mirror of
https://github.com/holub/mame
synced 2025-04-30 03:47:13 +03:00
906 lines
26 KiB
C++
906 lines
26 KiB
C++
// license:BSD-3-Clause
|
|
// copyright-holders:Aaron Giles
|
|
/***************************************************************************
|
|
|
|
MAME source code to HTML converter
|
|
|
|
****************************************************************************
|
|
|
|
Template file format:
|
|
|
|
<html header>
|
|
<!--PATH--> = insert path
|
|
<!--CONTENT--> = insert content
|
|
<html footer>
|
|
|
|
***************************************************************************/
|
|
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <string.h>
|
|
#include <ctype.h>
|
|
#include <assert.h>
|
|
#include "osdcore.h"
|
|
#include "corestr.h"
|
|
#include "corefile.h"
|
|
|
|
|
|
/***************************************************************************
|
|
CONSTANTS & DEFINES
|
|
***************************************************************************/
|
|
|
|
#define PREPROC_CLASS "preproc"
|
|
#define KEYWORD_CLASS "keyword"
|
|
#define CUSTOM_CLASS "custom"
|
|
|
|
|
|
|
|
/***************************************************************************
|
|
TYPE DEFINITIONS
|
|
***************************************************************************/
|
|
|
|
enum file_type
|
|
{
|
|
FILE_TYPE_INVALID,
|
|
FILE_TYPE_C,
|
|
FILE_TYPE_MAKE,
|
|
FILE_TYPE_XML,
|
|
FILE_TYPE_TEXT
|
|
};
|
|
|
|
|
|
struct ext_to_type
|
|
{
|
|
const char * extension;
|
|
file_type type;
|
|
};
|
|
|
|
|
|
struct token_entry
|
|
{
|
|
const char * token;
|
|
const char * color;
|
|
};
|
|
|
|
|
|
struct include_path
|
|
{
|
|
include_path * next;
|
|
std::string path;
|
|
};
|
|
|
|
|
|
struct list_entry
|
|
{
|
|
list_entry * next;
|
|
std::string name;
|
|
};
|
|
|
|
|
|
|
|
/***************************************************************************
|
|
GLOBAL VARIABLES
|
|
***************************************************************************/
|
|
|
|
static include_path *incpaths;
|
|
|
|
static const ext_to_type extension_lookup[] =
|
|
{
|
|
{ ".c", FILE_TYPE_C },
|
|
{ ".h", FILE_TYPE_C },
|
|
{ ".lst", FILE_TYPE_C },
|
|
{ ".mak", FILE_TYPE_MAKE },
|
|
{ "makefile", FILE_TYPE_MAKE },
|
|
{ ".lay", FILE_TYPE_XML },
|
|
{ ".xml", FILE_TYPE_XML },
|
|
{ ".txt", FILE_TYPE_TEXT },
|
|
};
|
|
|
|
|
|
static const token_entry dummy_token_table[] =
|
|
{
|
|
{ nullptr, KEYWORD_CLASS }
|
|
};
|
|
|
|
|
|
static const token_entry c_token_table[] =
|
|
{
|
|
{ "#define", PREPROC_CLASS },
|
|
{ "#elif", PREPROC_CLASS },
|
|
{ "#else", PREPROC_CLASS },
|
|
{ "#endif", PREPROC_CLASS },
|
|
{ "#error", PREPROC_CLASS },
|
|
{ "#if", PREPROC_CLASS },
|
|
{ "#ifdef", PREPROC_CLASS },
|
|
{ "#ifndef", PREPROC_CLASS },
|
|
{ "#include", PREPROC_CLASS },
|
|
{ "#line", PREPROC_CLASS },
|
|
{ "#pragma", PREPROC_CLASS },
|
|
{ "#undef", PREPROC_CLASS },
|
|
|
|
{ "auto", KEYWORD_CLASS },
|
|
{ "bool", KEYWORD_CLASS },
|
|
{ "break", KEYWORD_CLASS },
|
|
{ "case", KEYWORD_CLASS },
|
|
{ "catch", KEYWORD_CLASS },
|
|
{ "char", KEYWORD_CLASS },
|
|
{ "class", KEYWORD_CLASS },
|
|
{ "const", KEYWORD_CLASS },
|
|
{ "const_cast", KEYWORD_CLASS },
|
|
{ "continue", KEYWORD_CLASS },
|
|
{ "default", KEYWORD_CLASS },
|
|
{ "delete", KEYWORD_CLASS },
|
|
{ "do", KEYWORD_CLASS },
|
|
{ "double", KEYWORD_CLASS },
|
|
{ "dynamic_cast", KEYWORD_CLASS },
|
|
{ "else", KEYWORD_CLASS },
|
|
{ "enum", KEYWORD_CLASS },
|
|
{ "explicit", KEYWORD_CLASS },
|
|
{ "export", KEYWORD_CLASS },
|
|
{ "extern", KEYWORD_CLASS },
|
|
{ "false", KEYWORD_CLASS },
|
|
{ "float", KEYWORD_CLASS },
|
|
{ "for", KEYWORD_CLASS },
|
|
{ "friend", KEYWORD_CLASS },
|
|
{ "goto", KEYWORD_CLASS },
|
|
{ "if", KEYWORD_CLASS },
|
|
{ "inline", KEYWORD_CLASS },
|
|
{ "int", KEYWORD_CLASS },
|
|
{ "long", KEYWORD_CLASS },
|
|
{ "mutable", KEYWORD_CLASS },
|
|
{ "namespace", KEYWORD_CLASS },
|
|
{ "new", KEYWORD_CLASS },
|
|
{ "operator", KEYWORD_CLASS },
|
|
{ "private", KEYWORD_CLASS },
|
|
{ "protected", KEYWORD_CLASS },
|
|
{ "public", KEYWORD_CLASS },
|
|
{ "register", KEYWORD_CLASS },
|
|
{ "reinterpret_cast", KEYWORD_CLASS },
|
|
{ "return", KEYWORD_CLASS },
|
|
{ "short", KEYWORD_CLASS },
|
|
{ "signed", KEYWORD_CLASS },
|
|
{ "sizeof", KEYWORD_CLASS },
|
|
{ "static", KEYWORD_CLASS },
|
|
{ "static_cast", KEYWORD_CLASS },
|
|
{ "struct", KEYWORD_CLASS },
|
|
{ "switch", KEYWORD_CLASS },
|
|
{ "template", KEYWORD_CLASS },
|
|
{ "this", KEYWORD_CLASS },
|
|
{ "throw", KEYWORD_CLASS },
|
|
{ "true", KEYWORD_CLASS },
|
|
{ "try", KEYWORD_CLASS },
|
|
{ "typedef", KEYWORD_CLASS },
|
|
{ "typeid", KEYWORD_CLASS },
|
|
{ "typename", KEYWORD_CLASS },
|
|
{ "union", KEYWORD_CLASS },
|
|
{ "unsigned", KEYWORD_CLASS },
|
|
{ "virtual", KEYWORD_CLASS },
|
|
{ "void", KEYWORD_CLASS },
|
|
{ "volatile", KEYWORD_CLASS },
|
|
{ "wchar_t", KEYWORD_CLASS },
|
|
{ "while", KEYWORD_CLASS },
|
|
|
|
/*
|
|
{ "INLINE", CUSTOM_CLASS },
|
|
{ "int8_t", CUSTOM_CLASS },
|
|
{ "uint8_t", CUSTOM_CLASS },
|
|
{ "int16_t", CUSTOM_CLASS },
|
|
{ "uint16_t", CUSTOM_CLASS },
|
|
{ "int32_t", CUSTOM_CLASS },
|
|
{ "uint32_t", CUSTOM_CLASS },
|
|
{ "int64_t", CUSTOM_CLASS },
|
|
{ "uint64_t", CUSTOM_CLASS },
|
|
{ "ARRAY_LENGTH", CUSTOM_CLASS },
|
|
*/
|
|
{ nullptr, KEYWORD_CLASS }
|
|
};
|
|
|
|
|
|
|
|
/***************************************************************************
|
|
PROTOTYPES
|
|
***************************************************************************/
|
|
|
|
// core output functions
|
|
static int recurse_dir(int srcrootlen, int dstrootlen, std::string &srcdir, std::string &dstdir, std::string &tempheader, std::string &tempfooter);
|
|
static int output_file(file_type type, int srcrootlen, int dstrootlen, std::string &srcfile, std::string &dstfile, bool link_to_file, std::string &tempheader, std::string &tempfooter);
|
|
|
|
// HTML helpers
|
|
static util::core_file::ptr create_file_and_output_header(std::string &filename, std::string &templatefile, std::string &path);
|
|
static void output_footer_and_close_file(util::core_file::ptr &&file, std::string &templatefile, std::string &path);
|
|
|
|
// path helpers
|
|
static std::string &normalized_subpath(std::string &dest, std::string &path, int start);
|
|
static void output_path_as_links(util::core_file &file, std::string &path, bool end_is_directory, bool link_to_file);
|
|
static bool find_include_file(std::string &srcincpath, int srcrootlen, int dstrootlen, std::string &srcfile, std::string &dstfile, std::string &filename);
|
|
|
|
|
|
|
|
/***************************************************************************
|
|
MAIN
|
|
***************************************************************************/
|
|
|
|
/*-------------------------------------------------
|
|
main - main entry point
|
|
-------------------------------------------------*/
|
|
|
|
void usage(const char *argv0)
|
|
{
|
|
fprintf(stderr, "Usage:\n%s <srcroot> <destroot> <template.html> [-Iincpath [-Iincpath [...]]]\n", argv0);
|
|
exit(1);
|
|
}
|
|
|
|
int main(int argc, char *argv[])
|
|
{
|
|
// loop over arguments
|
|
include_path **incpathhead = &incpaths;
|
|
std::string srcdir, dstdir, tempfilename, tempheader, tempfooter;
|
|
int unadorned = 0;
|
|
for (int argnum = 1; argnum < argc; argnum++)
|
|
{
|
|
char *arg = argv[argnum];
|
|
|
|
// include path?
|
|
if (arg[0] == '-' && arg[1] == 'I')
|
|
{
|
|
*incpathhead = new include_path;
|
|
if (*incpathhead != nullptr)
|
|
{
|
|
(*incpathhead)->next = nullptr;
|
|
strreplacechr((*incpathhead)->path.assign(&arg[2]), '/', PATH_SEPARATOR[0]);
|
|
incpathhead = &(*incpathhead)->next;
|
|
}
|
|
}
|
|
|
|
// other parameter
|
|
else if (arg[0] != '-' && unadorned == 0)
|
|
{
|
|
srcdir.assign(arg);
|
|
unadorned++;
|
|
}
|
|
else if (arg[0] != '-' && unadorned == 1)
|
|
{
|
|
dstdir.assign(arg);
|
|
unadorned++;
|
|
}
|
|
else if (arg[0] != '-' && unadorned == 2)
|
|
{
|
|
tempfilename.assign(arg);
|
|
unadorned++;
|
|
}
|
|
else
|
|
usage(argv[0]);
|
|
}
|
|
|
|
// make sure we got 3 parameters
|
|
if (srcdir.length() == 0 || dstdir.length() == 0 || tempfilename.length() == 0)
|
|
usage(argv[0]);
|
|
|
|
// read the template file into an std::string
|
|
uint32_t bufsize;
|
|
void *buffer;
|
|
if (util::core_file::load(tempfilename.c_str(), &buffer, bufsize) == osd_file::error::NONE)
|
|
{
|
|
tempheader.assign((const char *)buffer, bufsize);
|
|
free(buffer);
|
|
}
|
|
|
|
// verify the template
|
|
if (tempheader.length() == 0)
|
|
{
|
|
fprintf(stderr, "Unable to read template file\n");
|
|
return 1;
|
|
}
|
|
int result = tempheader.find("<!--CONTENT-->");
|
|
if (result == -1)
|
|
{
|
|
fprintf(stderr, "Template is missing a <!--CONTENT--> marker\n");
|
|
return 1;
|
|
}
|
|
tempfooter.assign(tempheader).erase(0, result + 14);
|
|
tempheader.erase(result);
|
|
|
|
// recurse over subdirectories
|
|
return recurse_dir(srcdir.length(), dstdir.length(), srcdir, dstdir, tempheader, tempfooter);
|
|
}
|
|
|
|
|
|
|
|
/***************************************************************************
|
|
CORE OUTPUT FUNCTIONS
|
|
***************************************************************************/
|
|
|
|
static int compare_list_entries(const void *p1, const void *p2)
|
|
{
|
|
const list_entry *entry1 = *(const list_entry **)p1;
|
|
const list_entry *entry2 = *(const list_entry **)p2;
|
|
return entry1->name.compare(entry2->name);
|
|
}
|
|
|
|
|
|
/*-------------------------------------------------
|
|
recurse_dir - recurse through a directory
|
|
-------------------------------------------------*/
|
|
|
|
static int recurse_dir(int srcrootlen, int dstrootlen, std::string &srcdir, std::string &dstdir, std::string &tempheader, std::string &tempfooter)
|
|
{
|
|
static const osd::directory::entry::entry_type typelist[] = { osd::directory::entry::entry_type::DIR, osd::directory::entry::entry_type::FILE };
|
|
|
|
// extract a normalized subpath
|
|
std::string srcdir_subpath;
|
|
normalized_subpath(srcdir_subpath, srcdir, srcrootlen + 1);
|
|
|
|
// create an index file
|
|
std::string indexname;
|
|
indexname = string_format("%s%c%s", dstdir.c_str(), PATH_SEPARATOR[0], "index.html");
|
|
util::core_file::ptr indexfile = create_file_and_output_header(indexname, tempheader, srcdir_subpath);
|
|
|
|
// output the directory navigation
|
|
indexfile->printf("<h3>Viewing Directory: ");
|
|
output_path_as_links(*indexfile, srcdir_subpath, true, false);
|
|
indexfile->printf("</h3>");
|
|
|
|
// iterate first over directories, then over files
|
|
int result = 0;
|
|
for (int entindex = 0; entindex < ARRAY_LENGTH(typelist) && result == 0; entindex++)
|
|
{
|
|
auto entry_type = typelist[entindex];
|
|
|
|
// open the directory and iterate through it
|
|
auto dir = osd::directory::open(srcdir.c_str());
|
|
if (dir == nullptr)
|
|
{
|
|
result = 1;
|
|
break;
|
|
}
|
|
|
|
// build up the list of files
|
|
const osd::directory::entry *entry;
|
|
int found = 0;
|
|
list_entry *list = nullptr;
|
|
while ((entry = dir->read()) != nullptr)
|
|
if (entry->type == entry_type && entry->name[0] != '.')
|
|
{
|
|
auto lentry = new list_entry;
|
|
lentry->name.assign(entry->name);
|
|
lentry->next = list;
|
|
list = lentry;
|
|
found++;
|
|
}
|
|
|
|
// close the directory
|
|
dir.reset();
|
|
|
|
// skip if nothing found
|
|
if (found == 0)
|
|
continue;
|
|
|
|
// allocate memory for sorting
|
|
auto listarray = new list_entry *[found];
|
|
found = 0;
|
|
for (list_entry *curlist = list; curlist != nullptr; curlist = curlist->next)
|
|
listarray[found++] = curlist;
|
|
|
|
// sort the list
|
|
qsort(listarray, found, sizeof(listarray[0]), compare_list_entries);
|
|
|
|
// rebuild the list
|
|
list = nullptr;
|
|
while (--found >= 0)
|
|
{
|
|
listarray[found]->next = list;
|
|
list = listarray[found];
|
|
}
|
|
delete[] listarray;
|
|
|
|
// iterate through each file
|
|
for (list_entry *curlist = list; curlist != nullptr && result == 0; curlist = curlist->next)
|
|
{
|
|
// add a header
|
|
if (curlist == list)
|
|
indexfile->printf("\t<h2>%s</h2>\n\t<ul>\n", (entry_type == osd::directory::entry::entry_type::DIR) ? "Directories" : "Files");
|
|
|
|
// build the source filename
|
|
std::string srcfile;
|
|
srcfile = string_format("%s%c%s", srcdir.c_str(), PATH_SEPARATOR[0], curlist->name.c_str());
|
|
|
|
// if we have a file, output it
|
|
std::string dstfile;
|
|
if (entry_type == osd::directory::entry::entry_type::FILE)
|
|
{
|
|
// make sure we care, first
|
|
file_type type = FILE_TYPE_INVALID;
|
|
for (auto & elem : extension_lookup)
|
|
if (core_filename_ends_with(curlist->name, elem.extension))
|
|
{
|
|
type = elem.type;
|
|
break;
|
|
}
|
|
|
|
// if we got a valid file, process it
|
|
if (type != FILE_TYPE_INVALID)
|
|
{
|
|
dstfile = string_format("%s%c%s.html", dstdir.c_str(), PATH_SEPARATOR[0], curlist->name.c_str());
|
|
if (indexfile != nullptr)
|
|
indexfile->printf("\t<li><a href=\"%s.html\">%s</a></li>\n", curlist->name.c_str(), curlist->name.c_str());
|
|
result = output_file(type, srcrootlen, dstrootlen, srcfile, dstfile, srcdir.compare(dstdir) == 0, tempheader, tempfooter);
|
|
}
|
|
}
|
|
|
|
// if we have a directory, recurse
|
|
else
|
|
{
|
|
dstfile = string_format("%s%c%s", dstdir.c_str(), PATH_SEPARATOR[0], curlist->name.c_str());
|
|
if (indexfile != nullptr)
|
|
indexfile->printf("\t<li><a href=\"%s/index.html\">%s/</a></li>\n", curlist->name.c_str(), curlist->name.c_str());
|
|
result = recurse_dir(srcrootlen, dstrootlen, srcfile, dstfile, tempheader, tempfooter);
|
|
}
|
|
}
|
|
|
|
// close the list if we found some stuff
|
|
if (list != nullptr)
|
|
indexfile->printf("\t</ul>\n");
|
|
|
|
// free all the allocated entries
|
|
while (list != nullptr)
|
|
{
|
|
list_entry *next = list->next;
|
|
delete list;
|
|
list = next;
|
|
}
|
|
}
|
|
|
|
if (indexfile != nullptr)
|
|
output_footer_and_close_file(std::move(indexfile), tempfooter, srcdir_subpath);
|
|
return result;
|
|
}
|
|
|
|
|
|
/*-------------------------------------------------
|
|
output_file - output a file, converting to
|
|
HTML
|
|
-------------------------------------------------*/
|
|
|
|
static int output_file(file_type type, int srcrootlen, int dstrootlen, std::string &srcfile, std::string &dstfile, bool link_to_file, std::string &tempheader, std::string &tempfooter)
|
|
{
|
|
// extract a normalized subpath
|
|
std::string srcfile_subpath;
|
|
normalized_subpath(srcfile_subpath, srcfile, srcrootlen + 1);
|
|
|
|
fprintf(stderr, "Processing %s\n", srcfile_subpath.c_str());
|
|
|
|
// set some defaults
|
|
bool color_quotes = false;
|
|
const char *comment_start = "";
|
|
const char *comment_start_esc = "";
|
|
const char *comment_end = "";
|
|
const char *comment_end_esc = "";
|
|
const char *comment_inline = "";
|
|
const char *comment_inline_esc = "";
|
|
const char *token_chars = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789_#";
|
|
const token_entry *token_table = dummy_token_table;
|
|
|
|
// based on the file type, set the comment info
|
|
switch (type)
|
|
{
|
|
case FILE_TYPE_C:
|
|
color_quotes = true;
|
|
comment_start = comment_start_esc = "/*";
|
|
comment_end = comment_end_esc = "*/";
|
|
comment_inline = comment_inline_esc = "//";
|
|
token_table = c_token_table;
|
|
break;
|
|
|
|
case FILE_TYPE_MAKE:
|
|
color_quotes = true;
|
|
comment_inline = comment_inline_esc = "#";
|
|
break;
|
|
|
|
case FILE_TYPE_XML:
|
|
color_quotes = true;
|
|
comment_start = "<!--";
|
|
comment_start_esc = "<!--";
|
|
comment_end = "-->";
|
|
comment_end_esc = "-->";
|
|
break;
|
|
|
|
default:
|
|
case FILE_TYPE_TEXT:
|
|
break;
|
|
}
|
|
|
|
// make the token lookup table
|
|
bool is_token[256];
|
|
memset(is_token, 0, sizeof(is_token));
|
|
for (int toknum = 0; token_chars[toknum] != 0; toknum++)
|
|
is_token[(uint8_t)token_chars[toknum]] = true;
|
|
|
|
// open the source file
|
|
util::core_file::ptr src;
|
|
if (util::core_file::open(srcfile, OPEN_FLAG_READ, src) != osd_file::error::NONE)
|
|
{
|
|
fprintf(stderr, "Unable to read file '%s'\n", srcfile.c_str());
|
|
return 1;
|
|
}
|
|
|
|
// open the output file
|
|
util::core_file::ptr dst = create_file_and_output_header(dstfile, tempheader, srcfile_subpath);
|
|
if (dst == nullptr)
|
|
{
|
|
fprintf(stderr, "Unable to write file '%s'\n", dstfile.c_str());
|
|
return 1;
|
|
}
|
|
|
|
// output the directory navigation
|
|
dst->printf("<h3>Viewing File: ");
|
|
output_path_as_links(*dst, srcfile_subpath, false, link_to_file);
|
|
dst->printf("</h3>");
|
|
|
|
// start with some tags
|
|
dst->printf("\t<pre class=\"source\">\n");
|
|
|
|
// iterate over lines in the source file
|
|
int linenum = 1;
|
|
bool in_comment = false;
|
|
char srcline[4096];
|
|
std::ostringstream dstline;
|
|
while (src->gets(srcline, ARRAY_LENGTH(srcline)) != nullptr)
|
|
{
|
|
// start with the line number
|
|
dstline.str("");
|
|
util::stream_format(dstline, "<span class=\"linenum\">%5d</span> ", linenum++);
|
|
|
|
// iterate over characters in the source line
|
|
bool escape = false;
|
|
bool in_quotes = false;
|
|
bool in_inline_comment = false;
|
|
bool last_token_was_include = false;
|
|
bool last_was_token = false;
|
|
bool quotes_are_linked = false;
|
|
uint8_t curquote = 0;
|
|
int curcol = 0;
|
|
for (char *srcptr = srcline; *srcptr != 0; )
|
|
{
|
|
uint8_t ch = *srcptr++;
|
|
|
|
// track whether or not we are within an extended (C-style) comment
|
|
if (!in_quotes && !in_inline_comment)
|
|
{
|
|
if (!in_comment && ch == comment_start[0] && strncmp(srcptr - 1, comment_start, strlen(comment_start)) == 0)
|
|
{
|
|
util::stream_format(dstline, "<span class=\"comment\">%s", comment_start_esc);
|
|
curcol += strlen(comment_start);
|
|
srcptr += strlen(comment_start) - 1;
|
|
ch = 0;
|
|
in_comment = true;
|
|
}
|
|
else if (in_comment && ch == comment_end[0] && strncmp(srcptr - 1, comment_end, strlen(comment_end)) == 0)
|
|
{
|
|
util::stream_format(dstline, "%s</span>", comment_end_esc);
|
|
curcol += strlen(comment_end);
|
|
srcptr += strlen(comment_end) - 1;
|
|
ch = 0;
|
|
in_comment = false;
|
|
}
|
|
}
|
|
|
|
// track whether or not we are within an inline (C++-style) comment
|
|
if (!in_quotes && !in_comment && !in_inline_comment && ch == comment_inline[0] && strncmp(srcptr - 1, comment_inline, strlen(comment_inline)) == 0)
|
|
{
|
|
util::stream_format(dstline, "<span class=\"comment\">%s", comment_inline_esc);
|
|
curcol += strlen(comment_inline);
|
|
srcptr += strlen(comment_inline) - 1;
|
|
ch = 0;
|
|
in_inline_comment = true;
|
|
}
|
|
|
|
// if this is the start of a new token, see if we want to color it
|
|
if (!in_quotes && !in_comment && !in_inline_comment && !last_was_token && is_token[ch])
|
|
{
|
|
const token_entry *curtoken;
|
|
char *temp = srcptr;
|
|
int toklength;
|
|
|
|
// find the end of the token
|
|
while (*temp != 0 && is_token[(uint8_t)*temp])
|
|
temp++;
|
|
toklength = temp - (srcptr - 1);
|
|
|
|
// scan the token table
|
|
last_token_was_include = false;
|
|
for (curtoken = token_table; curtoken->token != nullptr; curtoken++)
|
|
if (strncmp(srcptr - 1, curtoken->token, toklength) == 0 && strlen(curtoken->token) == toklength)
|
|
{
|
|
util::stream_format(dstline, "<span class=\"%s\">%s</span>", curtoken->color, curtoken->token);
|
|
curcol += strlen(curtoken->token);
|
|
srcptr += strlen(curtoken->token) - 1;
|
|
ch = 0;
|
|
|
|
// look for include tokens specially
|
|
if (type == FILE_TYPE_C && strcmp(curtoken->token, "#include") == 0)
|
|
last_token_was_include = true;
|
|
break;
|
|
}
|
|
}
|
|
last_was_token = is_token[ch];
|
|
|
|
// if we hit a tab, expand it
|
|
if (ch == 0x09)
|
|
{
|
|
// compute how many spaces
|
|
int spaces = 4 - curcol % 4;
|
|
while (spaces--)
|
|
{
|
|
dstline.put(' ');
|
|
curcol++;
|
|
}
|
|
}
|
|
|
|
// otherwise, copy the source character
|
|
else if (ch != 0x0a && ch != 0x0d && ch != 0)
|
|
{
|
|
// track opening quotes
|
|
if (!in_comment && !in_inline_comment && !in_quotes && (ch == '"' || ch == '\''))
|
|
{
|
|
if (color_quotes)
|
|
util::stream_format(dstline, "<span class=\"string\">%c", ch);
|
|
else
|
|
dstline.put(ch);
|
|
in_quotes = true;
|
|
curquote = ch;
|
|
|
|
// handle includes
|
|
if (last_token_was_include)
|
|
{
|
|
char *endquote = strchr(srcptr, ch);
|
|
if (endquote != nullptr)
|
|
{
|
|
std::string filename(srcptr, endquote - srcptr);
|
|
std::string target;
|
|
if (find_include_file(target, srcrootlen, dstrootlen, srcfile, dstfile, filename))
|
|
{
|
|
util::stream_format(dstline, "<a href=\"%s\">", target.c_str());
|
|
quotes_are_linked = true;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// track closing quotes
|
|
else if (!in_comment && !in_inline_comment && in_quotes && (ch == curquote) && !escape)
|
|
{
|
|
if (quotes_are_linked)
|
|
dstline << "</a>";
|
|
if (color_quotes)
|
|
util::stream_format(dstline, "%c</span>", ch);
|
|
else
|
|
dstline.put(ch);
|
|
in_quotes = false;
|
|
curquote = 0;
|
|
quotes_are_linked = false;
|
|
}
|
|
|
|
// else just output the current character
|
|
else if (ch == '&')
|
|
dstline << "&";
|
|
else if (ch == '<')
|
|
dstline << "<";
|
|
else if (ch == '>')
|
|
dstline << ">";
|
|
else
|
|
dstline.put(ch);
|
|
curcol++;
|
|
}
|
|
|
|
// Update escape state
|
|
if (in_quotes)
|
|
escape = (ch == '\\' && type == FILE_TYPE_C) ? !escape : false;
|
|
}
|
|
|
|
// finish inline comments
|
|
if (in_inline_comment)
|
|
{
|
|
dstline << "</span>";
|
|
in_inline_comment = false;
|
|
}
|
|
|
|
// append a break and move on
|
|
dstline.put('\n');
|
|
dst->puts(dstline.str().c_str());
|
|
}
|
|
|
|
// close tags
|
|
dst->printf("\t</pre>\n");
|
|
|
|
// close the file
|
|
output_footer_and_close_file(std::move(dst), tempfooter, srcfile_subpath);
|
|
return 0;
|
|
}
|
|
|
|
|
|
|
|
/***************************************************************************
|
|
HTML OUTPUT HELPERS
|
|
***************************************************************************/
|
|
|
|
/*-------------------------------------------------
|
|
create_file_and_output_header - create a new
|
|
HTML file with a standard header
|
|
-------------------------------------------------*/
|
|
|
|
static util::core_file::ptr create_file_and_output_header(std::string &filename, std::string &templatefile, std::string &path)
|
|
{
|
|
// create the indexfile
|
|
util::core_file::ptr file;
|
|
if (util::core_file::open(filename, OPEN_FLAG_WRITE | OPEN_FLAG_CREATE | OPEN_FLAG_CREATE_PATHS | OPEN_FLAG_NO_BOM, file) != osd_file::error::NONE)
|
|
return util::core_file::ptr();
|
|
|
|
// print a header
|
|
std::string modified(templatefile);
|
|
strreplace(modified, "<!--PATH-->", path.c_str());
|
|
file->write(modified.c_str(), modified.length());
|
|
|
|
// return the file
|
|
return file;
|
|
}
|
|
|
|
|
|
/*-------------------------------------------------
|
|
output_footer_and_close_file - write a
|
|
standard footer to an HTML file and close it
|
|
-------------------------------------------------*/
|
|
|
|
static void output_footer_and_close_file(util::core_file::ptr &&file, std::string &templatefile, std::string &path)
|
|
{
|
|
std::string modified(templatefile);
|
|
strreplace(modified, "<!--PATH-->", path.c_str());
|
|
file->write(modified.c_str(), modified.length());
|
|
file.reset();
|
|
}
|
|
|
|
|
|
|
|
/***************************************************************************
|
|
HTML OUTPUT HELPERS
|
|
***************************************************************************/
|
|
|
|
/*-------------------------------------------------
|
|
normalized_subpath - normalize a path to
|
|
forward slashes and extract a subpath
|
|
-------------------------------------------------*/
|
|
|
|
static std::string &normalized_subpath(std::string &dest, std::string &path, int start)
|
|
{
|
|
strreplacechr(dest.assign(path.substr(start, -1)),PATH_SEPARATOR[0], '/');
|
|
return dest;
|
|
}
|
|
|
|
|
|
/*-------------------------------------------------
|
|
output_path_as_links - output a path as a
|
|
series of links
|
|
-------------------------------------------------*/
|
|
|
|
static void output_path_as_links(util::core_file &file, std::string &path, bool end_is_directory, bool link_to_file)
|
|
{
|
|
// first count how deep we are
|
|
int srcdepth = 0;
|
|
for (int slashindex = path.find_first_of('/'); slashindex != -1; slashindex = path.find_first_of('/', slashindex + 1))
|
|
srcdepth++;
|
|
if (end_is_directory)
|
|
srcdepth++;
|
|
|
|
// output a link to the root
|
|
file.printf("<a href=\"");
|
|
for (int depth = 0; depth < srcdepth; depth++)
|
|
file.printf("../");
|
|
file.printf("index.html\"><root></a>/");
|
|
|
|
// now output links to each path up the chain
|
|
int curdepth = 0;
|
|
int lastslash = 0;
|
|
for (int slashindex = path.find_first_of('/', lastslash); slashindex != -1; slashindex = path.find_first_of('/', lastslash))
|
|
{
|
|
std::string substr(path, lastslash, slashindex - lastslash);
|
|
|
|
curdepth++;
|
|
file.printf("<a href=\"");
|
|
for (int depth = curdepth; depth < srcdepth; depth++)
|
|
file.printf("../");
|
|
file.printf("index.html\">%s</a>/", substr.c_str());
|
|
|
|
lastslash = slashindex + 1;
|
|
}
|
|
|
|
// and a final link to the current directory
|
|
std::string substr(path, lastslash, -1);
|
|
if (end_is_directory)
|
|
file.printf("<a href=\"index.html\">%s</a>", substr.c_str());
|
|
else if (link_to_file)
|
|
file.printf("<a href=\"%s\">%s</a>", substr.c_str(), substr.c_str());
|
|
else
|
|
file.printf("<a href=\"%s.html\">%s</a>", substr.c_str(), substr.c_str());
|
|
}
|
|
|
|
|
|
/*-------------------------------------------------
|
|
find_include_file - find an include file
|
|
-------------------------------------------------*/
|
|
|
|
static bool find_include_file(std::string &srcincpath, int srcrootlen, int dstrootlen, std::string &srcfile, std::string &dstfile, std::string &filename)
|
|
{
|
|
// iterate over include paths and find the file
|
|
for (include_path *curpath = incpaths; curpath != nullptr; curpath = curpath->next)
|
|
{
|
|
// a '.' include path is specially treated
|
|
if (curpath->path.compare(".") == 0)
|
|
srcincpath.assign(srcfile.substr(0, srcfile.find_last_of(PATH_SEPARATOR[0])));
|
|
else
|
|
srcincpath.assign(srcfile.substr(0, srcrootlen + 1)).append(curpath->path);
|
|
|
|
// append the filename piecemeal to account for directories
|
|
int lastsepindex = 0;
|
|
int sepindex;
|
|
while ((sepindex = filename.find_first_of('/', lastsepindex)) != -1)
|
|
{
|
|
// handle .. by removing a chunk from the incpath
|
|
std::string pathpart(filename, lastsepindex, sepindex - lastsepindex);
|
|
if (pathpart.compare("..")==0)
|
|
{
|
|
sepindex = srcincpath.find_last_of(PATH_SEPARATOR[0]);
|
|
if (sepindex != -1)
|
|
srcincpath.erase(sepindex);
|
|
}
|
|
|
|
// otherwise, append a path separator and the pathpart
|
|
else
|
|
srcincpath.append(PATH_SEPARATOR).append(pathpart);
|
|
|
|
// advance past the previous index
|
|
lastsepindex = sepindex + 1;
|
|
}
|
|
|
|
// now append the filename
|
|
srcincpath.append(PATH_SEPARATOR).append(filename.substr(lastsepindex, -1));
|
|
|
|
// see if we can open it
|
|
util::core_file::ptr testfile;
|
|
if (util::core_file::open(srcincpath, OPEN_FLAG_READ, testfile) == osd_file::error::NONE)
|
|
{
|
|
// close the file
|
|
testfile.reset();
|
|
|
|
// find the longest matching directory substring between the include and source file
|
|
lastsepindex = 0;
|
|
while ((sepindex = srcincpath.find_first_of(PATH_SEPARATOR[0], lastsepindex)) != -1)
|
|
{
|
|
// get substrings up to the current directory
|
|
std::string tempfile(srcfile, 0, sepindex);
|
|
std::string tempinc(srcincpath, 0, sepindex);
|
|
|
|
// if we don't match, stop
|
|
if (tempfile.compare(tempinc)!=0)
|
|
break;
|
|
lastsepindex = sepindex + 1;
|
|
}
|
|
|
|
// chop off the common parts of the paths
|
|
std::string tempfile(srcfile, lastsepindex, -1);
|
|
srcincpath = srcincpath.substr(lastsepindex, -1);
|
|
strreplacechr(srcincpath, PATH_SEPARATOR[0], '/');
|
|
|
|
// for each directory left in the filename, we need to prepend a "../"
|
|
while ((sepindex = tempfile.find_first_of(PATH_SEPARATOR[0])) != -1)
|
|
{
|
|
tempfile.erase(0, sepindex + 1);
|
|
srcincpath.insert(0, "../");
|
|
}
|
|
srcincpath.append(".html");
|
|
|
|
// free the strings and return the include path
|
|
return true;
|
|
}
|
|
}
|
|
return false;
|
|
}
|