RE2::GlobalReplace case insensitive fail with OR-ed pattern AB|AC

[bigF93]

Hi,
it seems that the commit b2af9b3 has changed the behaviour of the case insensitive call to GlobalReplace.
Before b2af9b3 , the replace call was successful, but now it seems to have stopped working.
I am using ubuntu22 with g++.

To reproduce the issue, you can build and run the code snippet below (include iostream and iomanip which are not rendered by github)

#include <re2/re2.h>

int main()
{

// example of an array of bytes from which we want to remove some couples of bytes, like 0xa5 0xd1 and 0xa5 0x64
unsigned char rawBytes[] = { 0xa5, 0xd1, 0xa5, 0xd1, 0x61, 0x63, 0xa5, 0x64 };

std::string constructedString( reinterpret_cast<char*>( rawBytes ), sizeof( rawBytes )/sizeof( rawBytes[0] ) );

for ( const auto& elem : constructedString ) {
                std::cout << "0x" << std::hex << std::setw(2) << std::setfill('0') << static_cast<int>(static_cast<unsigned char>(elem)) << " ";
}
std::cout << "\n";

RE2::Options opts{};
opts.set_encoding( RE2::Options::Encoding::EncodingLatin1 );
opts.set_case_sensitive( false );
opts.set_posix_syntax( false );
opts.set_perl_classes( false );
RE2 regex{ "\xa5\xd1|\xa5\x64", opts };

int result = RE2::GlobalReplace( &constructedString, regex , "" );

for ( const uint8_t elem : constructedString) {
                std::cout << "0x" << std::hex << std::setw(2) << std::setfill('0') << static_cast<int>(static_cast<unsigned char>(elem)) << " ";
}

std::cout << "\nReplacements made = " << result << "\n";
return 0;

}

thank you!

[junyer]

Many thanks for the report! It seems that you have uncovered an ancient bug: this line fails to preserve the Latin1 flag. The case sensitivity doesn't appear to matter and, as such, I'm simplifying the options in the test that I'm committing with the fix.

[junyer]

I noticed some additional, seemingly long-standing sadness around Latin-1 handling that does involve case sensitivity. T_T