scuffed-code/icu4c/source/test/iotest/filetst.c
Michael Ow dc88968faf ICU-5621 Apply patch for u_finit.
X-SVN-Rev: 27163
2010-01-07 20:36:40 +00:00

1538 lines
56 KiB
C

/*
**********************************************************************
* Copyright (C) 2004-2010, International Business Machines
* Corporation and others. All Rights Reserved.
**********************************************************************
* file name: filetst.c
* encoding: US-ASCII
* tab size: 8 (not used)
* indentation:4
*
* created on: 2004apr06
* created by: George Rhoten
*/
#include "iotest.h"
#include "unicode/ustdio.h"
#include "unicode/ustring.h"
#include "unicode/uloc.h"
#include <string.h>
#include <stdlib.h>
const char *STANDARD_TEST_FILE = "iotest-c.txt";
#if !UCONFIG_NO_FORMATTING
static void TestFileFromICU(UFILE *myFile) {
int32_t n[1];
float myFloat = -1234.0;
int32_t newValuePtr[1];
double newDoubleValuePtr[1];
UChar myUString[256];
UChar uStringBuf[256];
char myString[256] = "";
char testBuf[256] = "";
void *origPtr, *ptr;
U_STRING_DECL(myStringOrig, "My-String", 9);
U_STRING_INIT(myStringOrig, "My-String", 9);
u_memset(myUString, 0x2a, sizeof(myUString)/sizeof(*myUString));
u_memset(uStringBuf, 0x2a, sizeof(uStringBuf)/sizeof(*uStringBuf));
memset(myString, '*', sizeof(myString)/sizeof(*myString));
memset(testBuf, '*', sizeof(testBuf)/sizeof(*testBuf));
if (myFile == NULL) {
log_err("Can't write test file.\n");
return;
}
*n = -1234;
if (sizeof(void *) == 4) {
origPtr = (void *)0xdeadbeef;
} else if (sizeof(void *) == 8) {
origPtr = (void *) INT64_C(0x1000200030004000);
} else if (sizeof(void *) == 16) {
/* iSeries */
union {
int32_t arr[4];
void *ptr;
} massiveBigEndianPtr = {{ 0x10002000, 0x30004000, 0x50006000, 0x70008000 }};
origPtr = massiveBigEndianPtr.ptr;
} else {
log_err("sizeof(void*)=%d hasn't been tested before", (int)sizeof(void*));
}
/* Test fprintf */
u_fprintf(myFile, "Signed decimal integer %%d: %d\n", *n);
u_fprintf(myFile, "Signed decimal integer %%i: %i\n", *n);
u_fprintf(myFile, "Unsigned octal integer %%o: %o\n", *n);
u_fprintf(myFile, "Unsigned decimal integer %%u: %u\n", *n);
u_fprintf(myFile, "Lowercase unsigned hexadecimal integer %%x: %x\n", *n);
u_fprintf(myFile, "Uppercase unsigned hexadecimal integer %%X: %X\n", *n);
u_fprintf(myFile, "Float %%f: %f\n", myFloat);
u_fprintf(myFile, "Lowercase float %%e: %e\n", myFloat);
u_fprintf(myFile, "Uppercase float %%E: %E\n", myFloat);
u_fprintf(myFile, "Lowercase float %%g: %g\n", myFloat);
u_fprintf(myFile, "Uppercase float %%G: %G\n", myFloat);
u_fprintf(myFile, "Pointer %%p: %p\n", origPtr);
u_fprintf(myFile, "Char %%c: %c\n", 'A');
u_fprintf(myFile, "UChar %%C: %C\n", (UChar)0x0041); /*'A'*/
u_fprintf(myFile, "String %%s: %s\n", "My-String");
u_fprintf(myFile, "NULL String %%s: %s\n", NULL);
u_fprintf(myFile, "Unicode String %%S: %S\n", myStringOrig);
u_fprintf(myFile, "NULL Unicode String %%S: %S\n", NULL);
u_fprintf(myFile, "Percent %%P (non-ANSI): %P\n", myFloat);
u_fprintf(myFile, "Spell Out %%V (non-ANSI): %V\n", myFloat);
if (u_feof(myFile)) {
log_err("Got feof while writing the file.\n");
}
*n = 1;
u_fprintf(myFile, "\t\nPointer to integer (Count) %%n: n=%d %n n=%d\n", *n, n, *n);
u_fprintf(myFile, "Pointer to integer Value: %d\n", *n);
u_fprintf(myFile, "This is a long test123456789012345678901234567890123456789012345678901234567890\n");
*n = 1;
u_fprintf(myFile, "\tNormal fprintf count: n=%d %n n=%d\n", (int)*n, (int*)n, (int)*n);
fprintf(u_fgetfile(myFile), "\tNormal fprintf count value: n=%d\n", (int)*n); /* Should be 27 as stated later on. */
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "r", "en_US_POSIX", NULL);
if (myFile == NULL) {
log_err("Can't read test file.");
return;
}
if (u_feof(myFile)) {
log_err("Got feof while reading the file and not at the end of the file.\n");
}
myUString[0] = u_fgetc(myFile);
if (myUString[0] != 0x53 /* S */) {
log_err("u_fgetc 1 returned %X. Expected 'S'.", myString[0]);
}
u_fungetc(myUString[0], myFile);
myUString[0] = u_fgetc(myFile);
if (myUString[0] != 0x53 /* S */) {
log_err("u_fgetc 2 returned %X. Expected 'S'.", myString[0]);
}
u_fungetc(myUString[0], myFile);
myUString[0] = u_fgetc(myFile);
if (myUString[0] != 0x53 /* S */) {
log_err("u_fgetc 3 returned %X. Expected 'S'.", myString[0]);
}
u_fungetc(myUString[0], myFile);
myUString[0] = u_fgetc(myFile);
myUString[1] = (UChar)u_fgetcx(myFile); /* Mix getc and getcx and see what happens. */
myUString[2] = u_fgetc(myFile);
if (myUString[0] != 0x53 /* S */ && myUString[1] != 0x69 /* i */ && myUString[2] != 0x6E /* n */) {
log_err("u_fgetcx returned \\u%04X\\u%04X\\u%04X. Expected 'Sin'.", myString[0], myString[1], myString[2]);
}
u_fungetc(myUString[2], myFile);
u_fungetc(myUString[1], myFile);
u_fungetc(myUString[0], myFile);
*n = -1234;
*newValuePtr = 1;
u_fscanf(myFile, "Signed decimal integer %%d: %d\n", newValuePtr);
if (*n != *newValuePtr) {
log_err("%%d Got: %d, Expected: %d\n", *newValuePtr, *n);
}
*newValuePtr = 1;
u_fscanf(myFile, "Signed decimal integer %%i: %i\n", newValuePtr);
if (*n != *newValuePtr) {
log_err("%%i Got: %i, Expected: %i\n", *newValuePtr, *n);
}
*newValuePtr = 1;
u_fscanf(myFile, "Unsigned octal integer %%o: %o\n", newValuePtr);
if (*n != *newValuePtr) {
log_err("%%o Got: %o, Expected: %o\n", *newValuePtr, *n);
}
*newValuePtr = 1;
u_fscanf(myFile, "Unsigned decimal integer %%u: %u\n", newValuePtr);
if (*n != *newValuePtr) {
log_err("%%u Got: %u, Expected: %u\n", *newValuePtr, *n);
}
*newValuePtr = 1;
u_fscanf(myFile, "Lowercase unsigned hexadecimal integer %%x: %x\n", newValuePtr);
if (*n != *newValuePtr) {
log_err("%%x Got: %x, Expected: %x\n", *newValuePtr, *n);
}
*newValuePtr = 1;
u_fscanf(myFile, "Uppercase unsigned hexadecimal integer %%X: %X\n", newValuePtr);
if (*n != *newValuePtr) {
log_err("%%X Got: %X, Expected: %X\n", *newValuePtr, *n);
}
*newDoubleValuePtr = -1.0;
u_fscanf(myFile, "Float %%f: %lf\n", newDoubleValuePtr);
if (myFloat != *newDoubleValuePtr) {
log_err("%%f Got: %f, Expected: %f\n", *newDoubleValuePtr, myFloat);
}
*newDoubleValuePtr = -1.0;
u_fscanf(myFile, "Lowercase float %%e: %le\n", newDoubleValuePtr);
if (myFloat != *newDoubleValuePtr) {
log_err("%%e Got: %e, Expected: %e\n", *newDoubleValuePtr, myFloat);
}
*newDoubleValuePtr = -1.0;
u_fscanf(myFile, "Uppercase float %%E: %lE\n", newDoubleValuePtr);
if (myFloat != *newDoubleValuePtr) {
log_err("%%E Got: %E, Expected: %E\n", *newDoubleValuePtr, myFloat);
}
*newDoubleValuePtr = -1.0;
u_fscanf(myFile, "Lowercase float %%g: %lg\n", newDoubleValuePtr);
if (myFloat != *newDoubleValuePtr) {
log_err("%%g Got: %g, Expected: %g\n", *newDoubleValuePtr, myFloat);
}
*newDoubleValuePtr = -1.0;
u_fscanf(myFile, "Uppercase float %%G: %lG\n", newDoubleValuePtr);
if (myFloat != *newDoubleValuePtr) {
log_err("%%G Got: %G, Expected: %G\n", *newDoubleValuePtr, myFloat);
}
ptr = NULL;
u_fscanf(myFile, "Pointer %%p: %p\n", &ptr);
if (ptr != origPtr) {
log_err("%%p Got: %p, Expected: %p\n", ptr, origPtr);
}
u_fscanf(myFile, "Char %%c: %c\n", myString);
if (*myString != 'A') {
log_err("%%c Got: %c, Expected: A\n", *myString);
}
u_fscanf(myFile, "UChar %%C: %C\n", myUString);
if (*myUString != (UChar)0x0041) { /*'A'*/
log_err("%%C Got: %C, Expected: A\n", *myUString);
}
u_fscanf(myFile, "String %%s: %s\n", myString);
if (strcmp(myString, "My-String")) {
log_err("%%s Got: %s, Expected: My String\n", myString);
}
u_fscanf(myFile, "NULL String %%s: %s\n", myString);
if (strcmp(myString, "(null)")) {
log_err("%%s Got: %s, Expected: My String\n", myString);
}
u_fscanf(myFile, "Unicode String %%S: %S\n", myUString);
u_austrncpy(myString, myUString, sizeof(myUString)/sizeof(*myUString));
if (strcmp(myString, "My-String")) {
log_err("%%S Got: %S, Expected: My String\n", myUString);
}
u_fscanf(myFile, "NULL Unicode String %%S: %S\n", myUString);
u_austrncpy(myString, myUString, sizeof(myUString)/sizeof(*myUString));
if (strcmp(myString, "(null)")) {
log_err("%%S Got: %S, Expected: My String\n", myUString);
}
*newDoubleValuePtr = -1.0;
u_fscanf(myFile, "Percent %%P (non-ANSI): %P\n", newDoubleValuePtr);
if (myFloat != *newDoubleValuePtr) {
log_err("%%P Got: %f, Expected: %f\n", *newDoubleValuePtr, myFloat);
}
*newDoubleValuePtr = -1.0;
u_fscanf(myFile, "Spell Out %%V (non-ANSI): %V\n", newDoubleValuePtr);
if (myFloat != *newDoubleValuePtr) {
log_err("%%V Got: %f, Expected: %f\n", *newDoubleValuePtr, myFloat);
}
u_fgets(myUString, 4, myFile);
myString[2] = '!';
myString[3] = '!';
u_austrncpy(myString, myUString, sizeof(myUString)/sizeof(*myUString));
if (myString == NULL || strcmp(myString, "\t\n") != 0) {
log_err("u_fgets got \"%s\"\n", myString);
}
if (u_fgets(myUString, sizeof(myUString)/sizeof(*myUString), myFile) != myUString) {
log_err("u_fgets did not return myUString\n");
}
u_austrncpy(myString, myUString, sizeof(myUString)/sizeof(*myUString));
if (myString == NULL || strcmp(myString, "Pointer to integer (Count) %n: n=1 n=1\n") != 0) {
log_err("u_fgets got \"%s\"\n", myString);
}
if (u_fgets(myUString, sizeof(myUString)/sizeof(*myUString), myFile) != myUString) {
log_err("u_fgets did not return myUString\n");
}
u_austrncpy(myString, myUString, sizeof(myUString)/sizeof(*myUString));
if (myString == NULL || strcmp(myString, "Pointer to integer Value: 37\n") != 0) {
log_err("u_fgets got \"%s\"\n", myString);
}
if (u_fgets(myUString, sizeof(myUString)/sizeof(*myUString), myFile) != myUString) {
log_err("u_fgets did not return myUString\n");
}
u_austrncpy(myString, myUString, sizeof(myUString)/sizeof(*myUString));
if (myString == NULL || strcmp(myString, "This is a long test123456789012345678901234567890123456789012345678901234567890\n") != 0) {
log_err("u_fgets got \"%s\"\n", myString);
}
if (u_fgets(myUString, 0, myFile) != NULL) {
log_err("u_fgets got \"%s\" and it should have returned NULL\n", myString);
}
if (u_fgets(myUString, 1, myFile) != myUString) {
log_err("u_fgets did not return myUString\n");
}
u_austrncpy(myString, myUString, sizeof(myUString)/sizeof(*myUString));
if (myString == NULL || strcmp(myString, "") != 0) {
log_err("u_fgets got \"%s\"\n", myString);
}
if (u_fgets(myUString, 2, myFile) != myUString) {
log_err("u_fgets did not return myUString\n");
}
u_austrncpy(myString, myUString, sizeof(myUString)/sizeof(*myUString));
if (myString == NULL || strcmp(myString, "\t") != 0) {
log_err("u_fgets got \"%s\"\n", myString);
}
u_austrncpy(myString, u_fgets(myUString, sizeof(myUString)/sizeof(*myUString), myFile),
sizeof(myUString)/sizeof(*myUString));
if (strcmp(myString, "Normal fprintf count: n=1 n=1\n") != 0) {
log_err("u_fgets got \"%s\"\n", myString);
}
if (u_feof(myFile)) {
log_err("Got feof while reading the file and not at the end of the file.\n");
}
u_austrncpy(myString, u_fgets(myUString, sizeof(myUString)/sizeof(*myUString), myFile),
sizeof(myUString)/sizeof(*myUString));
if (strcmp(myString, "\tNormal fprintf count value: n=27\n") != 0) {
log_err("u_fgets got \"%s\"\n", myString);
}
if (!u_feof(myFile)) {
log_err("Did not get feof while reading the end of the file.\n");
}
if (u_fscanf(myFile, "%S\n", myUString) != 0) {
log_err("u_fscanf read data while reading the end of the file.\n");
}
u_fclose(myFile);
}
static void TestFile(void) {
/* FILE *standardFile;*/
log_verbose("Testing u_fopen\n");
TestFileFromICU(u_fopen(STANDARD_TEST_FILE, "w", "en_US_POSIX", NULL));
/* Don't know how to make this work without stdout or stderr */
/*
log_verbose("Testing u_finit\n");
standardFile = fopen(STANDARD_TEST_FILE, "wb");
TestFileFromICU(u_finit(standardFile, NULL, NULL));
fclose(standardFile);
log_verbose("Testing u_fadopt\n");
standardFile = fopen(STANDARD_TEST_FILE, "wb");
TestFileFromICU(u_fadopt(standardFile, NULL, NULL));
*/
}
#endif
static void StdinBuffering(void) {
#if 0
UChar buff[255];
int32_t num = 0;
UFILE *uStdIn = NULL;
UFILE *uStdOut = NULL;
uStdIn = u_finit(stdin, NULL, NULL);
uStdOut = u_finit(stdout, NULL, NULL);
if (uStdIn == NULL)
return;
buff[0] = 0x40;
buff[1] = 0;
u_fgets(buff, sizeof(buff)/sizeof(buff[0]), uStdIn);
u_fprintf(uStdOut, "%S\n", buff);
u_fscanf(uStdIn, "%d", &num);
u_fprintf(uStdOut, "%d\n", num);
u_fscanf(uStdIn, "%d", &num);
u_fprintf(uStdOut, "%d\n", num);
#else
log_verbose("Test disabled because it requires user interaction");
#endif
}
static void TestCodepageAndLocale(void) {
UFILE *myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, NULL);
if (myFile == NULL) {
log_err("Can't write test file.\n");
return;
}
if (u_fgetcodepage(myFile) == NULL
|| strcmp(u_fgetcodepage(myFile), ucnv_getDefaultName()) != 0)
{
log_err("Didn't get the proper default codepage. Got %s expected: %s\n",
u_fgetcodepage(myFile), ucnv_getDefaultName());
}
#if !UCONFIG_NO_FORMATTING
if (u_fgetlocale(myFile) == NULL
|| strcmp(u_fgetlocale(myFile), uloc_getDefault()) != 0)
{
log_err("Didn't get the proper default locale. Got %s expected: %s\n",
u_fgetlocale(myFile), uloc_getDefault());
}
#endif
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "w", "es", NULL);
if (u_fgetcodepage(myFile) == NULL
|| strcmp(u_fgetcodepage(myFile), ucnv_getDefaultName()) != 0)
{
log_err("Didn't get the proper default codepage for \"es\". Got %s expected: iso-8859-1\n",
u_fgetcodepage(myFile));
}
#if !UCONFIG_NO_FORMATTING
if (u_fgetlocale(myFile) == NULL
|| strcmp(u_fgetlocale(myFile), "es") != 0)
{
log_err("Didn't get the proper default locale. Got %s expected: %s\n",
u_fgetlocale(myFile), "es");
}
#endif
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, "UTF-16");
if (u_fgetcodepage(myFile) == NULL
|| strcmp(u_fgetcodepage(myFile), "UTF-16") != 0)
{
log_err("Didn't get the proper default codepage for \"en\". Got %s expected: iso-8859-1\n",
u_fgetcodepage(myFile));
}
#if !UCONFIG_NO_FORMATTING
if (u_fgetlocale(myFile) == NULL
|| strcmp(u_fgetlocale(myFile), uloc_getDefault()) != 0)
{
log_err("Didn't get the proper default locale. Got %s expected: %s\n",
u_fgetlocale(myFile), uloc_getDefault());
}
#endif
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "w", "zh", "UTF-16");
if (u_fgetcodepage(myFile) == NULL
|| strcmp(u_fgetcodepage(myFile), "UTF-16") != 0)
{
log_err("Didn't get the proper default codepage for \"en\". Got %s expected: iso-8859-1\n",
u_fgetcodepage(myFile));
}
#if !UCONFIG_NO_FORMATTING
if (u_fgetlocale(myFile) == NULL
|| strcmp(u_fgetlocale(myFile), "zh") != 0)
{
log_err("Didn't get the proper default locale. Got %s expected: %s\n",
u_fgetlocale(myFile), "zh");
}
#endif
u_fclose(myFile);
}
static void TestfgetsBuffers(void) {
UChar buffer[2048];
UChar expectedBuffer[2048];
static const char testStr[] = "This is a test string that tests u_fgets. It makes sure that we don't try to read too much!";
UFILE *myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, "UTF-16");
int32_t expectedSize = (int32_t)strlen(testStr);
int32_t readSize;
int32_t repetitions;
if (myFile == NULL) {
log_err("Can't write test file.\n");
return;
}
u_fputc(0x3BC, myFile);
if (u_fputc(0x110000, myFile) != U_EOF) {
log_err("u_fputc should return U_EOF for 0x110000.\n");
}
if (u_fputc((UChar32)0xFFFFFFFFu, myFile) != U_EOF) {
log_err("u_fputc should return U_EOF for 0xFFFFFFFF.\n");
}
u_fputc(0xFF41, myFile);
u_memset(buffer, 0xBEEF, sizeof(buffer)/sizeof(buffer[0]));
u_memset(expectedBuffer, 0, sizeof(expectedBuffer)/sizeof(expectedBuffer[0]));
u_uastrncpy(buffer, testStr, expectedSize+1);
for (repetitions = 0; repetitions < 16; repetitions++) {
u_file_write(buffer, expectedSize, myFile);
u_strcat(expectedBuffer, buffer);
}
u_fclose(myFile);
u_memset(buffer, 0xBEEF, sizeof(buffer)/sizeof(buffer[0]));
myFile = u_fopen(STANDARD_TEST_FILE, "r", NULL, "UTF-16");
if (u_fgetc(myFile) != 0x3BC) {
log_err("The first character is wrong\n");
}
if (u_fgetc(myFile) != 0xFF41) {
log_err("The second character is wrong\n");
}
if (u_fgets(buffer, sizeof(buffer)/sizeof(buffer[0]), myFile) != buffer) {
log_err("Didn't get the buffer back\n");
return;
}
readSize = u_strlen(buffer);
if (readSize != expectedSize*repetitions) {
log_err("Buffer is the wrong size. Got %d Expected %d\n", u_strlen(buffer), expectedSize*repetitions);
}
if (buffer[(expectedSize*repetitions) + 1] != 0xBEEF) {
log_err("u_fgets wrote too much data\n");
}
if (u_strcmp(buffer, expectedBuffer) != 0) {
log_err("Did get expected string back\n");
}
if (strcmp(u_fgetcodepage(myFile), "UTF-16") != 0) {
log_err("Got %s instead of UTF-16\n", u_fgetcodepage(myFile));
}
u_fclose(myFile);
log_verbose("Now trying a multi-byte encoding (UTF-8).\n");
myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, "UTF-8");
u_fputc(0x3BC, myFile);
u_fputc(0xFF41, myFile);
u_memset(buffer, 0xBEEF, sizeof(buffer)/sizeof(buffer[0]));
u_memset(expectedBuffer, 0, sizeof(expectedBuffer)/sizeof(expectedBuffer[0]));
u_uastrncpy(buffer, testStr, expectedSize+1);
for (repetitions = 0; repetitions < 16; repetitions++) {
u_file_write(buffer, expectedSize, myFile);
u_strcat(expectedBuffer, buffer);
}
u_fclose(myFile);
u_memset(buffer, 0xBEEF, sizeof(buffer)/sizeof(buffer[0]));
myFile = u_fopen(STANDARD_TEST_FILE, "r", NULL, "UTF-8");
if (strcmp(u_fgetcodepage(myFile), "UTF-8") != 0) {
log_err("Got %s instead of UTF-8\n", u_fgetcodepage(myFile));
}
if (u_fgetc(myFile) != 0x3BC) {
log_err("The first character is wrong\n");
}
if (u_fgetc(myFile) != 0xFF41) {
log_err("The second character is wrong\n");
}
if (u_fgets(buffer, sizeof(buffer)/sizeof(buffer[0]), myFile) != buffer) {
log_err("Didn't get the buffer back\n");
return;
}
readSize = u_strlen(buffer);
if (readSize != expectedSize*repetitions) {
log_err("Buffer is the wrong size. Got %d Expected %d\n", u_strlen(buffer), expectedSize*repetitions);
}
if (buffer[(expectedSize*repetitions) + 1] != 0xBEEF) {
log_err("u_fgets wrote too much data\n");
}
if (u_strcmp(buffer, expectedBuffer) != 0) {
log_err("Did get expected string back\n");
}
u_fclose(myFile);
log_verbose("Now trying a multi-byte encoding (UTF-8) with a really small buffer.\n");
myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, "UTF-8");
u_fputc(0xFF41, myFile);
u_memset(buffer, 0xBEEF, sizeof(buffer)/sizeof(buffer[0]));
u_memset(expectedBuffer, 0, sizeof(expectedBuffer)/sizeof(expectedBuffer[0]));
u_uastrncpy(buffer, testStr, expectedSize+1);
for (repetitions = 0; repetitions < 1; repetitions++) {
u_file_write(buffer, expectedSize, myFile);
u_strcat(expectedBuffer, buffer);
}
u_fclose(myFile);
u_memset(buffer, 0xBEEF, sizeof(buffer)/sizeof(buffer[0]));
myFile = u_fopen(STANDARD_TEST_FILE, "r", NULL, "UTF-8");
if (u_fgets(buffer, 2, myFile) != buffer) {
log_err("Didn't get the buffer back\n");
return;
}
readSize = u_strlen(buffer);
if (readSize != 1) {
log_err("Buffer is the wrong size. Got %d Expected %d\n", u_strlen(buffer), 1);
}
if (buffer[0] != 0xFF41 || buffer[1] != 0) {
log_err("Did get expected string back\n");
}
if (buffer[2] != 0xBEEF) {
log_err("u_fgets wrote too much data\n");
}
u_fclose(myFile);
}
static void TestFileReadBuffering(void) {
UChar buffer[1024];
UFILE *myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, "UTF-16");
int32_t how_many;
int32_t repetitions;
u_memset(buffer, 0xBEEF, sizeof(buffer)/sizeof(buffer[0]));
for (repetitions = 0; repetitions < 2; repetitions++) {
u_file_write(buffer, sizeof(buffer)/sizeof(buffer[0]), myFile);
}
u_fclose(myFile);
u_memset(buffer, 0xDEAD, sizeof(buffer)/sizeof(buffer[0]));
myFile = u_fopen(STANDARD_TEST_FILE, "r", NULL, "UTF-16");
how_many = u_file_read(buffer, 1024, myFile);
if (how_many != 1024 || buffer[1023] != 0xBEEF) {
log_err("u_file_read read too much or not enough data\n");
}
u_fclose(myFile);
}
static void TestfgetsLineCount(void) {
UChar buffer[2048];
UChar expectedBuffer[2048];
char charBuffer[2048];
static const char testStr[] = "This is a test string that tests u_fgets. It makes sure that we don't try to read too much!";
UFILE *myFile = NULL;
FILE *stdFile = fopen(STANDARD_TEST_FILE, "w");
int32_t expectedSize = (int32_t)strlen(testStr);
int32_t repetitions;
int32_t nlRepetitions;
if (stdFile == NULL) {
log_err("Can't write test file.\n");
return;
}
u_memset(expectedBuffer, 0, sizeof(expectedBuffer)/sizeof(expectedBuffer[0]));
for (repetitions = 0; repetitions < 16; repetitions++) {
fwrite(testStr, sizeof(testStr[0]), expectedSize, stdFile);
for (nlRepetitions = 0; nlRepetitions < repetitions; nlRepetitions++) {
fwrite("\n", sizeof(testStr[0]), 1, stdFile);
}
}
fclose(stdFile);
myFile = u_fopen(STANDARD_TEST_FILE, "r", NULL, NULL);
stdFile = fopen(STANDARD_TEST_FILE, "r");
for (;;) {
char *returnedCharBuffer;
UChar *returnedUCharBuffer;
u_memset(buffer, 0xBEEF, sizeof(buffer)/sizeof(buffer[0]));
returnedCharBuffer = fgets(charBuffer, sizeof(charBuffer)/sizeof(charBuffer[0]), stdFile);
returnedUCharBuffer = u_fgets(buffer, sizeof(buffer)/sizeof(buffer[0]), myFile);
if (!returnedCharBuffer && !returnedUCharBuffer) {
/* Both returned NULL. stop. */
break;
}
if (returnedCharBuffer != charBuffer) {
log_err("Didn't get the charBuffer back\n");
continue;
}
u_uastrncpy(expectedBuffer, charBuffer, (int32_t)strlen(charBuffer)+1);
if (returnedUCharBuffer != buffer) {
log_err("Didn't get the buffer back\n");
continue;
}
if (u_strcmp(buffer, expectedBuffer) != 0) {
log_err("buffers are different\n");
}
if (buffer[u_strlen(buffer)+1] != 0xBEEF) {
log_err("u_fgets wrote too much\n");
}
}
fclose(stdFile);
u_fclose(myFile);
}
static void TestfgetsNewLineHandling(void) {
UChar buffer[256];
static const UChar testUStr[][16] = {
{0x000D, 0},
{0x000D, 0x000A, 0},
{0x000D, 0},
{0x000D, 0},
{0x0085, 0},
{0x000A, 0},
{0x000D, 0},
{0x000B, 0},
{0x000C, 0},
{0x000C, 0},
{0x2028, 0},
{0x0085, 0},
{0x2029, 0},
{0x0085, 0},
{0x008B, 0x000D, 0},
{0x00A0, 0x000D, 0x000A, 0},
{0x3000, 0x000D, 0},
{0xd800, 0xdfff, 0x000D, 0},
{0x00AB, 0x0085, 0},
{0x00AC, 0x000A, 0},
{0x00AD, 0x000D, 0},
{0x00BA, 0x000B, 0},
{0x00AB, 0x000C, 0},
{0x00B1, 0x000C, 0},
{0x30BB, 0x2028, 0},
{0x00A5, 0x0085, 0},
{0x0080, 0x2029, 0},
{0x00AF, 0x0085, 0}
};
UFILE *myFile = NULL;
int32_t lineIdx;
myFile = u_fopen(STANDARD_TEST_FILE, "wb", NULL, "UTF-8");
if (myFile == NULL) {
log_err("Can't write test file.\n");
return;
}
for (lineIdx = 0; lineIdx < (int32_t)(sizeof(testUStr)/sizeof(testUStr[0])); lineIdx++) {
u_file_write(testUStr[lineIdx], u_strlen(testUStr[lineIdx]), myFile);
}
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "rb", NULL, "UTF-8");
for (lineIdx = 0; lineIdx < (int32_t)(sizeof(testUStr)/sizeof(testUStr[0])); lineIdx++) {
UChar *returnedUCharBuffer;
u_memset(buffer, 0xBEEF, sizeof(buffer)/sizeof(buffer[0]));
returnedUCharBuffer = u_fgets(buffer, sizeof(buffer)/sizeof(buffer[0]), myFile);
if (!returnedUCharBuffer) {
/* Returned NULL. stop. */
break;
}
if (u_strcmp(buffer, testUStr[lineIdx]) != 0) {
log_err("buffers are different at index = %d\n", lineIdx);
}
if (buffer[u_strlen(buffer)+1] != 0xBEEF) {
log_err("u_fgets wrote too much\n");
}
}
if (lineIdx != (int32_t)(sizeof(testUStr)/sizeof(testUStr[0]))) {
log_err("u_fgets read too much\n");
}
if (u_fgets(buffer, sizeof(buffer)/sizeof(buffer[0]), myFile) != NULL) {
log_err("u_file_write wrote too much\n");
}
u_fclose(myFile);
}
static void TestLineCount(const char *prefixLine, const char *line, int32_t numRepititions) {
UChar buffer[64];
UChar expectedBuffer[64];
int32_t lineLen = strlen(line);
UChar *returnedUCharBuffer;
int32_t repetitions;
UFILE *myFile = NULL;
FILE *stdFile = fopen(STANDARD_TEST_FILE, "wb");
if (stdFile == NULL) {
log_err("Can't write test file.\n");
return;
}
/* Write a prefix line and then write a bunch of lines */
fwrite(prefixLine, strlen(prefixLine), 1, stdFile);
for (repetitions = 0; repetitions < numRepititions; repetitions++) {
fwrite(line, lineLen, 1, stdFile);
}
fclose(stdFile);
myFile = u_fopen(STANDARD_TEST_FILE, "rb", NULL, NULL);
if (myFile == NULL) {
log_err("Can't read test file.\n");
return;
}
/* Read the prefix line. This can make sure that a Windows newline is either on a boundary or before it. */
u_uastrncpy(expectedBuffer, prefixLine, (int32_t)strlen(prefixLine)+1);
returnedUCharBuffer = u_fgets(buffer, sizeof(buffer)/sizeof(buffer[0]), myFile);
if (u_strcmp(returnedUCharBuffer, expectedBuffer) != 0) {
log_err("prefix buffer is different. prefix=\"%s\"\n", prefixLine);
return;
}
u_uastrncpy(expectedBuffer, line, (int32_t)strlen(line)+1);
for (repetitions = 0; ; repetitions++) {
u_memset(buffer, 0xBEEF, sizeof(buffer)/sizeof(buffer[0]));
returnedUCharBuffer = u_fgets(buffer, sizeof(buffer)/sizeof(buffer[0]), myFile);
if (!returnedUCharBuffer) {
/* returned NULL. stop. */
break;
}
if (u_strcmp(buffer, expectedBuffer) != 0) {
log_err("buffers are different at count %d\n", repetitions);
}
if (buffer[u_strlen(buffer)+1] != 0xBEEF) {
log_err("u_fgets wrote too much\n");
}
}
if (repetitions != numRepititions) {
log_err("got wrong number of lines. got=%d expected=%d\n", repetitions, numRepititions);
}
u_fclose(myFile);
}
static void TestfgetsNewLineCount(void) {
/* This makes sure that lines are correctly handled between buffer boundaries. */
TestLineCount("\n", "\n", 1024); /* Unix newlines */
TestLineCount("\r\n", "\r\n", 1024);/* Windows newlines */
TestLineCount("a\r\n", "\r\n", 1024);/* Windows newlines offset by 1 byte */
TestLineCount("\r\n", "a\r\n", 1024);/* Windows newlines offset with data */
TestLineCount("\n", "a\n", 1024); /* Unix newlines offset with data */
TestLineCount("\n", "\r\n", 1024); /* a mixed number of lines. */
}
static void TestFgetsLineBuffering(void) {
UChar buffer[2003]; /* Use a non-power of 2 or 10 */
UChar *returnedUCharBuffer;
int32_t repetitions;
UFILE *myFile = NULL;
FILE *stdFile = fopen(STANDARD_TEST_FILE, "wb");
if (stdFile == NULL) {
log_err("Can't write test file.\n");
return;
}
u_memset(buffer, 0xBEEF, sizeof(buffer)/sizeof(buffer[0]));
/* Write one very long line */
for (repetitions = 0; repetitions < ((sizeof(buffer)/sizeof(buffer[0]))*2); repetitions++) {
fwrite(repetitions ? "1" : "2", 1, 1, stdFile);
}
fclose(stdFile);
myFile = u_fopen(STANDARD_TEST_FILE, "rb", NULL, NULL);
if (myFile == NULL) {
log_err("Can't read test file.\n");
return;
}
/* Read part of one very long line */
returnedUCharBuffer = u_fgets(buffer, (sizeof(buffer)/sizeof(buffer[0]))-1, myFile);
if (u_strlen(returnedUCharBuffer) != ((sizeof(buffer)/sizeof(buffer[0]))-2)) {
log_err("Line is wrong length. Got %d. Expected %d.\n",
u_strlen(returnedUCharBuffer), ((sizeof(buffer)/sizeof(buffer[0]))-2));
}
/* We better not read too much */
if (buffer[(sizeof(buffer)/sizeof(buffer[0]))-1] != 0xBEEF) {
log_err("Too much data was written\n");
}
u_fclose(myFile);
}
static void TestCodepage(void) {
UFILE *myFile = NULL;
static const UChar strABAccentA[] = { 0x0041, 0x0042, 0x00C1, 0x0043, 0};
static const UChar strBadConversion[] = { 0x0041, 0x0042, 0xfffd, 0x0043, 0};
UChar testBuf[sizeof(strABAccentA)/sizeof(strABAccentA[0])*2]; /* *2 to see if too much was */
char convName[UCNV_MAX_CONVERTER_NAME_LENGTH];
int32_t retVal;
UErrorCode status = U_ZERO_ERROR;
myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, "absurd converter that can't be opened");
if (myFile) {
log_err("Recieved a UFILE * with an invalid codepage parameter\n");
u_fclose(myFile);
}
myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, "ISO-8859-1");
if (myFile == NULL) {
log_err("Can't write test file for iso-8859-1.\n");
return;
}
if (strcmp("ISO-8859-1", u_fgetcodepage(myFile)) != 0) {
log_err("Couldn't get ISO-8859-1 back as opened codepage\n");
}
u_file_write(strABAccentA, u_strlen(strABAccentA), myFile);
u_fclose(myFile);
/* Now see what we got wrote */
myFile = u_fopen(STANDARD_TEST_FILE, "r", NULL, NULL);
if (u_fsetcodepage("ISO-8859-1", myFile) != 0) {
log_err("u_fsetcodepage didn't set the codepage\n");
}
retVal = u_file_read(testBuf, u_strlen(strABAccentA), myFile);
if (u_strncmp(strABAccentA, testBuf, u_strlen(strABAccentA)) != 0) {
log_err("The test data was read and written differently!\n");
}
if (retVal != u_strlen(strABAccentA)) {
log_err("The test data returned different lengths. Got: %d, Expected %d\n", retVal, u_strlen(strABAccentA));
}
u_fclose(myFile);
/* What happens on invalid input? */
myFile = u_fopen(STANDARD_TEST_FILE, "r", NULL, "ISO-8859-1");
if (strcmp(ucnv_getName(u_fgetConverter(myFile), &status), "ISO-8859-1") != 0) {
log_err("u_fgetConverter returned %s\n", ucnv_getName(u_fgetConverter(myFile), &status));
}
if (u_fsetcodepage("UTF-8", myFile) != 0) {
log_err("u_fsetcodepage didn't set the codepage to UTF-8\n");
}
if (strcmp(ucnv_getName(u_fgetConverter(myFile), &status), "UTF-8") != 0) {
log_err("u_fgetConverter returned %s\n", ucnv_getName(u_fgetConverter(myFile), &status));
}
retVal = u_file_read(testBuf, u_strlen(strBadConversion), myFile);
if (u_strncmp(strBadConversion, testBuf, u_strlen(strBadConversion)) != 0) {
log_err("The test data wasn't subsituted as expected\n");
}
u_fclose(myFile);
/* Can't currently swap codepages midstream */
myFile = u_fopen(STANDARD_TEST_FILE, "r", NULL, "ISO-8859-1");
strcpy(convName, u_fgetcodepage(myFile));
u_file_read(testBuf, 1, myFile);
if (u_fsetcodepage("UTF-8", myFile) == 0) {
log_err("u_fsetcodepage set the codepage after reading a byte\n");
}
retVal = u_file_read(testBuf + 1, u_strlen(strABAccentA) - 1, myFile);
if (u_strncmp(strABAccentA, testBuf, u_strlen(strABAccentA)) != 0) {
log_err("u_fsetcodepage changed the codepages after writing data\n");
}
if ((retVal + 1) != u_strlen(strABAccentA)) {
log_err("The test data returned different lengths. Got: %d, Expected %d\n", retVal, u_strlen(strABAccentA));
}
u_frewind(myFile);
retVal = u_file_read(testBuf, u_strlen(strABAccentA), myFile);
if (u_strncmp(strABAccentA, testBuf, u_strlen(strABAccentA)) != 0) {
log_err("The test data was read and written differently!\n");
}
if (retVal != u_strlen(strABAccentA)) {
log_err("The test data returned different lengths. Got: %d, Expected %d\n", retVal, u_strlen(strABAccentA));
}
u_fclose(myFile);
}
#if !UCONFIG_NO_FORMATTING
static void TestFilePrintCompatibility(void) {
UFILE *myFile = u_fopen(STANDARD_TEST_FILE, "wb", "en_US_POSIX", NULL);
FILE *myCFile;
int32_t num;
char cVal;
static const UChar emptyStr[] = {0};
char readBuf[512] = "";
char testBuf[512] = "";
if (myFile == NULL) {
log_err("Can't write test file.\n");
return;
}
#if !UCONFIG_NO_FORMATTING
if (strcmp(u_fgetlocale(myFile), "en_US_POSIX") != 0) {
log_err("Got %s instead of en_US_POSIX for locale\n", u_fgetlocale(myFile));
}
#endif
/* Compare against C API compatibility */
for (num = -STANDARD_TEST_NUM_RANGE; num < STANDARD_TEST_NUM_RANGE; num++) {
u_fprintf(myFile, "%x ", num);
u_fprintf(myFile, "%X ", num);
u_fprintf(myFile, "%o ", num);
u_fprintf(myFile, "%d ", num);
u_fprintf(myFile, "%i ", num);
u_fprintf(myFile, "%f ", (double)num);
/* u_fprintf(myFile, "%e ", (double)num);
u_fprintf(myFile, "%E ", (double)num);*/
u_fprintf(myFile, "%g ", (double)num);
u_fprintf(myFile, "%G", (double)num);
u_fputs(emptyStr, myFile);
}
u_fprintf_u(myFile, NEW_LINE);
for (num = 0; num < 0x80; num++) {
u_fprintf(myFile, "%c", num);
}
u_fclose(myFile);
myCFile = fopen(STANDARD_TEST_FILE, "rb");
if (myCFile == NULL) {
log_err("Can't read test file.");
return;
}
for (num = -STANDARD_TEST_NUM_RANGE; num < STANDARD_TEST_NUM_RANGE; num++) {
fscanf(myCFile, "%s", readBuf);
sprintf(testBuf, "%x", (int)num);
if (strcmp(readBuf, testBuf) != 0) {
log_err("%%x Got: \"%s\", Expected: \"%s\"\n", readBuf, testBuf);
}
fscanf(myCFile, "%s", readBuf);
sprintf(testBuf, "%X", (int)num);
if (strcmp(readBuf, testBuf) != 0) {
log_err("%%X Got: \"%s\", Expected: \"%s\"\n", readBuf, testBuf);
}
fscanf(myCFile, "%s", readBuf);
sprintf(testBuf, "%o", (int)num);
if (strcmp(readBuf, testBuf) != 0) {
log_err("%%o Got: \"%s\", Expected: \"%s\"\n", readBuf, testBuf);
}
/* fprintf is not compatible on all platforms e.g. the iSeries */
fscanf(myCFile, "%s", readBuf);
sprintf(testBuf, "%d", (int)num);
if (strcmp(readBuf, testBuf) != 0) {
log_err("%%d Got: \"%s\", Expected: \"%s\"\n", readBuf, testBuf);
}
fscanf(myCFile, "%s", readBuf);
sprintf(testBuf, "%i", (int)num);
if (strcmp(readBuf, testBuf) != 0) {
log_err("%%i Got: \"%s\", Expected: \"%s\"\n", readBuf, testBuf);
}
fscanf(myCFile, "%s", readBuf);
sprintf(testBuf, "%f", (double)num);
if (strcmp(readBuf, testBuf) != 0) {
log_err("%%f Got: \"%s\", Expected: \"%s\"\n", readBuf, testBuf);
}
/* fscanf(myCFile, "%s", readBuf);
sprintf(testBuf, "%e", (double)num);
if (strcmp(readBuf, testBuf) != 0) {
log_err("%%e Got: \"%s\", Expected: \"%s\"\n", readBuf, testBuf);
}
fscanf(myCFile, "%s", readBuf);
sprintf(testBuf, "%E", (double)num);
if (strcmp(readBuf, testBuf) != 0) {
log_err("%%E Got: \"%s\", Expected: \"%s\"\n", readBuf, testBuf);
}*/
fscanf(myCFile, "%s", readBuf);
sprintf(testBuf, "%g", (double)num);
if (strcmp(readBuf, testBuf) != 0) {
log_err("%%g Got: \"%s\", Expected: \"%s\"\n", readBuf, testBuf);
}
fscanf(myCFile, "%s", readBuf);
sprintf(testBuf, "%G", (double)num);
if (strcmp(readBuf, testBuf) != 0) {
log_err("%%G Got: \"%s\", Expected: \"%s\"\n", readBuf, testBuf);
}
}
/* Properly eat the newlines */
for (num = 0; num < (int32_t)strlen(C_NEW_LINE); num++) {
fscanf(myCFile, "%c", &cVal);
if (cVal != C_NEW_LINE[num]) {
log_err("OS newline error\n");
}
}
for (num = 0; num < (int32_t)strlen(C_NEW_LINE); num++) {
fscanf(myCFile, "%c", &cVal);
if (cVal != C_NEW_LINE[num]) {
log_err("ustdio newline error\n");
}
}
for (num = 0; num < 0x80; num++) {
cVal = -1;
fscanf(myCFile, "%c", &cVal);
if (num != cVal) {
log_err("%%c Got: 0x%x, Expected: 0x%x\n", cVal, num);
}
}
fclose(myCFile);
}
#endif
#define TestFPrintFormat(uFormat, uValue, cFormat, cValue) \
myFile = u_fopen(STANDARD_TEST_FILE, "w", "en_US_POSIX", NULL);\
if (myFile == NULL) {\
log_err("Can't write test file for %s.\n", uFormat);\
return;\
}\
/* Reinitialize the buffer to verify null termination works. */\
u_memset(uBuffer, 0x2a, sizeof(uBuffer)/sizeof(*uBuffer));\
memset(buffer, '*', sizeof(buffer)/sizeof(*buffer));\
\
uNumPrinted = u_fprintf(myFile, uFormat, uValue);\
u_fclose(myFile);\
myFile = u_fopen(STANDARD_TEST_FILE, "r", "en_US_POSIX", NULL);\
u_fgets(uBuffer, sizeof(uBuffer)/sizeof(*uBuffer), myFile);\
u_fclose(myFile);\
u_austrncpy(compBuffer, uBuffer, sizeof(uBuffer)/sizeof(*uBuffer));\
cNumPrinted = sprintf(buffer, cFormat, cValue);\
if (strcmp(buffer, compBuffer) != 0) {\
log_err("%" uFormat " Got: \"%s\", Expected: \"%s\"\n", compBuffer, buffer);\
}\
if (cNumPrinted != uNumPrinted) {\
log_err("%" uFormat " number printed Got: %d, Expected: %d\n", uNumPrinted, cNumPrinted);\
}\
if (buffer[uNumPrinted+1] != '*') {\
log_err("%" uFormat " too much stored\n");\
}\
#if !UCONFIG_NO_FORMATTING
static void TestFprintfFormat(void) {
static const UChar abcUChars[] = {0x61,0x62,0x63,0};
static const char abcChars[] = "abc";
UChar uBuffer[256];
char buffer[256];
char compBuffer[256];
int32_t uNumPrinted;
int32_t cNumPrinted;
UFILE *myFile;
TestFPrintFormat("%8S", abcUChars, "%8s", abcChars);
TestFPrintFormat("%-8S", abcUChars, "%-8s", abcChars);
TestFPrintFormat("%.2S", abcUChars, "%.2s", abcChars); /* strlen is 3 */
TestFPrintFormat("%8s", abcChars, "%8s", abcChars);
TestFPrintFormat("%-8s", abcChars, "%-8s", abcChars);
TestFPrintFormat("%.2s", abcChars, "%.2s", abcChars); /* strlen is 3 */
TestFPrintFormat("%8c", (char)'e', "%8c", (char)'e');
TestFPrintFormat("%-8c", (char)'e', "%-8c", (char)'e');
TestFPrintFormat("%8C", (UChar)0x65, "%8c", (char)'e');
TestFPrintFormat("%-8C", (UChar)0x65, "%-8c", (char)'e');
TestFPrintFormat("%f", 1.23456789, "%f", 1.23456789);
TestFPrintFormat("%f", 12345.6789, "%f", 12345.6789);
TestFPrintFormat("%f", 123456.789, "%f", 123456.789);
TestFPrintFormat("%f", 1234567.89, "%f", 1234567.89);
TestFPrintFormat("%10f", 1.23456789, "%10f", 1.23456789);
TestFPrintFormat("%-10f", 1.23456789, "%-10f", 1.23456789);
TestFPrintFormat("%10f", 123.456789, "%10f", 123.456789);
TestFPrintFormat("%10.4f", 123.456789, "%10.4f", 123.456789);
TestFPrintFormat("%-10f", 123.456789, "%-10f", 123.456789);
/* TestFPrintFormat("%g", 12345.6789, "%g", 12345.6789);
TestFPrintFormat("%g", 123456.789, "%g", 123456.789);
TestFPrintFormat("%g", 1234567.89, "%g", 1234567.89);
TestFPrintFormat("%G", 123456.789, "%G", 123456.789);
TestFPrintFormat("%G", 1234567.89, "%G", 1234567.89);*/
TestFPrintFormat("%10g", 1.23456789, "%10g", 1.23456789);
TestFPrintFormat("%10.4g", 1.23456789, "%10.4g", 1.23456789);
TestFPrintFormat("%-10g", 1.23456789, "%-10g", 1.23456789);
TestFPrintFormat("%10g", 123.456789, "%10g", 123.456789);
TestFPrintFormat("%-10g", 123.456789, "%-10g", 123.456789);
TestFPrintFormat("%8x", 123456, "%8x", 123456);
TestFPrintFormat("%-8x", 123456, "%-8x", 123456);
TestFPrintFormat("%08x", 123456, "%08x", 123456);
TestFPrintFormat("%8X", 123456, "%8X", 123456);
TestFPrintFormat("%-8X", 123456, "%-8X", 123456);
TestFPrintFormat("%08X", 123456, "%08X", 123456);
TestFPrintFormat("%#x", 123456, "%#x", 123456);
TestFPrintFormat("%#x", -123456, "%#x", -123456);
TestFPrintFormat("%8o", 123456, "%8o", 123456);
TestFPrintFormat("%-8o", 123456, "%-8o", 123456);
TestFPrintFormat("%08o", 123456, "%08o", 123456);
TestFPrintFormat("%#o", 123, "%#o", 123);
TestFPrintFormat("%#o", -123, "%#o", -123);
TestFPrintFormat("%8u", 123456, "%8u", 123456);
TestFPrintFormat("%-8u", 123456, "%-8u", 123456);
TestFPrintFormat("%08u", 123456, "%08u", 123456);
TestFPrintFormat("%8u", -123456, "%8u", -123456);
TestFPrintFormat("%-8u", -123456, "%-8u", -123456);
TestFPrintFormat("%.5u", 123456, "%.5u", 123456);
TestFPrintFormat("%.6u", 123456, "%.6u", 123456);
TestFPrintFormat("%.7u", 123456, "%.7u", 123456);
TestFPrintFormat("%8d", 123456, "%8d", 123456);
TestFPrintFormat("%-8d", 123456, "%-8d", 123456);
TestFPrintFormat("%08d", 123456, "%08d", 123456);
TestFPrintFormat("% d", 123456, "% d", 123456);
TestFPrintFormat("% d", -123456, "% d", -123456);
TestFPrintFormat("%8i", 123456, "%8i", 123456);
TestFPrintFormat("%-8i", 123456, "%-8i", 123456);
TestFPrintFormat("%08i", 123456, "%08i", 123456);
log_verbose("Get really crazy with the formatting.\n");
TestFPrintFormat("%-#12x", 123, "%-#12x", 123);
TestFPrintFormat("%-#12x", -123, "%-#12x", -123);
TestFPrintFormat("%#12x", 123, "%#12x", 123);
TestFPrintFormat("%#12x", -123, "%#12x", -123);
TestFPrintFormat("%-+12d", 123, "%-+12d", 123);
TestFPrintFormat("%-+12d", -123, "%-+12d", -123);
TestFPrintFormat("%- 12d", 123, "%- 12d", 123);
TestFPrintFormat("%- 12d", -123, "%- 12d", -123);
TestFPrintFormat("%+12d", 123, "%+12d", 123);
TestFPrintFormat("%+12d", -123, "%+12d", -123);
TestFPrintFormat("% 12d", 123, "% 12d", 123);
TestFPrintFormat("% 12d", -123, "% 12d", -123);
TestFPrintFormat("%12d", 123, "%12d", 123);
TestFPrintFormat("%12d", -123, "%12d", -123);
TestFPrintFormat("%.12d", 123, "%.12d", 123);
TestFPrintFormat("%.12d", -123, "%.12d", -123);
TestFPrintFormat("%-+12.1f", 1.234, "%-+12.1f", 1.234);
TestFPrintFormat("%-+12.1f", -1.234, "%-+12.1f", -1.234);
TestFPrintFormat("%- 12.10f", 1.234, "%- 12.10f", 1.234);
TestFPrintFormat("%- 12.1f", -1.234, "%- 12.1f", -1.234);
TestFPrintFormat("%+12.1f", 1.234, "%+12.1f", 1.234);
TestFPrintFormat("%+12.1f", -1.234, "%+12.1f", -1.234);
TestFPrintFormat("% 12.1f", 1.234, "% 12.1f", 1.234);
TestFPrintFormat("% 12.1f", -1.234, "% 12.1f", -1.234);
TestFPrintFormat("%12.1f", 1.234, "%12.1f", 1.234);
TestFPrintFormat("%12.1f", -1.234, "%12.1f", -1.234);
TestFPrintFormat("%.2f", 1.234, "%.2f", 1.234);
TestFPrintFormat("%.2f", -1.234, "%.2f", -1.234);
TestFPrintFormat("%3f", 1.234, "%3f", 1.234);
TestFPrintFormat("%3f", -1.234, "%3f", -1.234);
myFile = u_fopen(STANDARD_TEST_FILE, "w", "en_US_POSIX", NULL);
/* Reinitialize the buffer to verify null termination works. */
u_memset(uBuffer, 0x2a, sizeof(uBuffer)/sizeof(*uBuffer));
memset(buffer, '*', sizeof(buffer)/sizeof(*buffer));
uNumPrinted = u_fprintf(myFile, "%d % d %d", -1234, 1234, 1234);
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "r", "en_US_POSIX", NULL);
u_fgets(uBuffer, sizeof(uBuffer)/sizeof(*uBuffer), myFile);
u_fclose(myFile);
u_austrncpy(compBuffer, uBuffer, sizeof(uBuffer)/sizeof(*uBuffer));
cNumPrinted = sprintf(buffer, "%d % d %d", -1234, 1234, 1234);
if (strcmp(buffer, compBuffer) != 0) {
log_err("%%d %% d %%d Got: \"%s\", Expected: \"%s\"\n", compBuffer, buffer);
}
if (cNumPrinted != uNumPrinted) {
log_err("%%d %% d %%d number printed Got: %d, Expected: %d\n", uNumPrinted, cNumPrinted);
}
if (buffer[uNumPrinted+1] != '*') {
log_err("%%d %% d %%d too much stored\n");
}
}
#endif
#undef TestFPrintFormat
#if !UCONFIG_NO_FORMATTING
static void TestFScanSetFormat(const char *format, const UChar *uValue, const char *cValue, UBool expectedToPass) {
UFILE *myFile;
UChar uBuffer[256];
char buffer[256];
char compBuffer[256];
int32_t uNumScanned;
int32_t cNumScanned;
myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, NULL);
if (myFile == NULL) {
log_err("Can't write test file for %s.\n", format);
return;
}
/* Reinitialize the buffer to verify null termination works. */
u_memset(uBuffer, 0x2a, sizeof(uBuffer)/sizeof(*uBuffer));
uBuffer[sizeof(uBuffer)/sizeof(*uBuffer)-1] = 0;
memset(buffer, '*', sizeof(buffer)/sizeof(*buffer));
buffer[sizeof(buffer)/sizeof(*buffer)-1] = 0;
u_fprintf(myFile, "%S", uValue);
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "r", "en_US_POSIX", NULL);
uNumScanned = u_fscanf(myFile, format, uBuffer);
u_fclose(myFile);
if (expectedToPass) {
u_austrncpy(compBuffer, uBuffer, sizeof(uBuffer)/sizeof(*uBuffer));
cNumScanned = sscanf(cValue, format, buffer);
if (strncmp(buffer, compBuffer, sizeof(buffer)/sizeof(*buffer)) != 0) {
log_err("%s Got: \"%s\", Expected: \"%s\"\n", format, compBuffer, buffer);
}
if (cNumScanned != uNumScanned) {
log_err("%s number printed Got: %d, Expected: %d\n", format, uNumScanned, cNumScanned);
}
if (uNumScanned > 0 && uBuffer[u_strlen(uBuffer)+1] != 0x2a) {
log_err("%s too much stored\n", format);
}
}
else {
if (uNumScanned != 0 || uBuffer[0] != 0x2a || uBuffer[1] != 0x2a) {
log_err("%s too much stored on a failure\n", format);
}
}
}
#endif
#if !UCONFIG_NO_FORMATTING
static void TestFScanset(void) {
static const UChar abcUChars[] = {0x61,0x62,0x63,0x63,0x64,0x65,0x66,0x67,0};
static const char abcChars[] = "abccdefg";
TestFScanSetFormat("%[bc]S", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[cb]S", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[ab]S", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[ba]S", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[ab]", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[ba]", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[abcdefgh]", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[;hgfedcba]", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[^a]", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[^e]", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[^ed]", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[^dc]", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[^e] ", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%1[ab] ", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%2[^f]", abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[qrst]", abcUChars, abcChars, TRUE);
/* Extra long string for testing */
TestFScanSetFormat(" %[qrst]",
abcUChars, abcChars, TRUE);
TestFScanSetFormat("%[a-]", abcUChars, abcChars, TRUE);
/* Bad format */
TestFScanSetFormat("%[f-a]", abcUChars, abcChars, FALSE);
TestFScanSetFormat("%[c-a]", abcUChars, abcChars, FALSE);
TestFScanSetFormat("%[a", abcUChars, abcChars, FALSE);
/* The following is not deterministic on Windows */
/* TestFScanSetFormat("%[a-", abcUChars, abcChars);*/
/* TODO: Need to specify precision with a "*" */
}
#endif
#if !UCONFIG_NO_FORMATTING
static void TestBadFScanfFormat(const char *format, const UChar *uValue, const char *cValue) {
UFILE *myFile;
UChar uBuffer[256];
int32_t uNumScanned;
myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, NULL);
if (myFile == NULL) {
log_err("Can't write test file for %s.\n", format);
return;
}
/* Reinitialize the buffer to verify null termination works. */
u_memset(uBuffer, 0x2a, sizeof(uBuffer)/sizeof(*uBuffer));
uBuffer[sizeof(uBuffer)/sizeof(*uBuffer)-1] = 0;
u_fprintf(myFile, "%S", uValue);
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "r", "en_US_POSIX", NULL);
uNumScanned = u_fscanf(myFile, format, uBuffer);
u_fclose(myFile);
if (uNumScanned != 0 || uBuffer[0] != 0x2a || uBuffer[1] != 0x2a) {
log_err("%s too much stored on a failure\n", format);
}
}
#endif
#if !UCONFIG_NO_FORMATTING
static void TestBadScanfFormat(void) {
static const UChar abcUChars[] = {0x61,0x62,0x63,0x63,0x64,0x65,0x66,0x67,0};
static const char abcChars[] = "abccdefg";
TestBadFScanfFormat("%[] ", abcUChars, abcChars);
}
#endif
#if !UCONFIG_NO_FORMATTING
static void Test_u_vfprintf(const char *expectedResult, const char *format, ...) {
UChar uBuffer[256];
UChar uBuffer2[256];
va_list ap;
int32_t count;
UFILE *myFile;
myFile = u_fopen(STANDARD_TEST_FILE, "w", "en_US_POSIX", "UTF-8");
if (!myFile) {
log_err("Test file can't be opened\n");
return;
}
va_start(ap, format);
count = u_vfprintf(myFile, format, ap);
va_end(ap);
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "r", "en_US_POSIX", "UTF-8");
if (!myFile) {
log_err("Test file can't be opened\n");
return;
}
u_fgets(uBuffer, sizeof(uBuffer)/sizeof(*uBuffer), myFile);
u_uastrcpy(uBuffer2, expectedResult);
if (u_strcmp(uBuffer, uBuffer2) != 0) {
log_err("Got two different results for \"%s\" expected \"%s\"\n", format, expectedResult);
}
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "w", "en_US_POSIX", NULL);
if (!myFile) {
log_err("Test file can't be opened\n");
return;
}
u_uastrcpy(uBuffer, format);
va_start(ap, format);
count = u_vfprintf_u(myFile, uBuffer, ap);
va_end(ap);
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "r", "en_US_POSIX", NULL);
if (!myFile) {
log_err("Test file can't be opened\n");
return;
}
u_fgets(uBuffer, sizeof(uBuffer)/sizeof(*uBuffer), myFile);
u_uastrcpy(uBuffer2, expectedResult);
if (u_strcmp(uBuffer, uBuffer2) != 0) {
log_err("Got two different results for \"%s\" expected \"%s\"\n", format, expectedResult);
}
u_fclose(myFile);
}
static void TestVargs(void) {
Test_u_vfprintf("8 9 a B 8.9", "%d %u %x %X %.1f", 8, 9, 10, 11, 8.9);
}
#endif
static void TestUnicodeFormat(void)
{
#if !UCONFIG_NO_FORMATTING
/* Make sure that invariant conversion doesn't happen on the _u formats. */
UChar myUString[256];
UFILE *myFile;
static const UChar TEST_STR[] = { 0x03BC, 0x0025, 0x0024, 0};
static const UChar PERCENT_S[] = { 0x03BC, 0x0025, 0x0053, 0};
u_memset(myUString, 0x2a, sizeof(myUString)/sizeof(*myUString));
myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, "UTF-8");
if (!myFile) {
log_err("Test file can't be opened\n");
return;
}
u_fprintf_u(myFile, PERCENT_S, TEST_STR);
u_fclose(myFile);
myFile = u_fopen(STANDARD_TEST_FILE, "r", NULL, "UTF-8");
if (!myFile) {
log_err("Test file can't be opened\n");
return;
}
u_fscanf_u(myFile, PERCENT_S, myUString);
u_fclose(myFile);
if (u_strcmp(TEST_STR, myUString) != 0) {
log_err("u_fscanf_u doesn't work.\n");
}
#endif
}
static void TestFileWriteRetval(const char * a_pszEncoding) {
UChar * buffer;
UFILE * myFile;
int32_t count;
int32_t expected = 10000; /* test with large data to test internal buffer looping */
UChar testChar = 0xBEEF;
if (!*a_pszEncoding || 0 == strcmp(a_pszEncoding, "ASCII")) {
testChar = 'A'; /* otherwise read test will fail */
}
buffer = (UChar*) malloc(expected * sizeof(UChar));
if (!buffer) {
log_err("Out of memory\n");
return;
}
/* write */
myFile = u_fopen(STANDARD_TEST_FILE, "w", NULL, a_pszEncoding);
if (!myFile) {
free(buffer);
log_err("Test file can't be opened for write\n");
return;
}
u_memset(buffer, testChar, expected);
count = u_file_write(buffer, expected, myFile);
u_fclose(myFile);
if (count != expected) {
free(buffer);
log_err("u_file_write returned incorrect number of characters written\n");
return;
}
free(buffer);
buffer = NULL;
/* read */
myFile = u_fopen(STANDARD_TEST_FILE, "r", NULL, a_pszEncoding);
if (!myFile) {
log_err("Test file can't be opened for read\n");
return;
}
for (count = 0; count < expected; ++count) {
if (u_fgetc(myFile) != testChar) {
log_err("u_fgetc returned unexpected character\n");
u_fclose(myFile);
return;
}
}
if (u_fgetc(myFile) != U_EOF) {
log_err("u_fgetc did not return expected EOF\n");
u_fclose(myFile);
return;
}
u_fclose(myFile);
}
static void TestFileWriteRetvalUTF16(void) {
TestFileWriteRetval("UTF-16");
}
static void TestFileWriteRetvalUTF8(void) {
TestFileWriteRetval("UTF-8");
}
static void TestFileWriteRetvalASCII(void) {
TestFileWriteRetval("ASCII");
}
static void TestFileWriteRetvalNONE(void) {
TestFileWriteRetval("");
}
U_CFUNC void
addFileTest(TestNode** root) {
#if !UCONFIG_NO_FORMATTING
addTest(root, &TestFile, "file/TestFile");
#endif
addTest(root, &StdinBuffering, "file/StdinBuffering");
addTest(root, &TestfgetsBuffers, "file/TestfgetsBuffers");
addTest(root, &TestFileReadBuffering, "file/TestFileReadBuffering");
addTest(root, &TestfgetsLineCount, "file/TestfgetsLineCount");
addTest(root, &TestfgetsNewLineHandling, "file/TestfgetsNewLineHandling");
addTest(root, &TestfgetsNewLineCount, "file/TestfgetsNewLineCount");
addTest(root, &TestFgetsLineBuffering, "file/TestFgetsLineBuffering");
addTest(root, &TestCodepage, "file/TestCodepage");
addTest(root, &TestFileWriteRetvalUTF16, "file/TestFileWriteRetvalUTF16");
addTest(root, &TestFileWriteRetvalUTF8, "file/TestFileWriteRetvalUTF8");
addTest(root, &TestFileWriteRetvalASCII, "file/TestFileWriteRetvalASCII");
addTest(root, &TestFileWriteRetvalNONE, "file/TestFileWriteRetvalNONE");
#if !UCONFIG_NO_FORMATTING
addTest(root, &TestCodepageAndLocale, "file/TestCodepageAndLocale");
addTest(root, &TestFprintfFormat, "file/TestFprintfFormat");
addTest(root, &TestFScanset, "file/TestFScanset");
addTest(root, &TestFilePrintCompatibility, "file/TestFilePrintCompatibility");
addTest(root, &TestBadScanfFormat, "file/TestBadScanfFormat");
addTest(root, &TestVargs, "file/TestVargs");
addTest(root, &TestUnicodeFormat, "file/TestUnicodeFormat");
#endif
}