zint/backend/tests/test_medical.c

373 lines
20 KiB
C
Raw Normal View History

/*
libzint - the open source barcode library
CODE128: reduce extended latch cut-off from 5 to 4 for better encodation in certain cases (and no pessimizations found so far), props lyngklip (BWIPP); fix extended char latching when exactly 3 extended chars at end; count code set C (not digits) in loop deciding when to shift/latch to extended for better estimate AZTEC: return warning if ECC < 5% (due to bit-stuffing when version given); return error if > 22 layers (Zint 26) for Reader Initialisation symbol requested for better error message AZTEC/HANXIN/QRCODE: consolidate different ECC data size tables into one indexed by ECC DBAR_EXP: check for reduced length <= 77 up front for better error message HANXIN: use `malloc()` rather than `z_alloca()` for large binary array QRCODE: `ecc_level` now 0-based (not 1-based) MICROQR: consolidate different version end routines into one `microqr_end()` and use new `microqr_data` table to simplify code MICROPDF417: use table for max codewords per column library: centralize all error messages using new `errtxt()`, `errtxtf()`, `errtxt_adj()` funcs that protect `symbol->errtxt` from overflow, & try to make error messages more consistent thru-out, adding more feedback info to many, & use positional args "%n$" in prep for l10n (maybe); `is_sane/is_sane_lookup()` -> `not_sane/not_sane_lookup()`, returning 1-based position (zero on failure) instead of bool; `long` ints -> plain `int` (except those dealing with `ftell()`, `fread()` etc) as depend on int being 32-bits already GUI: in "grpDATF.ui" use "PlainText" rather than "RichText" for tracker ratio examples as height of text messing up sometimes manual: clarify Codablock-F length maximum & add examples docs: README: pandoc 3.5, Ubuntu 24.04 CMake: use "-Wpedantic" for Clang only as GNU complains about `errtxtf()` positional args "%n$"
2024-10-28 10:33:33 +13:00
Copyright (C) 2020-2024 Robin Stuart <rstuart114@gmail.com>
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
1. Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
3. Neither the name of the project nor the names of its contributors
may be used to endorse or promote products derived from this software
without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
SUCH DAMAGE.
*/
/* SPDX-License-Identifier: BSD-3-Clause */
#include "testcommon.h"
static void test_large(const testCtx *const p_ctx) {
int debug = p_ctx->debug;
struct item {
int symbology;
int option_2;
char *pattern;
int length;
int ret;
int expected_rows;
int expected_width;
};
/* s/\/\*[ 0-9]*\*\//\=printf("\/\*%3d*\/", line(".") - line("'<")): */
CODE128: reduce extended latch cut-off from 5 to 4 for better encodation in certain cases (and no pessimizations found so far), props lyngklip (BWIPP); fix extended char latching when exactly 3 extended chars at end; count code set C (not digits) in loop deciding when to shift/latch to extended for better estimate AZTEC: return warning if ECC < 5% (due to bit-stuffing when version given); return error if > 22 layers (Zint 26) for Reader Initialisation symbol requested for better error message AZTEC/HANXIN/QRCODE: consolidate different ECC data size tables into one indexed by ECC DBAR_EXP: check for reduced length <= 77 up front for better error message HANXIN: use `malloc()` rather than `z_alloca()` for large binary array QRCODE: `ecc_level` now 0-based (not 1-based) MICROQR: consolidate different version end routines into one `microqr_end()` and use new `microqr_data` table to simplify code MICROPDF417: use table for max codewords per column library: centralize all error messages using new `errtxt()`, `errtxtf()`, `errtxt_adj()` funcs that protect `symbol->errtxt` from overflow, & try to make error messages more consistent thru-out, adding more feedback info to many, & use positional args "%n$" in prep for l10n (maybe); `is_sane/is_sane_lookup()` -> `not_sane/not_sane_lookup()`, returning 1-based position (zero on failure) instead of bool; `long` ints -> plain `int` (except those dealing with `ftell()`, `fread()` etc) as depend on int being 32-bits already GUI: in "grpDATF.ui" use "PlainText" rather than "RichText" for tracker ratio examples as height of text messing up sometimes manual: clarify Codablock-F length maximum & add examples docs: README: pandoc 3.5, Ubuntu 24.04 CMake: use "-Wpedantic" for Clang only as GNU complains about `errtxtf()` positional args "%n$"
2024-10-28 10:33:33 +13:00
static const struct item data[] = {
/* 0*/ { BARCODE_CODABAR, -1, "A+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++B", 103, 0, 1, 1133 },
/* 1*/ { BARCODE_CODABAR, -1, "A++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++B", 104, ZINT_ERROR_TOO_LONG, -1, -1 },
/* 2*/ { BARCODE_CODABAR, 1, "A+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++B", 103, 0, 1, 1143 },
/* 3*/ { BARCODE_CODABAR, 1, "A++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++B", 104, ZINT_ERROR_TOO_LONG, -1, -1 },
/* 4*/ { BARCODE_PHARMA, -1, "131070", 6, 0, 1, 78 },
/* 5*/ { BARCODE_PHARMA, -1, "1", 7, ZINT_ERROR_TOO_LONG, -1, -1 },
/* 6*/ { BARCODE_PHARMA_TWO, -1, "64570080", 8, 0, 2, 31 },
/* 7*/ { BARCODE_PHARMA_TWO, -1, "1", 9, ZINT_ERROR_TOO_LONG, -1, -1 },
/* 8*/ { BARCODE_CODE32, -1, "1", 8, 0, 1, 103 },
/* 9*/ { BARCODE_CODE32, -1, "1", 9, ZINT_ERROR_TOO_LONG, -1, -1 },
};
CODE128: reduce extended latch cut-off from 5 to 4 for better encodation in certain cases (and no pessimizations found so far), props lyngklip (BWIPP); fix extended char latching when exactly 3 extended chars at end; count code set C (not digits) in loop deciding when to shift/latch to extended for better estimate AZTEC: return warning if ECC < 5% (due to bit-stuffing when version given); return error if > 22 layers (Zint 26) for Reader Initialisation symbol requested for better error message AZTEC/HANXIN/QRCODE: consolidate different ECC data size tables into one indexed by ECC DBAR_EXP: check for reduced length <= 77 up front for better error message HANXIN: use `malloc()` rather than `z_alloca()` for large binary array QRCODE: `ecc_level` now 0-based (not 1-based) MICROQR: consolidate different version end routines into one `microqr_end()` and use new `microqr_data` table to simplify code MICROPDF417: use table for max codewords per column library: centralize all error messages using new `errtxt()`, `errtxtf()`, `errtxt_adj()` funcs that protect `symbol->errtxt` from overflow, & try to make error messages more consistent thru-out, adding more feedback info to many, & use positional args "%n$" in prep for l10n (maybe); `is_sane/is_sane_lookup()` -> `not_sane/not_sane_lookup()`, returning 1-based position (zero on failure) instead of bool; `long` ints -> plain `int` (except those dealing with `ftell()`, `fread()` etc) as depend on int being 32-bits already GUI: in "grpDATF.ui" use "PlainText" rather than "RichText" for tracker ratio examples as height of text messing up sometimes manual: clarify Codablock-F length maximum & add examples docs: README: pandoc 3.5, Ubuntu 24.04 CMake: use "-Wpedantic" for Clang only as GNU complains about `errtxtf()` positional args "%n$"
2024-10-28 10:33:33 +13:00
const int data_size = ARRAY_SIZE(data);
int i, length, ret;
struct zint_symbol *symbol = NULL;
char data_buf[128];
testStartSymbol("test_large", &symbol);
for (i = 0; i < data_size; i++) {
if (testContinue(p_ctx, i)) continue;
symbol = ZBarcode_Create();
assert_nonnull(symbol, "Symbol not created\n");
testUtilStrCpyRepeat(data_buf, data[i].pattern, data[i].length);
assert_equal(data[i].length, (int) strlen(data_buf), "i:%d length %d != strlen(data_buf) %d\n", i, data[i].length, (int) strlen(data_buf));
length = testUtilSetSymbol(symbol, data[i].symbology, -1 /*input_mode*/, -1 /*eci*/, -1 /*option_1*/, data[i].option_2, -1, -1 /*output_options*/, data_buf, data[i].length, debug);
ret = ZBarcode_Encode(symbol, (unsigned char *) data_buf, length);
assert_equal(ret, data[i].ret, "i:%d ZBarcode_Encode ret %d != %d (%s)\n", i, ret, data[i].ret, symbol->errtxt);
if (ret < ZINT_ERROR) {
assert_equal(symbol->rows, data[i].expected_rows, "i:%d symbol->rows %d != %d\n", i, symbol->rows, data[i].expected_rows);
assert_equal(symbol->width, data[i].expected_width, "i:%d symbol->width %d != %d\n", i, symbol->width, data[i].expected_width);
}
ZBarcode_Delete(symbol);
}
testFinish();
}
static void test_hrt(const testCtx *const p_ctx) {
int debug = p_ctx->debug;
struct item {
int symbology;
int option_2;
char *data;
char *expected;
};
/* s/\/\*[ 0-9]*\*\//\=printf("\/\*%3d*\/", line(".") - line("'<")): */
CODE128: reduce extended latch cut-off from 5 to 4 for better encodation in certain cases (and no pessimizations found so far), props lyngklip (BWIPP); fix extended char latching when exactly 3 extended chars at end; count code set C (not digits) in loop deciding when to shift/latch to extended for better estimate AZTEC: return warning if ECC < 5% (due to bit-stuffing when version given); return error if > 22 layers (Zint 26) for Reader Initialisation symbol requested for better error message AZTEC/HANXIN/QRCODE: consolidate different ECC data size tables into one indexed by ECC DBAR_EXP: check for reduced length <= 77 up front for better error message HANXIN: use `malloc()` rather than `z_alloca()` for large binary array QRCODE: `ecc_level` now 0-based (not 1-based) MICROQR: consolidate different version end routines into one `microqr_end()` and use new `microqr_data` table to simplify code MICROPDF417: use table for max codewords per column library: centralize all error messages using new `errtxt()`, `errtxtf()`, `errtxt_adj()` funcs that protect `symbol->errtxt` from overflow, & try to make error messages more consistent thru-out, adding more feedback info to many, & use positional args "%n$" in prep for l10n (maybe); `is_sane/is_sane_lookup()` -> `not_sane/not_sane_lookup()`, returning 1-based position (zero on failure) instead of bool; `long` ints -> plain `int` (except those dealing with `ftell()`, `fread()` etc) as depend on int being 32-bits already GUI: in "grpDATF.ui" use "PlainText" rather than "RichText" for tracker ratio examples as height of text messing up sometimes manual: clarify Codablock-F length maximum & add examples docs: README: pandoc 3.5, Ubuntu 24.04 CMake: use "-Wpedantic" for Clang only as GNU complains about `errtxtf()` positional args "%n$"
2024-10-28 10:33:33 +13:00
static const struct item data[] = {
/* 0*/ { BARCODE_CODABAR, -1, "A1234B", "A1234B" },
/* 1*/ { BARCODE_CODABAR, -1, "a1234c", "A1234C" }, /* Converts to upper */
/* 2*/ { BARCODE_CODABAR, 1, "A1234B", "A1234B" }, /* Check not included */
/* 3*/ { BARCODE_CODABAR, 2, "A1234B", "A12345B" }, /* Check included */
/* 4*/ { BARCODE_CODABAR, 1, "A123456A", "A123456A" }, /* Check not included */
/* 5*/ { BARCODE_CODABAR, 2, "A123456A", "A123456$A" }, /* Check included */
/* 6*/ { BARCODE_PHARMA, -1, "123456", "" }, /* None */
/* 7*/ { BARCODE_PHARMA_TWO, -1, "123456", "" }, /* None */
/* 8*/ { BARCODE_CODE32, -1, "123456", "A001234564" },
/* 9*/ { BARCODE_CODE32, -1, "12345678", "A123456788" },
};
CODE128: reduce extended latch cut-off from 5 to 4 for better encodation in certain cases (and no pessimizations found so far), props lyngklip (BWIPP); fix extended char latching when exactly 3 extended chars at end; count code set C (not digits) in loop deciding when to shift/latch to extended for better estimate AZTEC: return warning if ECC < 5% (due to bit-stuffing when version given); return error if > 22 layers (Zint 26) for Reader Initialisation symbol requested for better error message AZTEC/HANXIN/QRCODE: consolidate different ECC data size tables into one indexed by ECC DBAR_EXP: check for reduced length <= 77 up front for better error message HANXIN: use `malloc()` rather than `z_alloca()` for large binary array QRCODE: `ecc_level` now 0-based (not 1-based) MICROQR: consolidate different version end routines into one `microqr_end()` and use new `microqr_data` table to simplify code MICROPDF417: use table for max codewords per column library: centralize all error messages using new `errtxt()`, `errtxtf()`, `errtxt_adj()` funcs that protect `symbol->errtxt` from overflow, & try to make error messages more consistent thru-out, adding more feedback info to many, & use positional args "%n$" in prep for l10n (maybe); `is_sane/is_sane_lookup()` -> `not_sane/not_sane_lookup()`, returning 1-based position (zero on failure) instead of bool; `long` ints -> plain `int` (except those dealing with `ftell()`, `fread()` etc) as depend on int being 32-bits already GUI: in "grpDATF.ui" use "PlainText" rather than "RichText" for tracker ratio examples as height of text messing up sometimes manual: clarify Codablock-F length maximum & add examples docs: README: pandoc 3.5, Ubuntu 24.04 CMake: use "-Wpedantic" for Clang only as GNU complains about `errtxtf()` positional args "%n$"
2024-10-28 10:33:33 +13:00
const int data_size = ARRAY_SIZE(data);
int i, length, ret;
struct zint_symbol *symbol = NULL;
testStartSymbol("test_hrt", &symbol);
for (i = 0; i < data_size; i++) {
if (testContinue(p_ctx, i)) continue;
symbol = ZBarcode_Create();
assert_nonnull(symbol, "Symbol not created\n");
length = testUtilSetSymbol(symbol, data[i].symbology, -1 /*input_mode*/, -1 /*eci*/, -1 /*option_1*/, data[i].option_2, -1, -1 /*output_options*/, data[i].data, -1, debug);
ret = ZBarcode_Encode(symbol, (unsigned char *) data[i].data, length);
assert_zero(ret, "i:%d ZBarcode_Encode ret %d != 0 %s\n", i, ret, symbol->errtxt);
assert_zero(strcmp((char *) symbol->text, data[i].expected), "i:%d strcmp(%s, %s) != 0\n", i, symbol->text, data[i].expected);
ZBarcode_Delete(symbol);
}
testFinish();
}
static void test_input(const testCtx *const p_ctx) {
int debug = p_ctx->debug;
struct item {
int symbology;
char *data;
int ret;
int expected_rows;
int expected_width;
const char *expected_errtxt;
int bwipp_cmp;
const char *comment;
};
/* s/\/\*[ 0-9]*\*\//\=printf("\/\*%3d*\/", line(".") - line("'<")): */
CODE128: reduce extended latch cut-off from 5 to 4 for better encodation in certain cases (and no pessimizations found so far), props lyngklip (BWIPP); fix extended char latching when exactly 3 extended chars at end; count code set C (not digits) in loop deciding when to shift/latch to extended for better estimate AZTEC: return warning if ECC < 5% (due to bit-stuffing when version given); return error if > 22 layers (Zint 26) for Reader Initialisation symbol requested for better error message AZTEC/HANXIN/QRCODE: consolidate different ECC data size tables into one indexed by ECC DBAR_EXP: check for reduced length <= 77 up front for better error message HANXIN: use `malloc()` rather than `z_alloca()` for large binary array QRCODE: `ecc_level` now 0-based (not 1-based) MICROQR: consolidate different version end routines into one `microqr_end()` and use new `microqr_data` table to simplify code MICROPDF417: use table for max codewords per column library: centralize all error messages using new `errtxt()`, `errtxtf()`, `errtxt_adj()` funcs that protect `symbol->errtxt` from overflow, & try to make error messages more consistent thru-out, adding more feedback info to many, & use positional args "%n$" in prep for l10n (maybe); `is_sane/is_sane_lookup()` -> `not_sane/not_sane_lookup()`, returning 1-based position (zero on failure) instead of bool; `long` ints -> plain `int` (except those dealing with `ftell()`, `fread()` etc) as depend on int being 32-bits already GUI: in "grpDATF.ui" use "PlainText" rather than "RichText" for tracker ratio examples as height of text messing up sometimes manual: clarify Codablock-F length maximum & add examples docs: README: pandoc 3.5, Ubuntu 24.04 CMake: use "-Wpedantic" for Clang only as GNU complains about `errtxtf()` positional args "%n$"
2024-10-28 10:33:33 +13:00
static const struct item data[] = {
/* 0*/ { BARCODE_CODABAR, "A1234B", 0, 1, 62, "", 1, "" },
/* 1*/ { BARCODE_CODABAR, "1234B", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 358: Does not begin with \"A\", \"B\", \"C\" or \"D\"", 1, "" },
/* 2*/ { BARCODE_CODABAR, "A1234", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 359: Does not end with \"A\", \"B\", \"C\" or \"D\"", 1, "" },
/* 3*/ { BARCODE_CODABAR, "A1234E", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 359: Does not end with \"A\", \"B\", \"C\" or \"D\"", 1, "" },
/* 4*/ { BARCODE_CODABAR, "C123.D", 0, 1, 63, "", 1, "" },
CODE128: reduce extended latch cut-off from 5 to 4 for better encodation in certain cases (and no pessimizations found so far), props lyngklip (BWIPP); fix extended char latching when exactly 3 extended chars at end; count code set C (not digits) in loop deciding when to shift/latch to extended for better estimate AZTEC: return warning if ECC < 5% (due to bit-stuffing when version given); return error if > 22 layers (Zint 26) for Reader Initialisation symbol requested for better error message AZTEC/HANXIN/QRCODE: consolidate different ECC data size tables into one indexed by ECC DBAR_EXP: check for reduced length <= 77 up front for better error message HANXIN: use `malloc()` rather than `z_alloca()` for large binary array QRCODE: `ecc_level` now 0-based (not 1-based) MICROQR: consolidate different version end routines into one `microqr_end()` and use new `microqr_data` table to simplify code MICROPDF417: use table for max codewords per column library: centralize all error messages using new `errtxt()`, `errtxtf()`, `errtxt_adj()` funcs that protect `symbol->errtxt` from overflow, & try to make error messages more consistent thru-out, adding more feedback info to many, & use positional args "%n$" in prep for l10n (maybe); `is_sane/is_sane_lookup()` -> `not_sane/not_sane_lookup()`, returning 1-based position (zero on failure) instead of bool; `long` ints -> plain `int` (except those dealing with `ftell()`, `fread()` etc) as depend on int being 32-bits already GUI: in "grpDATF.ui" use "PlainText" rather than "RichText" for tracker ratio examples as height of text messing up sometimes manual: clarify Codablock-F length maximum & add examples docs: README: pandoc 3.5, Ubuntu 24.04 CMake: use "-Wpedantic" for Clang only as GNU complains about `errtxtf()` positional args "%n$"
2024-10-28 10:33:33 +13:00
/* 5*/ { BARCODE_CODABAR, "C123,D", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 357: Invalid character at position 5 in input (\"0123456789-$:/.+ABCD\" only)", 1, "" },
/* 6*/ { BARCODE_CODABAR, "D:C", 0, 1, 33, "", 1, "" },
CODE128: reduce extended latch cut-off from 5 to 4 for better encodation in certain cases (and no pessimizations found so far), props lyngklip (BWIPP); fix extended char latching when exactly 3 extended chars at end; count code set C (not digits) in loop deciding when to shift/latch to extended for better estimate AZTEC: return warning if ECC < 5% (due to bit-stuffing when version given); return error if > 22 layers (Zint 26) for Reader Initialisation symbol requested for better error message AZTEC/HANXIN/QRCODE: consolidate different ECC data size tables into one indexed by ECC DBAR_EXP: check for reduced length <= 77 up front for better error message HANXIN: use `malloc()` rather than `z_alloca()` for large binary array QRCODE: `ecc_level` now 0-based (not 1-based) MICROQR: consolidate different version end routines into one `microqr_end()` and use new `microqr_data` table to simplify code MICROPDF417: use table for max codewords per column library: centralize all error messages using new `errtxt()`, `errtxtf()`, `errtxt_adj()` funcs that protect `symbol->errtxt` from overflow, & try to make error messages more consistent thru-out, adding more feedback info to many, & use positional args "%n$" in prep for l10n (maybe); `is_sane/is_sane_lookup()` -> `not_sane/not_sane_lookup()`, returning 1-based position (zero on failure) instead of bool; `long` ints -> plain `int` (except those dealing with `ftell()`, `fread()` etc) as depend on int being 32-bits already GUI: in "grpDATF.ui" use "PlainText" rather than "RichText" for tracker ratio examples as height of text messing up sometimes manual: clarify Codablock-F length maximum & add examples docs: README: pandoc 3.5, Ubuntu 24.04 CMake: use "-Wpedantic" for Clang only as GNU complains about `errtxtf()` positional args "%n$"
2024-10-28 10:33:33 +13:00
/* 7*/ { BARCODE_CODABAR, "DCC", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 363: Invalid character at position 1 in input (cannot contain \"A\", \"B\", \"C\" or \"D\")", 1, "" },
/* 8*/ { BARCODE_CODABAR, "A234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123B", ZINT_ERROR_TOO_LONG, -1, -1, "Error 356: Input length 104 too long (maximum 103)", 1, "" },
/* 9*/ { BARCODE_CODABAR, "AB", ZINT_ERROR_TOO_LONG, -1, -1, "Error 362: Input length 2 too short (minimum 3)", 1, "" },
/* 10*/ { BARCODE_PHARMA, "131070", 0, 1, 78, "", 1, "" },
/* 11*/ { BARCODE_PHARMA, "1310700", ZINT_ERROR_TOO_LONG, -1, -1, "Error 350: Input length 7 too long (maximum 6)", 1, "" },
/* 12*/ { BARCODE_PHARMA, "131071", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 352: Input value '131071' out of range (3 to 131070)", 1, "" },
/* 13*/ { BARCODE_PHARMA, "3", 0, 1, 4, "", 1, "" },
/* 14*/ { BARCODE_PHARMA, "2", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 352: Input value '2' out of range (3 to 131070)", 1, "" },
/* 15*/ { BARCODE_PHARMA, "1", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 352: Input value '1' out of range (3 to 131070)", 1, "" },
/* 16*/ { BARCODE_PHARMA, "12A", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 351: Invalid character at position 3 in input (digits only)", 1, "" },
/* 17*/ { BARCODE_PHARMA_TWO, "64570080", 0, 2, 31, "", 1, "" },
/* 18*/ { BARCODE_PHARMA_TWO, "64570081", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 353: Input value '64570081' out of range (4 to 64570080)", 1, "" },
/* 19*/ { BARCODE_PHARMA_TWO, "064570080", ZINT_ERROR_TOO_LONG, -1, -1, "Error 354: Input length 9 too long (maximum 8)", 1, "" },
/* 20*/ { BARCODE_PHARMA_TWO, "4", 0, 2, 3, "", 1, "" },
/* 21*/ { BARCODE_PHARMA_TWO, "3", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 353: Input value '3' out of range (4 to 64570080)", 1, "" },
/* 22*/ { BARCODE_PHARMA_TWO, "2", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 353: Input value '2' out of range (4 to 64570080)", 1, "" },
/* 23*/ { BARCODE_PHARMA_TWO, "1", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 353: Input value '1' out of range (4 to 64570080)", 1, "" },
/* 24*/ { BARCODE_PHARMA_TWO, "123A", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 355: Invalid character at position 4 in input (digits only)", 1, "" },
/* 25*/ { BARCODE_CODE32, "12345678", 0, 1, 103, "", 1, "" },
/* 26*/ { BARCODE_CODE32, "9", 0, 1, 103, "", 0, "BWIPP requires length 8 or 9" },
/* 27*/ { BARCODE_CODE32, "0", 0, 1, 103, "", 0, "BWIPP requires length 8 or 9" },
/* 28*/ { BARCODE_CODE32, "123456789", ZINT_ERROR_TOO_LONG, -1, -1, "Error 360: Input length 9 too long (maximum 8)", 1, "" },
/* 29*/ { BARCODE_CODE32, "A", ZINT_ERROR_INVALID_DATA, -1, -1, "Error 361: Invalid character at position 1 in input (digits only)", 1, "" },
/* 30*/ { BARCODE_CODE32, "99999999", 0, 1, 103, "", 1, "" },
};
CODE128: reduce extended latch cut-off from 5 to 4 for better encodation in certain cases (and no pessimizations found so far), props lyngklip (BWIPP); fix extended char latching when exactly 3 extended chars at end; count code set C (not digits) in loop deciding when to shift/latch to extended for better estimate AZTEC: return warning if ECC < 5% (due to bit-stuffing when version given); return error if > 22 layers (Zint 26) for Reader Initialisation symbol requested for better error message AZTEC/HANXIN/QRCODE: consolidate different ECC data size tables into one indexed by ECC DBAR_EXP: check for reduced length <= 77 up front for better error message HANXIN: use `malloc()` rather than `z_alloca()` for large binary array QRCODE: `ecc_level` now 0-based (not 1-based) MICROQR: consolidate different version end routines into one `microqr_end()` and use new `microqr_data` table to simplify code MICROPDF417: use table for max codewords per column library: centralize all error messages using new `errtxt()`, `errtxtf()`, `errtxt_adj()` funcs that protect `symbol->errtxt` from overflow, & try to make error messages more consistent thru-out, adding more feedback info to many, & use positional args "%n$" in prep for l10n (maybe); `is_sane/is_sane_lookup()` -> `not_sane/not_sane_lookup()`, returning 1-based position (zero on failure) instead of bool; `long` ints -> plain `int` (except those dealing with `ftell()`, `fread()` etc) as depend on int being 32-bits already GUI: in "grpDATF.ui" use "PlainText" rather than "RichText" for tracker ratio examples as height of text messing up sometimes manual: clarify Codablock-F length maximum & add examples docs: README: pandoc 3.5, Ubuntu 24.04 CMake: use "-Wpedantic" for Clang only as GNU complains about `errtxtf()` positional args "%n$"
2024-10-28 10:33:33 +13:00
const int data_size = ARRAY_SIZE(data);
int i, length, ret;
struct zint_symbol *symbol = NULL;
char cmp_buf[8192];
char cmp_msg[1024];
int do_bwipp = (debug & ZINT_DEBUG_TEST_BWIPP) && testUtilHaveGhostscript(); /* Only do BWIPP test if asked, too slow otherwise */
testStartSymbol("test_input", &symbol);
for (i = 0; i < data_size; i++) {
if (testContinue(p_ctx, i)) continue;
symbol = ZBarcode_Create();
assert_nonnull(symbol, "Symbol not created\n");
length = testUtilSetSymbol(symbol, data[i].symbology, -1 /*input_mode*/, -1 /*eci*/, -1 /*option_1*/, -1, -1, -1 /*output_options*/, data[i].data, -1, debug);
ret = ZBarcode_Encode(symbol, (unsigned char *) data[i].data, length);
assert_equal(ret, data[i].ret, "i:%d ZBarcode_Encode ret %d != %d (%s)\n", i, ret, data[i].ret, symbol->errtxt);
assert_zero(strcmp(symbol->errtxt, data[i].expected_errtxt), "i:%d symbol->errtxt %s != %s\n", i, symbol->errtxt, data[i].expected_errtxt);
if (ret < ZINT_ERROR) {
assert_equal(symbol->rows, data[i].expected_rows, "i:%d symbol->rows %d != %d\n", i, symbol->rows, data[i].expected_rows);
assert_equal(symbol->width, data[i].expected_width, "i:%d symbol->width %d != %d\n", i, symbol->width, data[i].expected_width);
if (do_bwipp && testUtilCanBwipp(i, symbol, -1, -1, -1, debug)) {
if (!data[i].bwipp_cmp) {
if (debug & ZINT_DEBUG_TEST_PRINT) printf("i:%d %s not BWIPP compatible (%s)\n", i, testUtilBarcodeName(symbol->symbology), data[i].comment);
} else {
char modules_dump[4096];
assert_notequal(testUtilModulesDump(symbol, modules_dump, sizeof(modules_dump)), -1, "i:%d testUtilModulesDump == -1\n", i);
ret = testUtilBwipp(i, symbol, -1, -1, -1, data[i].data, length, NULL, cmp_buf, sizeof(cmp_buf), NULL);
assert_zero(ret, "i:%d %s testUtilBwipp ret %d != 0\n", i, testUtilBarcodeName(symbol->symbology), ret);
ret = testUtilBwippCmp(symbol, cmp_msg, cmp_buf, modules_dump);
assert_zero(ret, "i:%d %s testUtilBwippCmp %d != 0 %s\n actual: %s\nexpected: %s\n",
i, testUtilBarcodeName(symbol->symbology), ret, cmp_msg, cmp_buf, modules_dump);
}
}
}
ZBarcode_Delete(symbol);
}
testFinish();
}
static void test_encode(const testCtx *const p_ctx) {
int debug = p_ctx->debug;
struct item {
int symbology;
int option_2;
char *data;
int ret;
int expected_rows;
int expected_width;
char *comment;
char *expected;
};
CODE128: reduce extended latch cut-off from 5 to 4 for better encodation in certain cases (and no pessimizations found so far), props lyngklip (BWIPP); fix extended char latching when exactly 3 extended chars at end; count code set C (not digits) in loop deciding when to shift/latch to extended for better estimate AZTEC: return warning if ECC < 5% (due to bit-stuffing when version given); return error if > 22 layers (Zint 26) for Reader Initialisation symbol requested for better error message AZTEC/HANXIN/QRCODE: consolidate different ECC data size tables into one indexed by ECC DBAR_EXP: check for reduced length <= 77 up front for better error message HANXIN: use `malloc()` rather than `z_alloca()` for large binary array QRCODE: `ecc_level` now 0-based (not 1-based) MICROQR: consolidate different version end routines into one `microqr_end()` and use new `microqr_data` table to simplify code MICROPDF417: use table for max codewords per column library: centralize all error messages using new `errtxt()`, `errtxtf()`, `errtxt_adj()` funcs that protect `symbol->errtxt` from overflow, & try to make error messages more consistent thru-out, adding more feedback info to many, & use positional args "%n$" in prep for l10n (maybe); `is_sane/is_sane_lookup()` -> `not_sane/not_sane_lookup()`, returning 1-based position (zero on failure) instead of bool; `long` ints -> plain `int` (except those dealing with `ftell()`, `fread()` etc) as depend on int being 32-bits already GUI: in "grpDATF.ui" use "PlainText" rather than "RichText" for tracker ratio examples as height of text messing up sometimes manual: clarify Codablock-F length maximum & add examples docs: README: pandoc 3.5, Ubuntu 24.04 CMake: use "-Wpedantic" for Clang only as GNU complains about `errtxtf()` positional args "%n$"
2024-10-28 10:33:33 +13:00
static const struct item data[] = {
/* 0*/ { BARCODE_CODABAR, -1, "A37859B", 0, 1, 72, "BS EN 798:1995 Figure 1",
"101100100101100101010100101101010011010101101010010110100101010010010110"
},
/* 1*/ { BARCODE_CODABAR, -1, "A0123456789-$:/.+D", 0, 1, 186, "Verified manually against tec-it",
"101100100101010100110101011001010100101101100101010101101001011010100101001010110100101101010011010101101001010101001101010110010101101011011011011010110110110110101011011011010100110010"
},
/* 2*/ { BARCODE_CODABAR, 1, "A1B", 0, 1, 43, "Verified manually against tec-it",
"1011001001010101100101101101101010010010110"
},
/* 3*/ { BARCODE_CODABAR, 1, "A+B", 0, 1, 43, "Verified manually against tec-it",
"1011001001010110110110101010011010010010110"
},
/* 4*/ { BARCODE_CODABAR, 1, "B0123456789-$:/.+B", 0, 1, 196, "Verified manually against tec-it",
"1001001011010101001101010110010101001011011001010101011010010110101001010010101101001011010100110101011010010101010011010101100101011010110110110110101101101101101010110110110100101011010010010110"
},
/* 5*/ { BARCODE_PHARMA, -1, "131070", 0, 1, 78, "",
"111001110011100111001110011100111001110011100111001110011100111001110011100111"
},
/* 6*/ { BARCODE_PHARMA, -1, "123456", 0, 1, 58, "",
"1110011100111001001001001110010010011100100100100100100111"
},
/* 7*/ { BARCODE_PHARMA_TWO, -1, "64570080", 0, 2, 31, "Verified manually against tec-it",
"1010101010101010101010101010101"
"1010101010101010101010101010101"
},
/* 8*/ { BARCODE_PHARMA_TWO, -1, "29876543", 0, 2, 31, "Verified manually against tec-it",
"0010100010001010001010001000101"
"1000101010100000100000101010000"
},
/* 9*/ { BARCODE_CODE32, -1, "34567890", 0, 1, 103, "Verified manually against tec-it",
"1001011011010101101001011010110010110101011011010010101100101101011010010101101010101100110100101101101"
},
};
CODE128: reduce extended latch cut-off from 5 to 4 for better encodation in certain cases (and no pessimizations found so far), props lyngklip (BWIPP); fix extended char latching when exactly 3 extended chars at end; count code set C (not digits) in loop deciding when to shift/latch to extended for better estimate AZTEC: return warning if ECC < 5% (due to bit-stuffing when version given); return error if > 22 layers (Zint 26) for Reader Initialisation symbol requested for better error message AZTEC/HANXIN/QRCODE: consolidate different ECC data size tables into one indexed by ECC DBAR_EXP: check for reduced length <= 77 up front for better error message HANXIN: use `malloc()` rather than `z_alloca()` for large binary array QRCODE: `ecc_level` now 0-based (not 1-based) MICROQR: consolidate different version end routines into one `microqr_end()` and use new `microqr_data` table to simplify code MICROPDF417: use table for max codewords per column library: centralize all error messages using new `errtxt()`, `errtxtf()`, `errtxt_adj()` funcs that protect `symbol->errtxt` from overflow, & try to make error messages more consistent thru-out, adding more feedback info to many, & use positional args "%n$" in prep for l10n (maybe); `is_sane/is_sane_lookup()` -> `not_sane/not_sane_lookup()`, returning 1-based position (zero on failure) instead of bool; `long` ints -> plain `int` (except those dealing with `ftell()`, `fread()` etc) as depend on int being 32-bits already GUI: in "grpDATF.ui" use "PlainText" rather than "RichText" for tracker ratio examples as height of text messing up sometimes manual: clarify Codablock-F length maximum & add examples docs: README: pandoc 3.5, Ubuntu 24.04 CMake: use "-Wpedantic" for Clang only as GNU complains about `errtxtf()` positional args "%n$"
2024-10-28 10:33:33 +13:00
const int data_size = ARRAY_SIZE(data);
int i, length, ret;
struct zint_symbol *symbol = NULL;
char escaped[1024];
char cmp_buf[8192];
char cmp_msg[1024];
int do_bwipp = (debug & ZINT_DEBUG_TEST_BWIPP) && testUtilHaveGhostscript(); /* Only do BWIPP test if asked, too slow otherwise */
int do_zxingcpp = (debug & ZINT_DEBUG_TEST_ZXINGCPP) && testUtilHaveZXingCPPDecoder(); /* Only do ZXing-C++ test if asked, too slow otherwise */
testStartSymbol("test_encode", &symbol);
for (i = 0; i < data_size; i++) {
if (testContinue(p_ctx, i)) continue;
symbol = ZBarcode_Create();
assert_nonnull(symbol, "Symbol not created\n");
length = testUtilSetSymbol(symbol, data[i].symbology, -1 /*input_mode*/, -1 /*eci*/, -1 /*option_1*/, data[i].option_2, -1, -1 /*output_options*/, data[i].data, -1, debug);
ret = ZBarcode_Encode(symbol, (unsigned char *) data[i].data, length);
assert_equal(ret, data[i].ret, "i:%d ZBarcode_Encode ret %d != %d (%s)\n", i, ret, data[i].ret, symbol->errtxt);
if (p_ctx->generate) {
printf(" /*%3d*/ { %s, %d, \"%s\", %s, %d, %d, \"%s\",\n",
i, testUtilBarcodeName(data[i].symbology), data[i].option_2, testUtilEscape(data[i].data, length, escaped, sizeof(escaped)),
testUtilErrorName(data[i].ret), symbol->rows, symbol->width, data[i].comment);
testUtilModulesPrint(symbol, " ", "\n");
printf(" },\n");
} else {
if (ret < ZINT_ERROR) {
int width, row;
assert_equal(symbol->rows, data[i].expected_rows, "i:%d symbol->rows %d != %d (%s)\n", i, symbol->rows, data[i].expected_rows, data[i].data);
assert_equal(symbol->width, data[i].expected_width, "i:%d symbol->width %d != %d (%s)\n", i, symbol->width, data[i].expected_width, data[i].data);
ret = testUtilModulesCmp(symbol, data[i].expected, &width, &row);
assert_zero(ret, "i:%d testUtilModulesCmp ret %d != 0 width %d row %d (%s)\n", i, ret, width, row, data[i].data);
if (do_bwipp && testUtilCanBwipp(i, symbol, -1, data[i].option_2, -1, debug)) {
ret = testUtilBwipp(i, symbol, -1, data[i].option_2, -1, data[i].data, length, NULL, cmp_buf, sizeof(cmp_buf), NULL);
assert_zero(ret, "i:%d %s testUtilBwipp ret %d != 0\n", i, testUtilBarcodeName(symbol->symbology), ret);
ret = testUtilBwippCmp(symbol, cmp_msg, cmp_buf, data[i].expected);
assert_zero(ret, "i:%d %s testUtilBwippCmp %d != 0 %s\n actual: %s\nexpected: %s\n",
i, testUtilBarcodeName(symbol->symbology), ret, cmp_msg, cmp_buf, data[i].expected);
}
if (do_zxingcpp && testUtilCanZXingCPP(i, symbol, data[i].data, length, debug)) {
int cmp_len, ret_len;
char modules_dump[8192 + 1];
assert_notequal(testUtilModulesDump(symbol, modules_dump, sizeof(modules_dump)), -1, "i:%d testUtilModulesDump == -1\n", i);
ret = testUtilZXingCPP(i, symbol, data[i].data, length, modules_dump, cmp_buf, sizeof(cmp_buf), &cmp_len);
assert_zero(ret, "i:%d %s testUtilZXingCPP ret %d != 0\n", i, testUtilBarcodeName(symbol->symbology), ret);
ret = testUtilZXingCPPCmp(symbol, cmp_msg, cmp_buf, cmp_len, data[i].data, length, NULL /*primary*/, escaped, &ret_len);
assert_zero(ret, "i:%d %s testUtilZXingCPPCmp %d != 0 %s\n actual: %.*s\nexpected: %.*s\n",
i, testUtilBarcodeName(symbol->symbology), ret, cmp_msg, cmp_len, cmp_buf, ret_len, escaped);
}
}
}
ZBarcode_Delete(symbol);
}
testFinish();
}
int main(int argc, char *argv[]) {
testFunction funcs[] = { /* name, func */
{ "test_large", test_large },
{ "test_hrt", test_hrt },
{ "test_input", test_input },
{ "test_encode", test_encode },
};
testRun(argc, argv, funcs, ARRAY_SIZE(funcs));
testReport();
return 0;
}
/* vim: set ts=4 sw=4 et : */