re-import sources as agreed with the maintainer

This commit is contained in:
Adam Samalik 2023-07-10 08:45:03 +02:00
parent 5bd3de2283
commit 1c7fdeef52
4 changed files with 241 additions and 1 deletions

2
.gitignore vendored
View File

@ -1 +1 @@
/zaf-0-0.1.20080714svn.tar.bz2
zaf-0-0.1.20080714svn.tar.bz2

View File

@ -0,0 +1,15 @@
#!/bin/bash
set -e
gcc -Wall test-hyphen.c -o test-hyphen -lhyphen
echo "Test to hyphenate given word => woordafbreking"
echo "woordafbreking" | ./test-hyphen /usr/share/hyphen/hyph_af_ZA.dic /dev/stdin
echo ""
echo ""
echo "Test to give all possible ways to hyphenate the given word => woordafbreking"
echo ""woordafbreking | ./test-hyphen -d /usr/share/hyphen/hyph_af_ZA.dic /dev/stdin

212
tests/scripts/test-hyphen.c Normal file
View File

@ -0,0 +1,212 @@
/* This test file is part of upstream hyphen project */
#include <string.h>
#include <stdlib.h>
#include <stdio.h>
#include <ctype.h>
#include "hyphen.h"
#define BUFSIZE 1000
void help() {
fprintf(stderr,"correct syntax is:\n");
fprintf(stderr,"example [-d | -dd] hyphen_dictionary_file file_of_words_to_check\n");
fprintf(stderr,"-o = use old algorithm (without non-standard hyphenation)\n");
fprintf(stderr,"-d = hyphenation with listing of the possible hyphenations\n");
fprintf(stderr,"-n = print hyphenation vector\n");
}
/* get a pointer to the nth 8-bit or UTF-8 character of the word */
char * hindex(char * word, int n, int utf8) {
int j = 0;
while (j < n) {
j++;
word++;
while (utf8 && ((((unsigned char) *word) >> 6) == 2)) word++;
}
return word;
}
/* list possible hyphenations with -dd option (example for the usage of the hyphenate2() function) */
void single_hyphenations(char * word, char * hyphen, char ** rep, int * pos, int * cut, int utf8) {
int i, k, j = 0;
char r;
for (i = 0; (i + 1) < strlen(word); i++) {
if (utf8 && ((((unsigned char) word[i]) >> 6) == 2)) continue;
if ((hyphen[j] & 1)) {
if (rep && rep[j]) {
k = hindex(word, j - pos[j] + 1, utf8) - word;
r = word[k];
word[k] = 0;
printf(" - %s%s", word, rep[j]);
word[k] = r;
printf("%s\n", hindex(word + k, cut[j], utf8));
} else {
k = hindex(word, j + 1, utf8) - word;
r = word[k];
word[k] = 0;
printf(" - %s=", word);
word[k] = r;
printf("%s\n", word + k);
}
}
j++;
}
}
int
main(int argc, char** argv)
{
HyphenDict *dict;
int df;
int wtc;
FILE* wtclst;
int k, n, i, j, c;
char buf[BUFSIZE + 1];
int nHyphCount;
char *hyphens;
char *lcword;
char *hyphword;
char hword[BUFSIZE * 2];
int arg = 1;
int optd = 1;
int optn = 0;
int optdd = 0;
char ** rep;
int * pos;
int * cut;
/* first parse the command line options */
/* arg1 - hyphen dictionary file, arg2 - file of words to check */
if (argv[arg]) {
if (strcmp(argv[arg], "-o") == 0) {
optd = 0;
arg++;
}
if (strcmp(argv[arg], "-n") == 0) {
optn = 1;
arg++;
}
if (argv[arg] && strcmp(argv[arg], "-d") == 0) {
optd = 1;
optdd = 1;
arg++;
}
}
if (argv[arg]) {
df = arg++;
} else {
help();
exit(1);
}
if (argv[arg]) {
wtc = arg++;
} else {
help();
exit(1);
}
/* load the hyphenation dictionary */
if ((dict = hnj_hyphen_load(argv[df])) == NULL) {
fprintf(stderr, "Couldn't find file %s\n", argv[df]);
fflush(stderr);
exit(1);
}
/* open the words to check list */
wtclst = fopen(argv[wtc],"r");
if (!wtclst) {
fprintf(stderr,"Error - could not open file of words to check\n");
exit(1);
}
/* now read each word from the wtc file */
while(fgets(buf,BUFSIZE,wtclst) != NULL) {
k = strlen(buf);
if (k && buf[k - 1] == '\n') buf[k - 1] = '\0';
if (k >=2 && buf[k - 2] == '\r') buf[k-- - 2] = '\0';
/* set aside some buffers to hold lower cased */
/* and hyphen information */
lcword = (char *) malloc(k+1);
hyphens = (char *)malloc(k+5);
/* basic ascii lower-case, not suitable for real-world usage*/
for (i = 0; i < k; ++i) {
lcword[i] = buf[i];
if ( (lcword[i] >= 'A') && (lcword[i] <= 'Z') )
lcword[i] += 32;
}
/* first remove any trailing periods */
n = k-1;
while((n >=0) && (lcword[n] == '.')) n--;
n++;
/* now actually try to hyphenate the word */
rep = NULL;
pos = NULL;
cut = NULL;
hword[0] = '\0';
if ((!optd && hnj_hyphen_hyphenate(dict, lcword, n-1, hyphens)) ||
(optd && hnj_hyphen_hyphenate2(dict, lcword, n-1, hyphens, hword, &rep, &pos, &cut))) {
free(hyphens);
free(lcword);
fprintf(stderr, "hyphenation error\n");
exit(1);
}
if (optn) fprintf(stderr, "%s\n", hyphens);
if (!optd) {
/* now backfill hyphens[] for any removed periods */
for (c = n; c < k; c++) hyphens[c] = '0';
hyphens[k] = '\0';
/* now create a new char string showing hyphenation positions */
/* count the hyphens and allocate space for the new hypehanted string */
nHyphCount = 0;
for (i = 0; i < n; i++)
if (hyphens[i]&1)
nHyphCount++;
hyphword = (char *) malloc(k+1+nHyphCount);
j = 0;
for (i = 0; i < n; i++) {
hyphword[j++] = buf[i];
if (hyphens[i]&1) {
hyphword[j++] = '-';
}
}
hyphword[j] = '\0';
fprintf(stdout,"%s\n",hyphword);
fflush(stdout);
free(hyphword);
} else {
fprintf(stdout,"%s\n", hword);
if (optdd) single_hyphenations(lcword, hyphens, rep, pos, cut, dict->utf8);
if (rep) {
for (i = 0; i < n - 1; i++) {
if (rep[i]) free(rep[i]);
}
free(rep);
free(pos);
free(cut);
}
}
free(hyphens);
free(lcword);
}
fclose(wtclst);
hnj_hyphen_free(dict);
return 0;
}

13
tests/tests.yml Normal file
View File

@ -0,0 +1,13 @@
- hosts: localhost
roles:
- role: standard-test-basic
required_packages:
- hyphen-devel
- gcc
tags:
- classic
tests:
- sample:
dir: scripts/
run: ./run_tests.sh