[Xunfop]

As know IEEE 745 floating has precision problem, is it possible to integrate fix point from pawn?

[Black Rose]

I'm guessing that would require a huge amount of work to replace, if possible at all. So what is the problem with using it as is?

[Xunfop]

For correct floating mathematical.

[Black Rose]

For what reason? It is pretty accurate as is.
Why would you need it more accurate?

You should probably give a good example if you expect anything done.

If you want to do an equal check on a value that is near you could just knock off some of the last couple of bits to make them irrelevant without deviating much from the original value. The best part being of course that it scales. Doesn't work as I expected.

[Xunfop]

For kreedz time system, it count to 2nd decimal place, after it store the floating, the value goes wrong. It also make the mathematical value wrong.

[Black Rose]

You might see deviation at the 7th decimal (which isn't displayed by default). It may affect previous decimals if right on the edge. For example 0.1 might be 0.999999.
I don't know what Kreedz time system is but it sounds like a badly written plugin IMO.

Don't store values in float in any long time storage, round them to integers. If you want decimal points, multiply them by 100 first for 2 decimal points and divide them by 100 on load.
Same can be applied to strings. Turn them into strings and then shove a decimal point at the correct position.

Yes it's crude, but if we speak about efficiency it's completely trivial and won't matter at all unless you call it on every frame.

If you need a better answer than that, share the code.

Edit: Here are some examples of outputs using float_to_str() or a multiplied number used with num_to_str() with carefully placed decimal point at string level.
Both seems to work fine.

    new Float:test1 = -2.8;
    new Float:test2 = -2.9 + 0.1;

    server_print("%.2f", test1);
    server_print("%.2f", test2);





    new test3 = floatround(test1 * 100);
    new test4 = floatround(test2 * 100);

    new str1[32], str2[6];
    new len = num_to_str(test3, str1, charsmax(str1));
    str1[len + 1] = 0;
    str1[len] = str1[len - 1]
    str1[len - 1] = str1[len - 2]
    str1[len - 2] = '.';

    len = num_to_str(test4, str2, charsmax(str2));
    if ( len < charsmax(str2) ) {
        str2[len + 1] = 0;
        str2[len] = str2[len - 1]
        str2[len - 1] = str2[len - 2]
        str2[len - 2] = '.';
    }
    
    server_print("%s", str1);
    server_print("%s", str2);





    new str3[32], str4[32];
    float_to_str(test1, str3, charsmax(str3));
    float_to_str(test2, str4, charsmax(str4));

    new pos;
    if ( ( pos = strfind(str3, ".") ) != -1 )
        str3[pos + 3] = 0;

    if ( ( pos = strfind(str4, ".") ) != -1 )
        str4[pos + 3] = 0;

    server_print("%s", str3);
    server_print("%s", str4);

-2.79
-2.80
-2.80
-2.80
-2.80
-2.80

[fysiks]

Originally Posted by Xunfop For kreedz time system, it count to 2nd decimal place, after it store the floating, the value goes wrong. It also make the mathematical value wrong.

You'll have to be more specific with how you are actually using these numbers. If you only need accuracy to the hundredths place then there is no issue with using floating point numbers because their accuracy is much much better than that. Like I said before, it is probably an issue with displaying only (which include writing it to a file in any non-binary form).

[Xunfop]

Here's the code.

PHP Code:

public fnLoadWorldRecord() {
    g_bHasWR = false;
    
    for(new i = 0; i < sizeof(g_sSaveDemosFile); i++) {
        if(!g_bHasWR && file_exists(g_sSaveDemosFile[i])) {
            new hFile = fopen(g_sSaveDemosFile[i], "r");
            new sData[96];
            
            while(!feof(hFile)) {
                fgets(hFile, sData, charsmax(sData));
                
                if(!sData[0] || sData[0] == '^n' || equali(sData, g_sFirstDemosLine[i]) || !equali(sData, g_sMapName, g_iMapNameLen)) {
                    if(g_bHasWR) {
                        break;
                    }
                    else {
                        continue;
                    }
                }
                
                new sMapName[64], sTime[10], sAuthor[32], sCountry[2];
                
                trim(sData);
                
                if(i == 0) {
                    parse(sData, sMapName, charsmax(sMapName), sTime, charsmax(sTime), sAuthor, charsmax(sAuthor), sCountry, charsmax(sCountry));
                }
                else {
                    new sMapId[11], sTimeStamp[11], sAuthorId[11];
                    
                    parse(sData, sMapName, charsmax(sMapName), sTime, charsmax(sTime), sMapId, charsmax(sMapId), sTimeStamp, charsmax(sTimeStamp), sAuthorId, charsmax(sAuthorId), sCountry, charsmax(sCountry), sAuthor, charsmax(sAuthor));
                }
                
                if(sMapName[g_iMapNameLen] != '[' && g_iMapNameLen != strlen(sMapName)) {
                    if(g_bHasWR) {
                        break;
                    }
                    else {
                        continue;
                    }
                }
                
                if(!g_bHasWR) {
                    g_bHasWR = true;
                }
                
                if(sMapName[g_iMapNameLen] == '[') {
                    if(isdigit(sMapName[g_iMapNameLen + 1])) {
                        formatex(g_sWRExt[g_iWRAuthorsNum], charsmax(g_sWRExt[]), "[%s", sMapName[g_iMapNameLen + 1]);
                    }
                    else {
                        formatex(g_sWRExt[g_iWRAuthorsNum], charsmax(g_sWRExt[]), "[%c]", sMapName[g_iMapNameLen + 1]);
                    }
                }
                
                g_fWRTimes[g_iWRAuthorsNum] = str_to_float(sTime);
                g_sWRAuthors[g_iWRAuthorsNum] = sAuthor;
                
                if(g_iWRAuthorsNum >= 1) {
                    if((g_fWRMinTime <= 0.0 && g_fWRTimes[g_iWRAuthorsNum] > 0.0) || 
                      (0.0 < g_fWRTimes[g_iWRAuthorsNum] < g_fWRMinTime)) {
                            g_fWRMinTime = g_fWRTimes[g_iWRAuthorsNum];
                    }
                }
                else {
                    g_fWRMinTime = g_fWRTimes[g_iWRAuthorsNum];
                }
                
                g_iWRAuthorsNum++;
                
                continue;
            }
        }
    }
    
    if(g_bHasWR) {
        fnGenerateWRString();
    }
    
    return PLUGIN_HANDLED;
}

public fnGenerateWRString() {
    for(new i = 0; i < g_iWRAuthorsNum; i++) {
        if(i > 0) {
            add(g_sWorldRecords, charsmax(g_sWorldRecords), ", ");
        }
        
        new sWRTime[11];
        
        fnConvertTime("float", g_fWRTimes[i], sWRTime, charsmax(sWRTime));
        
        add(g_sWorldRecords, charsmax(g_sWorldRecords), g_sWRExt[i]);
        
        if(equal(sWRTime, "00:00.00")) {
            add(g_sWorldRecords, charsmax(g_sWorldRecords), "^x03");
        }
        else {
            add(g_sWorldRecords, charsmax(g_sWorldRecords), "^x04 ");
            add(g_sWorldRecords, charsmax(g_sWorldRecords), sWRTime);
        }
        
        add(g_sWorldRecords, charsmax(g_sWorldRecords), " ");
        add(g_sWorldRecords, charsmax(g_sWorldRecords), g_sWRAuthors[i]);
        add(g_sWorldRecords, charsmax(g_sWorldRecords), "^x01");
    }
    
    g_sWorldRecords[160] = 0;
    
    return PLUGIN_HANDLED;
}

public fnConvertTime(type[], Float:time, convert_time[], len) {
    new sTemp[11];
    new Float:fSeconds = time, iMinutes;
    
    iMinutes = floatround(fSeconds / 60, floatround_floor);
    fSeconds -= iMinutes * 60;
    
    if(equali(type, "float")) {
        formatex(sTemp, charsmax(sTemp), "%02i:%05.2f", iMinutes, fSeconds);
    }
    else {
        formatex(sTemp, charsmax(sTemp), "%02i:%02.0f", iMinutes, fSeconds);
        sTemp[strlen(sTemp) - 1] = 0;
    }
    
    formatex(convert_time, len, sTemp);
    
    return PLUGIN_HANDLED;
} 


Reading file format.

8b1_brickngrass 231.68 Nukk ru 642

How to fix this with -0.01 problem?

And probably should move this thread to scripting help area?

[Black Rose]

You have to give some pointers or create some kind of function that I could test it with.
I don't really want to spend time creating a wrapper for this code just to try it out.
Sorry for being an asshole here but I don't know what you want and what the purpose of the code is.

Where exactly in the code is the problem. What value are you getting? What value do you expect?

If your problem is saving to/reading from a file with 2 decimals, save with 3 decimals or more. Every time you save without preformatting the float, you risk it being off by .01 when you save with two decimals. 3 decimals: .001 and so on...

[Powerlord]

Originally Posted by Black Rose I'm guessing that would require a huge amount of work to replace, if possible at all. So what is the problem with using it as is?

I imagine it would since AMX and SM both just wrap the Pawn Floating Point Support Library (PDF).

In fact, I don't know enough about infinite precision math libs to know if they'd be feasible in Pawn.

Having said that, you can fake it by using cells with whole values and just displaying them as if they were decimals. Maybe using 16 bits as the fractional part and 16 bits as the whole part? (One of them would have to include the sign bit, though)