,9 @@ void TextNode::TextEmitPass(RegExpCompiler* compiler, TextEmitPassType pass,
// invalid by using the Latin1 equivalent instead.
quark = unibrow::Latin1::TryConvertToLatin1(quark);
}
bool needs_bounds_check =
*checked_up_to < cp_offset + j || read_backward();
bool bounds_checked = false;
switch (pass) {
case NON_LATIN1_MATCH:
DCHECK(one_byte);
......@@ -2331,24 +2329,24 @@ void TextNode::TextEmitPass(RegExpCompiler* compiler, TextEmitPassType pass,
}
break;
case NON_LETTER_CHARACTER_MATCH:
emit_function = &EmitAtomNonLetter;
bounds_checked =
EmitAtomNonLetter(isolate, compiler, quark, backtrack,
cp_offset + j, needs_bounds_check, preloaded);
break;
case SIMPLE_CHARACTER_MATCH:
emit_function = &EmitSimpleCharacter;
bounds_checked = EmitSimpleCharacter(isolate, compiler, quark,
backtrack, cp_offset + j,
needs_bounds_check, preloaded);
break;
case CASE_CHARACTER_MATCH:
emit_function = &EmitAtomLetter;
bounds_checked =
EmitAtomLetter(isolate, compiler, quark, backtrack,
cp_offset + j, needs_bounds_check, preloaded);
break;
default:
break;
}
if (emit_function != nullptr) {
bool bounds_check = *checked_up_to < cp_offset + j || read_backward();
bool bound_checked =
emit_function(isolate, compiler, quark, backtrack, cp_offset + j,
bounds_check, preloaded);
if (bound_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to);
}
if (bounds_checked) UpdateBoundsCheck(cp_offset + j, checked_up_to);
}
} else {
DCHECK_EQ(TextElement::CHAR_CLASS, elm.text_type());
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment