aboutsummaryrefslogtreecommitdiffstats
path: root/utilities/vpl2mod.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'utilities/vpl2mod.cpp')
-rw-r--r--utilities/vpl2mod.cpp254
1 files changed, 254 insertions, 0 deletions
diff --git a/utilities/vpl2mod.cpp b/utilities/vpl2mod.cpp
new file mode 100644
index 0000000..cddc923
--- /dev/null
+++ b/utilities/vpl2mod.cpp
@@ -0,0 +1,254 @@
+#include <ctype.h>
+#include <stdio.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <stdlib.h>
+
+#ifndef __GNUC__
+#include <io.h>
+#else
+#include <unistd.h>
+#endif
+
+#include <swmgr.h>
+#include <rawtext.h>
+#include <iostream>
+#include <string>
+
+#ifndef O_BINARY
+#define O_BINARY 0
+#endif
+
+char readline(int fd, char **buf) {
+ char ch;
+ if (*buf)
+ delete [] *buf;
+ *buf = 0;
+ int len;
+
+
+ long index = lseek(fd, 0, SEEK_CUR);
+ // clean up any preceding white space
+ while ((len = read(fd, &ch, 1)) == 1) {
+ if ((ch != 13) && (ch != ' ') && (ch != '\t'))
+ break;
+ else index++;
+ }
+
+
+ while (ch != 10) {
+ if ((len = read(fd, &ch, 1)) != 1)
+ break;
+ }
+
+ int size = (lseek(fd, 0, SEEK_CUR) - index) - 1;
+
+ *buf = new char [ size + 1 ];
+
+ if (size > 0) {
+ lseek(fd, index, SEEK_SET);
+ read(fd, *buf, size);
+ read(fd, &ch, 1); //pop terminating char
+ (*buf)[size] = 0;
+
+ // clean up any trailing junk on buf
+ for (char *it = *buf+(strlen(*buf)-1); it > *buf; it--) {
+ if ((*it != 10) && (*it != 13) && (*it != ' ') && (*it != '\t'))
+ break;
+ else *it = 0;
+ }
+ }
+ else **buf = 0;
+ return !len;
+}
+
+
+char *parseVReg(char *buf) {
+ char stage = 0;
+
+ while (*buf) {
+ switch (stage) {
+ case 0:
+ if (isalpha(*buf))
+ stage++;
+ break;
+ case 1:
+ if (isdigit(*buf))
+ stage++;
+ break;
+ case 2:
+ if (*buf == ':')
+ stage++;
+ break;
+ case 3:
+ if (isdigit(*buf))
+ stage++;
+ break;
+ case 4:
+ if (*buf == ' ') {
+ *buf = 0;
+ return ++buf;
+ }
+ break;
+ }
+ buf++;
+ }
+ return (stage == 4) ? buf : 0; // if we got to stage 4 return after key buf, else return 0;
+}
+
+
+bool isKJVRef(const char *buf) {
+ VerseKey vk, test;
+ vk.AutoNormalize(0);
+ vk.Headings(1); // turn on mod/testmnt/book/chap headings
+ vk.Persist(1);
+ // lets do some tests on the verse --------------
+ vk = buf;
+ test = buf;
+
+ if (vk.Testament() && vk.Book() && vk.Chapter() && vk.Verse()) { // if we're not a heading
+// std::cerr << (const char*)vk << " == " << (const char*)test << std::endl;
+ return (vk == test);
+ }
+ else return true; // no check if we're a heading... Probably bad.
+}
+
+
+void fixText(char *text) {
+ char *to = text;
+ while(*text) {
+ *to++ = *text++;
+ *to++ = *text++;
+ if (!*text)
+ break;
+ if (*text != ' ')
+ std::cerr << "problem\n";
+ else text++;
+ }
+ *to = 0;
+}
+
+int main(int argc, char **argv) {
+
+ // Let's test our command line arguments
+ if (argc < 2) {
+// fprintf(stderr, "usage: %s <vpl_file> </path/to/mod> [0|1 - file includes prepended verse references]\n", argv[0]);
+ fprintf(stderr, "usage: %s <source_vpl_file> </path/to/output/mod/> [0|1 - prepended verse refs] [0|1 - NT only]\n\n", argv[0]);
+ fprintf(stderr, "\tWith no verse refs, source file must contain exactly 31102 lines.\n");
+ fprintf(stderr, "\tThis is KJV verse count plus headings for MODULE,\n");
+ fprintf(stderr, "\tTESTAMENT, BOOK, CHAPTER. An example snippet follows:\n\n");
+ fprintf(stderr, "\t\tMODULE HEADER\n");
+ fprintf(stderr, "\t\tOLD TESTAMENT HEADER\n");
+ fprintf(stderr, "\t\tGENESIS HEADER\n");
+ fprintf(stderr, "\t\tCHAPTER 1 HEADER\n");
+ fprintf(stderr, "\t\tIn the beginning...\n\n");
+ fprintf(stderr, "\t... implying there must also be a CHAPTER2 HEADER,\n");
+ fprintf(stderr, "\tEXODUS HEADER, NEW TESTAMENT HEADER, etc. If there is no text for\n");
+ fprintf(stderr, "\tthe header, a blank line must, at least, hold place.\n\n");
+ fprintf(stderr, "\tWith verse refs, source file must simply contain any number of lines,\n");
+ fprintf(stderr, "\tthat begin with the verse reference for which it is an entry. e.g.:\n\n");
+ fprintf(stderr, "\t\tgen 1:0 CHAPTER 1 HEADER\n");
+ fprintf(stderr, "\t\tgen 1:1 In the beginning...\n\n");
+ exit(-1);
+ }
+
+ // Let's see if we can open our input file
+ int fd = open(argv[1], O_RDONLY|O_BINARY);
+ if (fd < 0) {
+ fprintf(stderr, "error: %s: couldn't open input file: %s \n", argv[0], argv[1]);
+ exit(-2);
+ }
+
+ // Try to initialize a default set of datafiles and indicies at our
+ // datapath location passed to us from the user.
+ if (RawText::createModule(argv[2])) {
+ fprintf(stderr, "error: %s: couldn't create module at path: %s \n", argv[0], argv[2]);
+ exit(-3);
+ }
+
+ // not used yet, but for future support of a vpl file with each line
+ // prepended with verse reference, eg. "Gen 1:1 In the beginning..."
+ bool vref = false;
+ if (argc > 3)
+ vref = (argv[3][0] == '0') ? false : true;
+
+ // if 'nt' is the 4th arg, our vpl file only has the NT
+ bool ntonly = false;
+ if (argc > 4)
+ ntonly = (argv[4][0] == '0') ? false : true;
+
+ // Do some initialization stuff
+ char *buffer = 0;
+ RawText mod(argv[2]); // open our datapath with our RawText driver.
+ VerseKey vk;
+ vk.AutoNormalize(0);
+ vk.Headings(1); // turn on mod/testmnt/book/chap headings
+ vk.Persist(1);
+
+ mod.SetKey(vk);
+
+ // Loop through module from TOP to BOTTOM and set next line from
+ // input file as text for this entry in the module
+ mod = TOP;
+ if (ntonly) vk = "Matthew 1:1";
+
+ int successive = 0; //part of hack below
+ while ((!mod.Error()) && (!readline(fd, &buffer))) {
+ if (*buffer == '|') // comments, ignore line
+ continue;
+ if (vref) {
+ const char *verseText = parseVReg(buffer);
+ if (!verseText) { // if we didn't find a valid verse ref
+ std::cerr << "No valid verse ref found on line: " << buffer << "\n";
+ exit(-4);
+ }
+
+ vk = buffer;
+ if (vk.Error()) {
+ std::cerr << "Error parsing key: " << buffer << "\n";
+ exit(-5);
+ }
+ string orig = mod.getRawEntry();
+
+ if (!isKJVRef(buffer)) {
+ VerseKey origVK = vk;
+ /* This block is functioning improperly -- problem with AutoNormalize???
+ do {
+ vk--;
+ }
+ while (!vk.Error() && !isKJVRef(vk)); */
+ //hack to replace above:
+ successive++;
+ vk -= successive;
+ orig = mod.getRawEntry();
+
+ std::cerr << "Not a valid KJV ref: " << origVK << "\n";
+ std::cerr << "appending to ref: " << vk << "\n";
+ orig += " [ (";
+ orig += origVK;
+ orig += ") ";
+ orig += verseText;
+ orig += " ] ";
+ verseText = orig.c_str();
+ }
+ else {
+ successive = 0;
+ }
+
+ if (orig.length() > 1)
+ std::cerr << "Warning, overwriting verse: " << vk << std::endl;
+
+ // ------------- End verse tests -----------------
+ mod << verseText; // save text to module at current position
+ }
+ else {
+ fixText(buffer);
+ mod << buffer; // save text to module at current position
+ mod++; // increment module position
+ }
+ }
+
+ // clear up our buffer that readline might have allocated
+ if (buffer)
+ delete [] buffer;
+}