[FEA] NVStrings regex Refactorings

[codereport]

After discussing with David, seeing as the NVStrings regex op is a very large port - any non-local refactorings or changes should be pushed to a follow up clean up issue (this issue). List will be updated below:

• rename LBRA, RBRA and LBRA_NC to be PAREN
• regex::print case statement consolidation
• change reprog_device::find to return result, instead of have output parameters; this:

__device__ inline int32_t reprog_device::find( int32_t idx, string_view const& dstr, int32_t& begin, int32_t& end )

leads to this unnecessarily confusing code:

        string_view d_str = d_strings.element<string_view>(idx);
        int32_t find_count = 0;
        size_type nchars = d_str.length();
        size_type begin = 0;
        while( begin <= nchars )
        {
            auto end = nchars;
            if( prog.find(idx,d_str,begin,end) <=0 )
                break;
            ++find_count;
            begin = end > begin ? end : begin + 1;
        }
        return find_count;

• Node op1, op2; should be initialized to some invalid state or the design should be changed
• Change this for loop to the any_of algorithm with strided adaptor or iterators

    for(int i = 0; i < count; i += 2) {
        if( (ch >= literals[i]) && (ch <= literals[i+1]) )
            return true;
    }

• remove bool nextc(char32_t& c) output parameter, and then we can directly initialise uninitialised variables:

        char32_t c;
        int quoted = nextc(c);

and

        int quoted;
        quoted = nextc(yy);

• clean this up:

char buff[8];
buff[0] = 0;
for (int i = 0; i < 7 && *exprp != '}' && *exprp != ',' && *exprp != 0; i++, exprp++)
{
    buff[i] = *exprp;
    buff[i + 1] = 0;
}

and this:

buff[0] = 0;
for (int i = 0; i < 7 && *exprp != '}' && *exprp != 0; i++, exprp++)
{
    buff[i] = *exprp;
    buff[i + 1] = 0;
}

• give Item a non-default explicit destructor and clean this up:

            Item item;
            item.t = token;
            if (token == CCLASS || token == NCCLASS)
                item.d.yyclass_id = yyclass_id;
            else if (token == COUNTED || token == COUNTED_LAZY)
            {
                item.d.yycount.n = yy_min_count;
                item.d.yycount.m = yy_max_count;
                m_has_counted = true;
            }
            else
                item.d.yy = yy;

• Put const everywhere
• std::vector<int> stack; - just use a stack
• come up with better names:

struct reljunk;
struct reinst;

• clean this up (in tandem with to_char_utf8)

std::vector<char32_t> string_to_char32_vector( std::string const& pattern )
{
    size_type size = static_cast<size_type>(pattern.size());
    size_type count = characters_in_string(pattern.c_str(),size);
    std::vector<char32_t> result(count+1);
    char32_t* output_ptr = result.data();
    const char* input_ptr = pattern.data();
    for( size_type idx=0; idx < size; ++idx )
    {
        char_utf8 output_character = 0;
        size_type ch_width = to_char_utf8(input_ptr,output_character);
        input_ptr += ch_width;
        idx += ch_width - 1;
        *output_ptr++ = output_character;
    }
    result[count] = 0; // last entry set to 0
    return result;
}

• use std::accumulate:

    auto classes_size = classes_count * sizeof(_classes[0]);
    for( int32_t idx=0; idx < classes_count; ++idx )
        classes_size += static_cast<int32_t>((h_prog.class_at(idx).literals.size())*sizeof(char32_t));

• remove goto BreakFor;
• rename ators and ands : [REVIEW] Port NVStrings regex contains ops #3292 (comment)
• rename Operand and Operator

[codereport]

from reviewing #3409 definitely refactor reprog_device::find

[kkraus14]

@davidwendt @codereport is this still relevant?

[codereport]

Yes, this is just a tech debt job - so not a top priority. Will put in backlog.

[davidwendt]

All or most of this has been addressed. We can open a new issue for any new or remaining concerns.