How to replace/ignore invalid Unicode/UTF8 charact

Question

On Python, there is this option errors='ignore' for the open Python function:

open( '/filepath.txt', 'r', encoding='UTF-8', errors='ignore' )

With this, reading a file with invalid UTF8 characters will replace them with nothing, i.e., they are ignored. For example, a file with the characthers Føö»BÃ¥r is going to be read as FøöBår.

If a line as Føö»BÃ¥r is read with getline() from stdio.h, it will be read as Føö�Bår:

FILE* cfilestream = fopen( "/filepath.txt", "r" );
int linebuffersize = 131072;
char* readline = (char*) malloc( linebuffersize );

while( true )
{
    if( getline( &readline, &linebuffersize, cfilestream ) != -1 ) {
        std::cerr << "readline=" readline << std::endl;
    }
    else {
        break;
    }
}

How can I make stdio.h getline() read it as FøöBår instead of Føö�Bår, i..e, ignoring invalid UTF8 characters?

One overwhelming solution I can think of it do iterate throughout all characters on each line read and build a new readline without any of these characters. For example:

FILE* cfilestream = fopen( "/filepath.txt", "r" );
int linebuffersize = 131072;
char* readline = (char*) malloc( linebuffersize );
char* fixedreadline = (char*) malloc( linebuffersize );

int index;
int charsread;
int invalidcharsoffset;

while( true )
{
    if( ( charsread = getline( &readline, &linebuffersize, cfilestream ) ) != -1 )
    {
        invalidcharsoffset = 0;
        for( index = 0; index < charsread; ++index )
        {
            if( readline[index] != '�' ) {
                fixedreadline[index-invalidcharsoffset] = readline[index];
            } 
            else {
                ++invalidcharsoffset;
            }
        }
        std::cerr << "fixedreadline=" << fixedreadline << std::endl;
    }
    else {
        break;
    }
}

Related questions:

1. Fixing invalid UTF8 characters
2. Replacing non UTF8 characters
3. python replace unicode characters
4. Python unicode: how to replace character that cannot be decoded using utf8 with whitespace?

Answer 1

You are confusing what you see with what is really going on. The getline function does not do any replacement of characters. [Note 1]

You are seeing a replacement character (U+FFFD) because your console outputs that character when it is asked to render an invalid UTF-8 code. Most consoles will do that if they are in UTF-8 mode; that is, the current locale is UTF-8.

Also, saying that a file contains the "characters Føö»BÃ¥r" is at best imprecise. A file does not really contain characters. It contains byte sequences which may be interpreted as characters -- for example, by a console or other user presentation software which renders them into glyphs -- according to some encoding. Different encodings produce different results; in this particular case, you have a file which was created by software using the Windows-1252 encoding (or, roughly equivalently, ISO 8859-15), and you are rendering it on a console using UTF-8.

What that means is that the data read by getline contains an invalid UTF-8 sequence, but it (probably) does not contain the replacement character code. Based on the character string you present, it contains the hex character \xbb, which is a guillemot (») in Windows code page 1252.

Finding all the invalid UTF-8 sequences in a string read by getline (or any other C library function which reads files) requires scanning the string, but not for a particular code sequence. Rather, you need to decode UTF-8 sequences one at a time, looking for the ones which are not valid. That's not a simple task, but the mbtowc function can help (if you have enabled a UTF-8 locale). As you'll see in the linked manpage, mbtowc returns the number of bytes contained in a valid "multibyte sequence" (which is UTF-8 in a UTF-8 locale), or -1 to indicate an invalid or incomplete sequence. In the scan, you should pass through the bytes in a valid sequence, or remove/ignore the single byte starting an invalid sequence, and then continue the scan until you reach the end of the string.

Here's some lightly-tested example code (in C):

#include <stdlib.h>
#include <string.h>

/* Removes in place any invalid UTF-8 sequences from at most 'len' characters of the
 * string pointed to by 's'. (If a NUL byte is encountered, conversion stops.)
 * If the length of the converted string is less than 'len', a NUL byte is
 * inserted.
 * Returns the length of the possibly modified string (with a maximum of 'len'),
 * not including the NUL terminator (if any).
 * Requires that a UTF-8 locale be active; since there is no way to test for
 * this condition, no attempt is made to do so. If the current locale is not UTF-8,
 * behaviour is undefined.
 */
size_t remove_bad_utf8(char* s, size_t len) {
  char* in = s;
  /* Skip over the initial correct sequence. Avoid relying on mbtowc returning
   * zero if n is 0, since Posix is not clear whether mbtowc returns 0 or -1.
   */
  int seqlen;
  while (len && (seqlen = mbtowc(NULL, in, len)) > 0) { len -= seqlen; in += seqlen; }
  char* out = in;

  if (len && seqlen < 0) {
    ++in;
    --len;
    /* If we find an invalid sequence, we need to start shifting correct sequences.  */
    for (; len; in += seqlen, len -= seqlen) {
      seqlen = mbtowc(NULL, in, len);
      if (seqlen > 0) {
        /* Shift the valid sequence (if one was found) */
        memmove(out, in, seqlen);
        out += seqlen;
      }
      else if (seqlen < 0) seqlen = 1;
      else /* (seqlen == 0) */ break;
    }
    *out++ = 0;
  }
  return out - s;
}

---

Notes

1. Aside from the possible line-end transformation of the underlying I/O library, which will replace CR-LF with a single \n on systems like Windows where the two character CR-LF sequence is used as a line-end indication.

Answer 2

I also managed to fix it by trailing/cutting down all Non-ASCII characters.

This one takes about 2.6 seconds to parse 319MB:

#include <stdlib.h>
#include <iostream>

int main(int argc, char const *argv[])
{
    FILE* cfilestream = fopen( "./test.txt", "r" );
    size_t linebuffersize = 131072;

    if( cfilestream == NULL ) {
        perror( "fopen cfilestream" );
        return -1;
    }

    char* readline = (char*) malloc( linebuffersize );
    char* fixedreadline = (char*) malloc( linebuffersize );

    if( readline == NULL ) {
        perror( "malloc readline" );
        return -1;
    }

    if( fixedreadline == NULL ) {
        perror( "malloc fixedreadline" );
        return -1;
    }

    char* source;
    if( ( source = std::setlocale( LC_ALL, "en_US.utf8" ) ) == NULL ) {
        perror( "setlocale" );
    }
    else {
        std::cerr << "locale='" << source << "'" << std::endl;
    }

    int index;
    int charsread;
    int invalidcharsoffset;
    unsigned int fixedchar;

    while( true )
    {
        if( ( charsread = getline( &readline, &linebuffersize, cfilestream ) ) != -1 )
        {
            invalidcharsoffset = 0;
            for( index = 0; index < charsread; ++index )
            {
                fixedchar = static_cast<unsigned int>( readline[index] );
                // std::cerr << "index " << std::setw(3) << index
                //         << " readline " << std::setw(10) << fixedchar
                //         << " -> '" << readline[index] << "'" << std::endl;

                if( 31 < fixedchar && fixedchar < 128 ) {
                    fixedreadline[index-invalidcharsoffset] = readline[index];
                }
                else {
                    ++invalidcharsoffset;
                }
            }

            fixedreadline[index-invalidcharsoffset] = '\0';
            // std::cerr << "fixedreadline=" << fixedreadline << std::endl;
        }
        else {
            break;
        }
    }
    std::cerr << "fixedreadline=" << fixedreadline << std::endl;

    free( readline );
    free( fixedreadline );

    fclose( cfilestream );
    return 0;
}

---

Alternative and slower version using memcpy

Using menmove does not improve much speed, so you could either one.

This one takes about 3.1 seconds to parse 319MB:

#include <stdlib.h>
#include <iostream>
#include <cstring>
#include <iomanip>

int main(int argc, char const *argv[])
{
    FILE* cfilestream = fopen( "./test.txt", "r" );
    size_t linebuffersize = 131072;

    if( cfilestream == NULL ) {
        perror( "fopen cfilestream" );
        return -1;
    }

    char* readline = (char*) malloc( linebuffersize );
    char* fixedreadline = (char*) malloc( linebuffersize );

    if( readline == NULL ) {
        perror( "malloc readline" );
        return -1;
    }

    if( fixedreadline == NULL ) {
        perror( "malloc fixedreadline" );
        return -1;
    }

    char* source;
    char* destination;
    char* finalresult;

    int index;
    int lastcopy;
    int charsread;
    int charstocopy;
    int invalidcharsoffset;

    bool hasignoredbytes;
    unsigned int fixedchar;

    if( ( source = std::setlocale( LC_ALL, "en_US.utf8" ) ) == NULL ) {
        perror( "setlocale" );
    }
    else {
        std::cerr << "locale='" << source << "'" << std::endl;
    }

    while( true )
    {
        if( ( charsread = getline( &readline, &linebuffersize, cfilestream ) ) != -1 )
        {
            hasignoredbytes = false;
            source = readline;
            destination = fixedreadline;
            lastcopy = 0;
            invalidcharsoffset = 0;

            for( index = 0; index < charsread; ++index )
            {
                fixedchar = static_cast<unsigned int>( readline[index] );
                // std::cerr << "fixedchar " << std::setw(10)
                //           << fixedchar << " -> '"
                //           << readline[index] << "'" << std::endl;

                if( 31 < fixedchar && fixedchar < 128 ) {
                    if( hasignoredbytes ) {
                        charstocopy = index - lastcopy - invalidcharsoffset;
                        memcpy( destination, source, charstocopy );

                        source += index - lastcopy;
                        lastcopy = index;
                        destination += charstocopy;

                        invalidcharsoffset = 0;
                        hasignoredbytes = false;
                    }
                }
                else {
                    ++invalidcharsoffset;
                    hasignoredbytes = true;
                }
            }

            if( destination != fixedreadline ) {
                charstocopy = charsread - static_cast<int>( source - readline )
                               - invalidcharsoffset;

                memcpy( destination, source, charstocopy );
                destination += charstocopy - 1;

                if( *destination == '\n' ) {
                    *destination = '\0';
                }
                else {
                    *++destination = '\0';
                }
                finalresult = fixedreadline;
            }
            else {
                finalresult = readline;
            }

            // std::cerr << "finalresult=" << finalresult << std::endl;
        }
        else {
            break;
        }
    }
    std::cerr << "finalresult=" << finalresult << std::endl;

    free( readline );
    free( fixedreadline );

    fclose( cfilestream );
    return 0;
}

---

Optimized solution using iconv

This takes about 4.6 seconds to parse 319MB of text.

#include <iconv.h>
#include <string.h>
#include <stdlib.h>
#include <iostream>

// Compile it with:
//     g++ -o main test.cpp -O3 -liconv
int main(int argc, char const *argv[])
{
    FILE* cfilestream = fopen( "./test.txt", "r" );
    size_t linebuffersize = 131072;

    if( cfilestream == NULL ) {
        perror( "fopen cfilestream" );
        return -1;
    }

    char* readline = (char*) malloc( linebuffersize );
    char* fixedreadline = (char*) malloc( linebuffersize );

    if( readline == NULL ) {
        perror( "malloc readline" );
        return -1;
    }

    if( fixedreadline == NULL ) {
        perror( "malloc fixedreadline" );
        return -1;
    }

    char* source;
    char* destination;

    int charsread;
    size_t inchars;
    size_t outchars;

    if( ( source = std::setlocale( LC_ALL, "en_US.utf8" ) ) == NULL ) {
        perror( "setlocale" );
    }
    else {
        std::cerr << "locale='" << source << "'" << std::endl;
    }

    iconv_t conversiondescriptor = iconv_open("UTF-8//IGNORE", "UTF-8");
    if( conversiondescriptor == (iconv_t)-1 ) {
        perror( "iconv_open conversiondescriptor" );
    }

    while( true )
    {
        if( ( charsread = getline( &readline, &linebuffersize, cfilestream ) ) != -1 )
        {
            source = readline;
            inchars = charsread;

            destination = fixedreadline;
            outchars = charsread;

            if( iconv( conversiondescriptor, &source, &inchars, &destination, &outchars ) )
            {
                perror( "iconv" );
            }

            // Trim out the new line character
            if( *--destination == '\n' ) {
                *--destination = '\0';
            }
            else {
                *destination = '\0';
            }

            // std::cerr << "fixedreadline='" << fixedreadline << "'" << std::endl;
        }
        else {
            break;
        }
    }
    std::cerr << "fixedreadline='" << fixedreadline << "'" << std::endl;

    free( readline );
    free( fixedreadline );

    if( fclose( cfilestream ) ) {
        perror( "fclose cfilestream" );
    }

    if( iconv_close( conversiondescriptor ) ) {
        perror( "iconv_close conversiondescriptor" );
    }

    return 0;
}

---

Slowest solution ever using mbtowc

This takes about 24.2 seconds to parse 319MB of text.

If you comment out the line fixedchar = mbtowc(NULL, source, charsread); and uncomment the line charsread -= fixedchar; (breaking the invalid characters removal) this will take 1.9 seconds instead of 24.2 seconds (also compiled with -O3 optimization level).

#include <stdlib.h>
#include <string.h>

#include <iostream>
#include <cstring>
#include <iomanip>

int main(int argc, char const *argv[])
{
    FILE* cfilestream = fopen( "./test.txt", "r" );
    size_t linebuffersize = 131072;

    if( cfilestream == NULL ) {
        perror( "fopen cfilestream" );
        return -1;
    }

    char* readline = (char*) malloc( linebuffersize );
    if( readline == NULL ) {
        perror( "malloc readline" );
        return -1;
    }

    char* source;
    char* lineend;
    char* destination;
    int charsread;
    int fixedchar;

    if( ( source = std::setlocale( LC_ALL, "en_US.utf8" ) ) == NULL ) {
        perror( "setlocale" );
    }
    else {
        std::cerr << "locale='" << source << "'" << std::endl;
    }

    while( true )
    {
        if( ( charsread = getline( &readline, &linebuffersize, cfilestream ) ) != -1 )
        {
            lineend = readline + charsread;
            destination = readline;
            for( source = readline; source != lineend; )
            {
                // fixedchar = 1;
                fixedchar = mbtowc(NULL, source, charsread);
                charsread -= fixedchar;

                // std::ostringstream contents;
                // for( int index = 0; index < fixedchar; ++index )
                //         contents << source[index];

                // std::cerr << "fixedchar=" << std::setw(10)
                //         << fixedchar << " -> '"
                //         << contents.str().c_str() << "'" << std::endl;

                if( fixedchar > 0 ) {
                    memmove( destination, source, fixedchar );
                    source += fixedchar;
                    destination += fixedchar;
                }
                else if( fixedchar < 0 ) {
                    source += 1;
                    // std::cerr << "errno=" << strerror( errno ) << std::endl;
                }
                else {
                    break;
                }
            }

            // Trim out the new line character
            if( *--destination == '\n' ) {
                *--destination = '\0';
            }
            else {
                *destination = '\0';
            }

            // std::cerr << "readline='" << readline << "'" << std::endl;
        }
        else {
            break;
        }
    }
    std::cerr << "readline='" << readline << "'" << std::endl;

    if( fclose( cfilestream ) ) {
        perror( "fclose cfilestream" );
    }

    free( readline );
    return 0;
}

---

Fastest version from all my others above using memmove

You cannot use memcpy here because the memory regions overlap!

This takes about 2.4 seconds to parse 319MB.

If you comment out the lines *destination = *source and memmove( destination, source, 1 ) (breaking the invalid characters removal) the performance still almost the same as when memmove is being called. Here in, calling memmove( destination, source, 1 ) is a little slower than directly doing *destination = *source;

#include <stdlib.h>
#include <iostream>
#include <cstring>
#include <iomanip>

int main(int argc, char const *argv[])
{
    FILE* cfilestream = fopen( "./test.txt", "r" );
    size_t linebuffersize = 131072;

    if( cfilestream == NULL ) {
        perror( "fopen cfilestream" );
        return -1;
    }

    char* readline = (char*) malloc( linebuffersize );
    if( readline == NULL ) {
        perror( "malloc readline" );
        return -1;
    }

    char* source;
    char* lineend;
    char* destination;

    int charsread;
    unsigned int fixedchar;

    if( ( source = std::setlocale( LC_ALL, "en_US.utf8" ) ) == NULL ) {
        perror( "setlocale" );
    }
    else {
        std::cerr << "locale='" << source << "'" << std::endl;
    }

    while( true )
    {
        if( ( charsread = getline( &readline, &linebuffersize, cfilestream ) ) != -1 )
        {
            lineend = readline + charsread;
            destination = readline;
            for( source = readline; source != lineend; ++source )
            {
                fixedchar = static_cast<unsigned int>( *source );
                // std::cerr << "fixedchar=" << std::setw(10)
                //         << fixedchar << " -> '" << *source << "'" << std::endl;

                if( 31 < fixedchar && fixedchar < 128 ) {
                    *destination = *source;
                    ++destination;
                }
            }

            // Trim out the new line character
            if( *source == '\n' ) {
                *--destination = '\0';
            }
            else {
                *destination = '\0';
            }

            // std::cerr << "readline='" << readline << "'" << std::endl;
        }
        else {
            break;
        }
    }
    std::cerr << "readline='" << readline << "'" << std::endl;

    if( fclose( cfilestream ) ) {
        perror( "fclose cfilestream" );
    }

    free( readline );
    return 0;
}

---

Bonus

You can also use Python C Extensions (API).

It takes about 2.3 seconds to parse 319MB without converting them to cached version UTF-8 char*

And takes about 3.2 seconds to parse 319MB converting them to UTF-8 char*. And also takes about 3.2 seconds to parse 319MB converting them to cached ASCII char*.

#define PY_SSIZE_T_CLEAN
#include <Python.h>
#include <iostream>

typedef struct
{
    PyObject_HEAD
}
PyFastFile;

static PyModuleDef fastfilepackagemodule =
{
    // https://docs.python.org/3/c-api/module.html#c.PyModuleDef
    PyModuleDef_HEAD_INIT,
    "fastfilepackage", /* name of module */
    "Example module that wrapped a C++ object", /* module documentation, may be NULL */
    -1, /* size of per-interpreter state of the module, or 
                -1 if the module keeps state in global variables. */

    NULL, /* PyMethodDef* m_methods */
    NULL, /* inquiry m_reload */
    NULL, /* traverseproc m_traverse */
    NULL, /* inquiry m_clear */
    NULL, /* freefunc m_free */
};

// initialize PyFastFile Object
static int PyFastFile_init(PyFastFile* self, PyObject* args, PyObject* kwargs) {
    char* filepath;

    if( !PyArg_ParseTuple( args, "s", &filepath ) ) {
        return -1;
    }

    int linecount = 0;
    PyObject* iomodule;
    PyObject* openfile;
    PyObject* fileiterator;

    iomodule = PyImport_ImportModule( "builtins" );
    if( iomodule == NULL ) {
        std::cerr << "ERROR: FastFile failed to import the io module '"
                "(and open the file " << filepath << "')!" << std::endl;
        PyErr_PrintEx(100);
        return -1;
    }
    PyObject* openfunction = PyObject_GetAttrString( iomodule, "open" );

    if( openfunction == NULL ) {
        std::cerr << "ERROR: FastFile failed get the io module open "
                << "function (and open the file '" << filepath << "')!" << std::endl;
        PyErr_PrintEx(100);
        return -1;
    }
    openfile = PyObject_CallFunction( 
            openfunction, "ssiss", filepath, "r", -1, "ASCII", "ignore" );

    if( openfile == NULL ) {
        std::cerr << "ERROR: FastFile failed to open the file'"
                << filepath << "'!" << std::endl;
        PyErr_PrintEx(100);
        return -1;
    }
    PyObject* iterfunction = PyObject_GetAttrString( openfile, "__iter__" );
    Py_DECREF( openfunction );

    if( iterfunction == NULL ) {
        std::cerr << "ERROR: FastFile failed get the io module iterator" 
                << "function (and open the file '" << filepath << "')!" << std::endl;
        PyErr_PrintEx(100);
        return -1;
    }
    PyObject* openiteratorobject = PyObject_CallObject( iterfunction, NULL );
    Py_DECREF( iterfunction );

    if( openiteratorobject == NULL ) {
        std::cerr << "ERROR: FastFile failed get the io module iterator object"
                << " (and open the file '" << filepath << "')!" << std::endl;
        PyErr_PrintEx(100);
        return -1;
    }
    fileiterator = PyObject_GetAttrString( openfile, "__next__" );
    Py_DECREF( openiteratorobject );

    if( fileiterator == NULL ) {
        std::cerr << "ERROR: FastFile failed get the io module iterator "
                << "object (and open the file '" << filepath << "')!" << std::endl;
        PyErr_PrintEx(100);
        return -1;
    }

    PyObject* readline;
    while( ( readline = PyObject_CallObject( fileiterator, NULL ) ) != NULL ) {
        linecount += 1;
        PyUnicode_AsUTF8( readline );
        Py_DECREF( readline );
        // std::cerr << "linecount " << linecount << " readline '" << readline
        //         << "' '" << PyUnicode_AsUTF8( readline ) << "'" << std::endl;
    }
    std::cerr << "linecount " << linecount << std::endl;

    // PyErr_PrintEx(100);
    PyErr_Clear();
    PyObject* closefunction = PyObject_GetAttrString( openfile, "close" );

    if( closefunction == NULL ) {
        std::cerr << "ERROR: FastFile failed get the close file function for '"
                << filepath << "')!" << std::endl;
        PyErr_PrintEx(100);
        return -1;
    }

    PyObject* closefileresult = PyObject_CallObject( closefunction, NULL );
    Py_DECREF( closefunction );

    if( closefileresult == NULL ) {
        std::cerr << "ERROR: FastFile failed close open file '"
                << filepath << "')!" << std::endl;
        PyErr_PrintEx(100);
        return -1;
    }
    Py_DECREF( closefileresult );

    Py_XDECREF( iomodule );
    Py_XDECREF( openfile );
    Py_XDECREF( fileiterator );

    return 0;
}

// destruct the object
static void PyFastFile_dealloc(PyFastFile* self) {
    Py_TYPE(self)->tp_free( (PyObject*) self );
}

static PyTypeObject PyFastFileType =
{
    PyVarObject_HEAD_INIT( NULL, 0 )
    "fastfilepackage.FastFile" /* tp_name */
};

// create the module
PyMODINIT_FUNC PyInit_fastfilepackage(void)
{
    PyObject* thismodule;

    // https://docs.python.org/3/c-api/typeobj.html
    PyFastFileType.tp_new = PyType_GenericNew;
    PyFastFileType.tp_basicsize = sizeof(PyFastFile);
    PyFastFileType.tp_dealloc = (destructor) PyFastFile_dealloc;
    PyFastFileType.tp_flags = Py_TPFLAGS_DEFAULT;
    PyFastFileType.tp_doc = "FastFile objects";
    PyFastFileType.tp_init = (initproc) PyFastFile_init;

    if( PyType_Ready( &PyFastFileType) < 0 ) {
        return NULL;
    }

    thismodule = PyModule_Create(&fastfilepackagemodule);
    if( thismodule == NULL ) {
        return NULL;
    }

    // Add FastFile class to thismodule allowing the use to create objects
    Py_INCREF( &PyFastFileType );
    PyModule_AddObject( thismodule, "FastFile", (PyObject*) &PyFastFileType );
    return thismodule;
}

To built it, create the file source/fastfilewrappar.cpp with the contents of the above file and the setup.py with the following contents:

#! /usr/bin/env python
# -*- coding: utf-8 -*-
from setuptools import setup, Extension

myextension = Extension(
    language = "c++",
    extra_link_args = ["-std=c++11"],
    extra_compile_args = ["-std=c++11"],
    name = 'fastfilepackage',
    sources = [
        'source/fastfilewrapper.cpp'
    ],
    include_dirs = [ 'source' ],
)

setup(
        name = 'fastfilepackage',
        ext_modules= [ myextension ],
    )

To run example, use following Python script:

import time
import datetime
import fastfilepackage

testfile = './test.txt'
timenow = time.time()
iterable = fastfilepackage.FastFile( testfile )

fastfile_time = time.time() - timenow
timedifference = datetime.timedelta( seconds=fastfile_time )
print( 'FastFile timedifference', timedifference, flush=True )

Example:

user@user-pc$ /usr/bin/pip3.6 install .
Processing /fastfilepackage
Building wheels for collected packages: fastfilepackage
  Building wheel for fastfilepackage (setup.py) ... done
  Stored in directory: /pip-ephem-wheel-cache-j313cpzc/wheels/e5/5f/bc/52c820
Successfully built fastfilepackage
Installing collected packages: fastfilepackage
  Found existing installation: fastfilepackage 0.0.0
    Uninstalling fastfilepackage-0.0.0:
      Successfully uninstalled fastfilepackage-0.0.0
Successfully installed fastfilepackage-0.0.0

user@user-pc$ /usr/bin/python3.6 fastfileperformance.py
linecount 820800
FastFile timedifference 0:00:03.204614

---

Using std::getline

This takes about 4.7 seconds to parse 319MB.

If you remove the UTF-8 removal algorithm borrowed from the fastest benchmark using stdlib.h getline(), it takes 1.7 seconds to run.

#include <stdlib.h>
#include <iostream>
#include <locale>
#include <fstream>
#include <iomanip>

int main(int argc, char const *argv[])
{
    unsigned int fixedchar;
    int linecount = -1;

    char* source;
    char* lineend;
    char* destination;

    if( ( source = setlocale( LC_ALL, "en_US.ascii" ) ) == NULL ) {
        perror( "setlocale" );
        return -1;
    }
    else {
        std::cerr << "locale='" << source << "'" << std::endl;
    }

    std::ifstream fileifstream{ "./test.txt" };
    if( fileifstream.fail() ) {
        std::cerr << "ERROR: FastFile failed to open the file!" << std::endl;
        return -1;
    }

    size_t linebuffersize = 131072;
    char* readline = (char*) malloc( linebuffersize );

    if( readline == NULL ) {
        perror( "malloc readline" );
        return -1;
    }

    while( true )
    {
        if( !fileifstream.eof() )
        {
            linecount += 1;
            fileifstream.getline( readline, linebuffersize );
            lineend = readline + fileifstream.gcount();
            destination = readline;

            for( source = readline; source != lineend; ++source )
            {
                fixedchar = static_cast<unsigned int>( *source );
                // std::cerr << "fixedchar=" << std::setw(10)
                //         << fixedchar << " -> '" << *source << "'" << std::endl;

                if( 31 < fixedchar && fixedchar < 128 ) {
                    *destination = *source;
                    ++destination;
                }
            }

            // Trim out the new line character
            if( *source == '\n' ) {
                *--destination = '\0';
            }
            else {
                *destination = '\0';
            }

            // std::cerr << "readline='" << readline << "'" << std::endl;
        }
        else {
            break;
        }
    }
    std::cerr << "linecount='" << linecount << "'" << std::endl;

    if( fileifstream.is_open() ) {
        fileifstream.close();
    }

    free( readline );
    return 0;
}

---

Resume

1. 2.6 seconds trimming UTF-8 using two buffers with indexing
2. 3.1 seconds trimming UTF-8 using two buffers with memcpy
3. 4.6 seconds removing invalid UTF-8 with iconv
4. 24.2 seconds removing invalid UTF-8 with mbtowc
5. 2.4 seconds trimming UTF-8 using one buffer with pointer direct assigning

Bonus

1. 2.3 seconds removing invalid UTF-8 without converting them to a cached UTF-8 char*
2. 3.2 seconds removing invalid UTF-8 converting them to a cached UTF-8 char*
3. 3.2 seconds trimming UTF-8 and caching as ASCII char*
4. 4.7 seconds trimming UTF-8 with std::getline() using one buffer with pointer direct assigning

The used file ./text.txt had 820.800 lines where each line was equal to:

id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char&id-é-char\r\n

And all versions where compiled with

1. g++ (GCC) 7.4.0
2. iconv (GNU libiconv 1.14)
3. g++ -o main test.cpp -O3 -liconv && time ./main

Answer 3

As @rici well explains in his answer, there can be several invalid UTF-8 sequences in a byte sequence.

Possibly iconv(3) could be worth a look, e.g. see https://linux.die.net/man/3/iconv_open.

When the string "//IGNORE" is appended to tocode, characters that cannot be represented in the target character set will be silently discarded.

Example

This byte sequence, if interpreted as UTF-8, contains some invalid UTF-8:

"some invalid\xFE\xFE\xFF\xFF stuff"

If you display this you would see something like

some invalid���� stuff

When this string passes through the remove_invalid_utf8 function in the following C program, the invalid UTF-8 bytes are removed using the iconv function mentioned above.

So the result is then:

some invalid stuff

C Program

#include <stdio.h>
#include <iconv.h>
#include <string.h>
#include <stdlib.h>
#include <stdbool.h>
#include <errno.h>

char *remove_invalid_utf8(char *utf8, size_t len) {
    size_t inbytes_len = len;
    char *inbuf = utf8;

    size_t outbytes_len = len;
    char *result = calloc(outbytes_len + 1, sizeof(char));
    char *outbuf = result;

    iconv_t cd = iconv_open("UTF-8//IGNORE", "UTF-8");
    if(cd == (iconv_t)-1) {
        perror("iconv_open");
    }
    if(iconv(cd, &inbuf, &inbytes_len, &outbuf, &outbytes_len)) {
        perror("iconv");
    }
    iconv_close(cd);
    return result;
}

int main() {
    char *utf8 = "some invalid\xFE\xFE\xFF\xFF stuff";
    char *converted = remove_invalid_utf8(utf8, strlen(utf8));
    printf("converted: %s to %s\n", utf8, converted);
    free(converted);
    return 0;
}