summaryrefslogtreecommitdiff
path: root/Build/source/utils/texinfo/makeinfo/makeinfo.c
diff options
context:
space:
mode:
Diffstat (limited to 'Build/source/utils/texinfo/makeinfo/makeinfo.c')
-rw-r--r--Build/source/utils/texinfo/makeinfo/makeinfo.c4157
1 files changed, 4157 insertions, 0 deletions
diff --git a/Build/source/utils/texinfo/makeinfo/makeinfo.c b/Build/source/utils/texinfo/makeinfo/makeinfo.c
new file mode 100644
index 00000000000..22ed4c4fd99
--- /dev/null
+++ b/Build/source/utils/texinfo/makeinfo/makeinfo.c
@@ -0,0 +1,4157 @@
+/* makeinfo -- convert Texinfo source into other formats.
+ $Id: makeinfo.c,v 1.74 2004/12/19 17:15:42 karl Exp $
+
+ Copyright (C) 1987, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
+ 2000, 2001, 2002, 2003, 2004 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 2, or (at your option)
+ any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program; if not, write to the Free Software
+ Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+
+ Original author of makeinfo: Brian Fox (bfox@ai.mit.edu). */
+
+#include "system.h"
+#include "getopt.h"
+
+#define COMPILING_MAKEINFO
+#include "makeinfo.h"
+#include "cmds.h"
+#include "files.h"
+#include "float.h"
+#include "footnote.h"
+#include "html.h"
+#include "index.h"
+#include "insertion.h"
+#include "lang.h"
+#include "macro.h"
+#include "node.h"
+#include "sectioning.h"
+#include "toc.h"
+#include "xml.h"
+
+/* You can change some of the behavior of Makeinfo by changing the
+ following defines: */
+
+/* Define INDENT_PARAGRAPHS_IN_TABLE if you want the paragraphs which
+ appear within an @table, @ftable, or @itemize environment to have
+ standard paragraph indentation. Without this, such paragraphs have
+ no starting indentation. */
+/* #define INDENT_PARAGRAPHS_IN_TABLE */
+
+/* Define PARAGRAPH_START_INDENT to be the amount of indentation that
+ the first lines of paragraphs receive by default, where no other
+ value has been specified. Users can change this value on the command
+ line, with the --paragraph-indent option, or within the texinfo file,
+ with the @paragraphindent command. */
+#define PARAGRAPH_START_INDENT 3
+
+/* Define DEFAULT_PARAGRAPH_SPACING as the number of blank lines that you
+ wish to appear between paragraphs. A value of 1 creates a single blank
+ line between paragraphs. Paragraphs are defined by 2 or more consecutive
+ newlines in the input file (i.e., one or more blank lines). */
+#define DEFAULT_PARAGRAPH_SPACING 1
+
+/* Global variables. */
+
+/* The output file name. */
+char *output_filename = NULL;
+
+/* Name of the output file that the user elected to pass on the command line.
+ Such a name overrides any name found with the @setfilename command. */
+char *command_output_filename = NULL;
+static char *save_command_output_filename = NULL;
+
+#define INITIAL_PARAGRAPH_SPACE 5000
+int paragraph_buffer_len = INITIAL_PARAGRAPH_SPACE;
+
+/* The amount of indentation to add at the starts of paragraphs.
+ 0 means don't change existing indentation at paragraph starts.
+ > 0 is amount to indent new paragraphs by.
+ < 0 means indent to column zero by removing indentation if necessary.
+
+ This is normally zero, but some people prefer paragraph starts to be
+ somewhat more indented than paragraph bodies. A pretty value for
+ this is 3. */
+int paragraph_start_indent = PARAGRAPH_START_INDENT;
+
+/* Indentation that is pending insertion. We have this for hacking lines
+ which look blank, but contain whitespace. We want to treat those as
+ blank lines. */
+int pending_indent = 0;
+
+/* The index in our internal command table of the currently
+ executing command. */
+int command_index;
+
+/* A search string which is used to find the first @setfilename. */
+char setfilename_search[] =
+ { COMMAND_PREFIX,
+ 's', 'e', 't', 'f', 'i', 'l', 'e', 'n', 'a', 'm', 'e', 0 };
+
+/* Values for calling handle_variable_internal (). */
+#define SET 1
+#define CLEAR 2
+#define IFSET 3
+#define IFCLEAR 4
+
+/* Flags controlling the operation of the program. */
+
+/* Default is to remove output if there were errors. */
+int force = 0;
+
+/* Default is to notify users of bad choices. */
+int print_warnings = 1;
+
+/* Number of errors that we tolerate on a given fileset. */
+int max_error_level = 100;
+
+/* The actual last inserted character. Note that this may be something
+ other than NEWLINE even if last_char_was_newline is 1. */
+int last_inserted_character = 0;
+
+/* Nonzero means that a newline character has already been
+ inserted, so close_paragraph () should insert one less. */
+int line_already_broken = 0;
+
+/* When nonzero we have finished an insertion (see end_insertion ()) and we
+ want to ignore false continued paragraph closings. */
+int insertion_paragraph_closed = 0;
+
+/* Nonzero means attempt to make all of the lines have fill_column width. */
+int do_justification = 0;
+
+/* Nonzero means don't replace whitespace with &nbsp; in HTML mode. */
+int in_html_elt = 0;
+
+/* Nonzero means we are inserting a block level HTML element that must not be
+ enclosed in a <p>, such as <ul>, <ol> and <h?>. */
+int in_html_block_level_elt = 0;
+
+/* True when expanding a macro definition. */
+static int executing_macro = 0;
+
+/* True when we are inside a <li> block of a menu. */
+static int in_menu_item = 0;
+
+typedef struct brace_element
+{
+ struct brace_element *next;
+ COMMAND_FUNCTION *proc;
+ char *command;
+ int pos, line;
+ int in_fixed_width_font;
+} BRACE_ELEMENT;
+
+BRACE_ELEMENT *brace_stack = NULL;
+
+static void convert_from_file (char *name);
+static void convert_from_loaded_file (char *name);
+static void convert_from_stream (FILE *stream, char *name);
+static void do_flush_right_indentation (void);
+static void handle_variable (int action);
+static void handle_variable_internal (int action, char *name);
+static void init_brace_stack (void);
+static void init_internals (void);
+static void pop_and_call_brace (void);
+static void remember_brace (COMMAND_FUNCTION (*proc));
+static int end_of_sentence_p (void);
+
+void maybe_update_execution_strings (char **text, unsigned int new_len);
+
+/* Error handling. */
+
+/* Number of errors encountered. */
+int errors_printed = 0;
+
+/* Remember that an error has been printed. If more than
+ max_error_level have been printed, then exit the program. */
+static void
+remember_error (void)
+{
+ errors_printed++;
+ if (max_error_level && (errors_printed > max_error_level))
+ {
+ fprintf (stderr, _("Too many errors! Gave up.\n"));
+ flush_file_stack ();
+ if (errors_printed - max_error_level < 2)
+ cm_bye ();
+ xexit (1);
+ }
+}
+
+/* Print the last error gotten from the file system. */
+int
+fs_error (char *filename)
+{
+ remember_error ();
+ perror (filename);
+ return 0;
+}
+
+/* Print an error message, and return false. */
+void
+#if defined (VA_FPRINTF) && __STDC__
+error (const char *format, ...)
+#else
+error (format, va_alist)
+ const char *format;
+ va_dcl
+#endif
+{
+#ifdef VA_FPRINTF
+ va_list ap;
+#endif
+
+ remember_error ();
+
+ VA_START (ap, format);
+#ifdef VA_FPRINTF
+ VA_FPRINTF (stderr, format, ap);
+#else
+ fprintf (stderr, format, a1, a2, a3, a4, a5, a6, a7, a8);
+#endif /* not VA_FPRINTF */
+ va_end (ap);
+
+ putc ('\n', stderr);
+}
+
+/* Just like error (), but print the input file and line number as well. */
+void
+#if defined (VA_FPRINTF) && __STDC__
+file_line_error (char *infile, int lno, const char *format, ...)
+#else
+file_line_error (infile, lno, format, va_alist)
+ char *infile;
+ int lno;
+ const char *format;
+ va_dcl
+#endif
+{
+#ifdef VA_FPRINTF
+ va_list ap;
+#endif
+
+ remember_error ();
+ fprintf (stderr, "%s:%d: ", infile, lno);
+
+ VA_START (ap, format);
+#ifdef VA_FPRINTF
+ VA_FPRINTF (stderr, format, ap);
+#else
+ fprintf (stderr, format, a1, a2, a3, a4, a5, a6, a7, a8);
+#endif /* not VA_FPRINTF */
+ va_end (ap);
+
+ fprintf (stderr, ".\n");
+}
+
+/* Just like file_line_error (), but take the input file and the line
+ number from global variables. */
+void
+#if defined (VA_FPRINTF) && __STDC__
+line_error (const char *format, ...)
+#else
+line_error (format, va_alist)
+ const char *format;
+ va_dcl
+#endif
+{
+#ifdef VA_FPRINTF
+ va_list ap;
+#endif
+
+ remember_error ();
+ fprintf (stderr, "%s:%d: ", input_filename, line_number);
+
+ VA_START (ap, format);
+#ifdef VA_FPRINTF
+ VA_FPRINTF (stderr, format, ap);
+#else
+ fprintf (stderr, format, a1, a2, a3, a4, a5, a6, a7, a8);
+#endif /* not VA_FPRINTF */
+ va_end (ap);
+
+ fprintf (stderr, ".\n");
+}
+
+void
+#if defined (VA_FPRINTF) && __STDC__
+warning (const char *format, ...)
+#else
+warning (format, va_alist)
+ const char *format;
+ va_dcl
+#endif
+{
+#ifdef VA_FPRINTF
+ va_list ap;
+#endif
+
+ if (print_warnings)
+ {
+ fprintf (stderr, _("%s:%d: warning: "), input_filename, line_number);
+
+ VA_START (ap, format);
+#ifdef VA_FPRINTF
+ VA_FPRINTF (stderr, format, ap);
+#else
+ fprintf (stderr, format, a1, a2, a3, a4, a5, a6, a7, a8);
+#endif /* not VA_FPRINTF */
+ va_end (ap);
+
+ fprintf (stderr, ".\n");
+ }
+}
+
+
+/* The other side of a malformed expression. */
+static void
+misplaced_brace (void)
+{
+ line_error (_("Misplaced %c"), '}');
+}
+
+/* Main. */
+
+/* Display the version info of this invocation of Makeinfo. */
+static void
+print_version_info (void)
+{
+ printf ("makeinfo (GNU %s) %s\n", PACKAGE, VERSION);
+}
+
+/* If EXIT_VALUE is zero, print the full usage message to stdout.
+ Otherwise, just say to use --help for more info.
+ Then exit with EXIT_VALUE. */
+static void
+usage (int exit_value)
+{
+ if (exit_value != 0)
+ fprintf (stderr, _("Try `%s --help' for more information.\n"), progname);
+ else
+ {
+ printf (_("Usage: %s [OPTION]... TEXINFO-FILE...\n"), progname);
+ puts ("");
+
+ puts (_("\
+Translate Texinfo source documentation to various other formats, by default\n\
+Info files suitable for reading online with Emacs or standalone GNU Info.\n"));
+
+ printf (_("\
+General options:\n\
+ --error-limit=NUM quit after NUM errors (default %d).\n\
+ --force preserve output even if errors.\n\
+ --help display this help and exit.\n\
+ --no-validate suppress node cross-reference validation.\n\
+ --no-warn suppress warnings (but not errors).\n\
+ --reference-limit=NUM warn about at most NUM references (default %d).\n\
+ -v, --verbose explain what is being done.\n\
+ --version display version information and exit.\n"),
+ max_error_level, reference_warning_limit);
+ puts ("");
+
+ /* xgettext: no-wrap */
+ puts (_("\
+Output format selection (default is to produce Info):\n\
+ --docbook output Docbook XML rather than Info.\n\
+ --html output HTML rather than Info.\n\
+ --xml output Texinfo XML rather than Info.\n\
+ --plaintext output plain text rather than Info.\n\
+"));
+
+ puts (_("\
+General output options:\n\
+ -E, --macro-expand FILE output macro-expanded source to FILE.\n\
+ ignoring any @setfilename.\n\
+ --no-headers suppress node separators, Node: lines, and menus\n\
+ from Info output (thus producing plain text)\n\
+ or from HTML (thus producing shorter output);\n\
+ also, write to standard output by default.\n\
+ --no-split suppress splitting of Info or HTML output,\n\
+ generate only one output file.\n\
+ --number-sections output chapter and sectioning numbers.\n\
+ -o, --output=FILE output to FILE (directory if split HTML),\n\
+"));
+
+ printf (_("\
+Options for Info and plain text:\n\
+ --enable-encoding output accented and special characters in\n\
+ Info output based on @documentencoding.\n\
+ --fill-column=NUM break Info lines at NUM characters (default %d).\n\
+ --footnote-style=STYLE output footnotes in Info according to STYLE:\n\
+ `separate' to put them in their own node;\n\
+ `end' to put them at the end of the node\n\
+ in which they are defined (default).\n\
+ --paragraph-indent=VAL indent Info paragraphs by VAL spaces (default %d).\n\
+ If VAL is `none', do not indent; if VAL is\n\
+ `asis', preserve existing indentation.\n\
+ --split-size=NUM split Info files at size NUM (default %d).\n"),
+ fill_column, paragraph_start_indent,
+ DEFAULT_SPLIT_SIZE);
+ puts ("");
+
+ puts (_("\
+Options for HTML:\n\
+ --css-include=FILE include FILE in HTML <style> output;\n\
+ read stdin if FILE is -.\n\
+"));
+
+ printf (_("\
+Options for XML and Docbook:\n\
+ --output-indent=VAL indent XML elements by VAL spaces (default %d).\n\
+ If VAL is 0, ignorable whitespace is dropped.\n\
+"), xml_indentation_increment);
+ puts ("");
+
+ puts (_("\
+Input file options:\n\
+ --commands-in-node-names allow @ commands in node names.\n\
+ -D VAR define the variable VAR, as with @set.\n\
+ -I DIR append DIR to the @include search path.\n\
+ -P DIR prepend DIR to the @include search path.\n\
+ -U VAR undefine the variable VAR, as with @clear.\n\
+"));
+
+ puts (_("\
+Conditional processing in input:\n\
+ --ifdocbook process @ifdocbook and @docbook even if\n\
+ not generating Docbook.\n\
+ --ifhtml process @ifhtml and @html even if not generating HTML.\n\
+ --ifinfo process @ifinfo even if not generating Info.\n\
+ --ifplaintext process @ifplaintext even if not generating plain text.\n\
+ --iftex process @iftex and @tex; implies --no-split.\n\
+ --ifxml process @ifxml and @xml.\n\
+ --no-ifdocbook do not process @ifdocbook and @docbook text.\n\
+ --no-ifhtml do not process @ifhtml and @html text.\n\
+ --no-ifinfo do not process @ifinfo text.\n\
+ --no-ifplaintext do not process @ifplaintext text.\n\
+ --no-iftex do not process @iftex and @tex text.\n\
+ --no-ifxml do not process @ifxml and @xml text.\n\
+\n\
+ Also, for the --no-ifFORMAT options, do process @ifnotFORMAT text.\n\
+"));
+
+ puts (_("\
+ The defaults for the @if... conditionals depend on the output format:\n\
+ if generating HTML, --ifhtml is on and the others are off;\n\
+ if generating Info, --ifinfo is on and the others are off;\n\
+ if generating plain text, --ifplaintext is on and the others are off;\n\
+ if generating XML, --ifxml is on and the others are off.\n\
+"));
+
+ fputs (_("\
+Examples:\n\
+ makeinfo foo.texi write Info to foo's @setfilename\n\
+ makeinfo --html foo.texi write HTML to @setfilename\n\
+ makeinfo --xml foo.texi write Texinfo XML to @setfilename\n\
+ makeinfo --docbook foo.texi write DocBook XML to @setfilename\n\
+ makeinfo --no-headers foo.texi write plain text to standard output\n\
+\n\
+ makeinfo --html --no-headers foo.texi write html without node lines, menus\n\
+ makeinfo --number-sections foo.texi write Info with numbered sections\n\
+ makeinfo --no-split foo.texi write one Info file however big\n\
+"), stdout);
+
+ puts (_("\n\
+Email bug reports to bug-texinfo@gnu.org,\n\
+general questions and discussion to help-texinfo@gnu.org.\n\
+Texinfo home page: http://www.gnu.org/software/texinfo/"));
+
+ } /* end of full help */
+
+ xexit (exit_value);
+}
+
+struct option long_options[] =
+{
+ { "commands-in-node-names", 0, &expensive_validation, 1 },
+ { "css-include", 1, 0, 'C' },
+ { "docbook", 0, 0, 'd' },
+ { "enable-encoding", 0, &enable_encoding, 1 },
+ { "error-limit", 1, 0, 'e' },
+ { "fill-column", 1, 0, 'f' },
+ { "footnote-style", 1, 0, 's' },
+ { "force", 0, &force, 1 },
+ { "help", 0, 0, 'h' },
+ { "html", 0, 0, 'w' },
+ { "ifdocbook", 0, &process_docbook, 1 },
+ { "ifhtml", 0, &process_html, 1 },
+ { "ifinfo", 0, &process_info, 1 },
+ { "ifplaintext", 0, &process_plaintext, 1 },
+ { "iftex", 0, &process_tex, 1 },
+ { "ifxml", 0, &process_xml, 1 },
+ { "macro-expand", 1, 0, 'E' },
+ { "no-headers", 0, &no_headers, 1 },
+ { "no-ifdocbook", 0, &process_docbook, 0 },
+ { "no-ifhtml", 0, &process_html, 0 },
+ { "no-ifinfo", 0, &process_info, 0 },
+ { "no-ifplaintext", 0, &process_plaintext, 0 },
+ { "no-iftex", 0, &process_tex, 0 },
+ { "no-ifxml", 0, &process_xml, 0 },
+ { "no-number-footnotes", 0, &number_footnotes, 0 },
+ { "no-number-sections", 0, &number_sections, 0 },
+ { "no-pointer-validate", 0, &validating, 0 },
+ { "no-split", 0, &splitting, 0 },
+ { "no-validate", 0, &validating, 0 },
+ { "no-warn", 0, &print_warnings, 0 },
+ { "number-footnotes", 0, &number_footnotes, 1 },
+ { "number-sections", 0, &number_sections, 1 },
+ { "output", 1, 0, 'o' },
+ { "output-indent", 1, 0, 'i' },
+ { "paragraph-indent", 1, 0, 'p' },
+ { "plaintext", 0, 0, 't' },
+ { "reference-limit", 1, 0, 'r' },
+ { "split-size", 1, 0, 'S'},
+ { "verbose", 0, &verbose_mode, 1 },
+ { "version", 0, 0, 'V' },
+ { "xml", 0, 0, 'x' },
+ {NULL, 0, NULL, 0}
+};
+
+/* We use handle_variable_internal for -D and -U, and it depends on
+ execute_string, which depends on input_filename, which is not defined
+ while we are handling options. :-\ So we save these defines in this
+ struct, and handle them later. */
+typedef struct command_line_define
+{
+ struct command_line_define *next;
+ int action;
+ char *define;
+} COMMAND_LINE_DEFINE;
+
+static COMMAND_LINE_DEFINE *command_line_defines = NULL;
+
+/* For each file mentioned in the command line, process it, turning
+ Texinfo commands into wonderfully formatted output text. */
+int
+main (int argc, char **argv)
+{
+ int c, ind;
+ int reading_from_stdin = 0;
+
+#ifdef HAVE_SETLOCALE
+ /* Do not use LC_ALL, because LC_NUMERIC screws up the scanf parsing
+ of the argument to @multicolumn. */
+ setlocale (LC_TIME, "");
+#ifdef LC_MESSAGES /* ultrix */
+ setlocale (LC_MESSAGES, "");
+#endif
+ setlocale (LC_CTYPE, "");
+ setlocale (LC_COLLATE, "");
+#endif
+
+#ifdef ENABLE_NLS
+ /* Set the text message domain. */
+ bindtextdomain (PACKAGE, LOCALEDIR);
+ textdomain (PACKAGE);
+#endif
+
+ /* If TEXINFO_OUTPUT_FORMAT envvar is set, use it to set default output.
+ Can be overridden with one of the output options. */
+ if (getenv ("TEXINFO_OUTPUT_FORMAT") != NULL)
+ {
+ if (STREQ (getenv ("TEXINFO_OUTPUT_FORMAT"), "docbook"))
+ {
+ splitting = 0;
+ html = 0;
+ docbook = 1;
+ xml = 1;
+ process_docbook = 1;
+ }
+ else if (STREQ (getenv ("TEXINFO_OUTPUT_FORMAT"), "html"))
+ {
+ html = 1;
+ docbook = 0;
+ xml = 0;
+ process_html = 1;
+ }
+ else if (STREQ (getenv ("TEXINFO_OUTPUT_FORMAT"), "info"))
+ {
+ html = 0;
+ docbook = 0;
+ xml = 0;
+ }
+ else if (STREQ (getenv ("TEXINFO_OUTPUT_FORMAT"), "plaintext"))
+ {
+ splitting = 0;
+ no_headers = 1;
+ html = 0;
+ docbook = 0;
+ xml = 0;
+ process_plaintext = 1;
+ }
+ else if (STREQ (getenv ("TEXINFO_OUTPUT_FORMAT"), "xml"))
+ {
+ splitting = 0;
+ html = 0;
+ docbook = 0;
+ xml = 1;
+ process_xml = 1;
+ }
+ else
+ fprintf (stderr,
+ _("%s: Ignoring unrecognized TEXINFO_OUTPUT_FORMAT value `%s'.\n"),
+ progname, getenv ("TEXINFO_OUTPUT_FORMAT"));
+ }
+
+ /* Parse argument flags from the input line. */
+ while ((c = getopt_long (argc, argv, "D:de:E:f:hI:i:o:p:P:r:s:t:U:vV:wx",
+ long_options, &ind)) != EOF)
+ {
+ if (c == 0 && long_options[ind].flag == 0)
+ c = long_options[ind].val;
+
+ switch (c)
+ {
+ case 'C': /* --css-include */
+ css_include = xstrdup (optarg);
+ break;
+
+ case 'D':
+ case 'U':
+ /* User specified variable to set or clear. */
+ if (xml && !docbook)
+ {
+ COMMAND_LINE_DEFINE *new = xmalloc (sizeof (COMMAND_LINE_DEFINE));
+ new->action = (c == 'D') ? SET : CLEAR;
+ new->define = xstrdup (optarg);
+ new->next = command_line_defines;
+ command_line_defines = new;
+ }
+ else
+ handle_variable_internal ((c == 'D' ? SET : CLEAR), optarg);
+ break;
+
+ case 'd': /* --docbook */
+ splitting = 0;
+ xml = 1;
+ docbook = 1;
+ html = 0;
+ process_docbook = 1;
+ break;
+
+ case 'e': /* --error-limit */
+ if (sscanf (optarg, "%d", &max_error_level) != 1)
+ {
+ fprintf (stderr,
+ _("%s: %s arg must be numeric, not `%s'.\n"),
+ progname, "--error-limit", optarg);
+ usage (1);
+ }
+ break;
+
+ case 'E': /* --macro-expand */
+ if (!macro_expansion_output_stream)
+ {
+ macro_expansion_filename = optarg;
+ macro_expansion_output_stream
+ = strcmp (optarg, "-") == 0 ? stdout : fopen (optarg, "w");
+ if (!macro_expansion_output_stream)
+ error (_("%s: could not open macro expansion output `%s'"),
+ progname, optarg);
+ }
+ else
+ fprintf (stderr,
+ _("%s: ignoring second macro expansion output `%s'.\n"),
+ progname, optarg);
+ break;
+
+ case 'f': /* --fill-column */
+ if (sscanf (optarg, "%d", &fill_column) != 1)
+ {
+ fprintf (stderr,
+ _("%s: %s arg must be numeric, not `%s'.\n"),
+ progname, "--fill-column", optarg);
+ usage (1);
+ }
+ break;
+
+ case 'h': /* --help */
+ usage (0);
+ break;
+
+ case 'I':
+ /* Append user-specified dir to include file path. */
+ append_to_include_path (optarg);
+ break;
+
+ case 'i':
+ if (sscanf (optarg, "%d", &xml_indentation_increment) != 1)
+ {
+ fprintf (stderr,
+ _("%s: %s arg must be numeric, not `%s'.\n"),
+ progname, "--output-indent", optarg);
+ usage (1);
+ }
+ break;
+
+ case 'o': /* --output */
+ command_output_filename = xstrdup (optarg);
+ save_command_output_filename = command_output_filename;
+ break;
+
+ case 'p': /* --paragraph-indent */
+ if (set_paragraph_indent (optarg) < 0)
+ {
+ fprintf (stderr,
+ _("%s: --paragraph-indent arg must be numeric/`none'/`asis', not `%s'.\n"),
+ progname, optarg);
+ usage (1);
+ }
+ break;
+
+ case 'P':
+ /* Prepend user-specified include dir to include path. */
+ prepend_to_include_path (optarg);
+ break;
+
+ case 'r': /* --reference-limit */
+ if (sscanf (optarg, "%d", &reference_warning_limit) != 1)
+ {
+ fprintf (stderr,
+ _("%s: %s arg must be numeric, not `%s'.\n"),
+ progname, "--reference-limit", optarg);
+ usage (1);
+ }
+ break;
+
+ case 's': /* --footnote-style */
+ if (set_footnote_style (optarg) < 0)
+ {
+ fprintf (stderr,
+ _("%s: --footnote-style arg must be `separate' or `end', not `%s'.\n"),
+ progname, optarg);
+ usage (1);
+ }
+ footnote_style_preset = 1;
+ break;
+
+ case 'S': /* --split-size */
+ if (sscanf (optarg, "%d", &split_size) != 1)
+ {
+ fprintf (stderr,
+ _("%s: %s arg must be numeric, not `%s'.\n"),
+ progname, "--split-size", optarg);
+ usage (1);
+ }
+ break;
+
+ case 't': /* --plaintext */
+ splitting = 0;
+ no_headers = 1;
+ html = 0;
+ docbook = 0;
+ xml = 0;
+ process_plaintext = 1;
+ break;
+
+ case 'v':
+ verbose_mode++;
+ break;
+
+ case 'V': /* --version */
+ print_version_info ();
+ puts ("");
+ puts ("Copyright (C) 2004 Free Software Foundation, Inc.");
+ printf (_("There is NO warranty. You may redistribute this software\n\
+under the terms of the GNU General Public License.\n\
+For more information about these matters, see the files named COPYING.\n"));
+ xexit (0);
+ break;
+
+ case 'w': /* --html */
+ xml = 0;
+ docbook = 0;
+ html = 1;
+ process_html = 1;
+ break;
+
+ case 'x': /* --xml */
+ splitting = 0;
+ html = 0;
+ docbook = 0;
+ xml = 1;
+ process_xml = 1;
+ break;
+
+ case '?':
+ usage (1);
+ break;
+ }
+ }
+
+ if (macro_expansion_output_stream)
+ validating = 0;
+
+ if (!validating)
+ expensive_validation = 0;
+
+ if (optind == argc)
+ {
+ /* Check to see if input is a file. If so, process that. */
+ if (!isatty (fileno (stdin)))
+ reading_from_stdin = 1;
+ else
+ {
+ fprintf (stderr, _("%s: missing file argument.\n"), progname);
+ usage (1);
+ }
+ }
+
+ if (no_headers)
+ {
+ /* If the user did not specify an output file, use stdout. */
+ if (!command_output_filename)
+ command_output_filename = xstrdup ("-");
+
+ if (html && splitting && !STREQ (command_output_filename, "-"))
+ { /* --no-headers --no-split --html indicates confusion. */
+ fprintf (stderr,
+ "%s: can't split --html output to `%s' with --no-headers.\n",
+ progname, command_output_filename);
+ usage (1);
+ }
+
+ /* --no-headers implies --no-split. */
+ splitting = 0;
+ }
+
+ if (process_info == -1)
+ { /* no explicit --[no-]ifinfo option, so we'll do @ifinfo
+ if we're generating info or (for compatibility) plain text. */
+ process_info = !html && !xml;
+ }
+
+ if (process_plaintext == -1)
+ { /* no explicit --[no-]ifplaintext option, so we'll do @ifplaintext
+ if we're generating plain text. */
+ process_plaintext = no_headers && !html && !xml;
+ }
+
+ if (verbose_mode)
+ print_version_info ();
+
+ /* Remaining arguments are file names of texinfo files.
+ Convert them, one by one. */
+ if (!reading_from_stdin)
+ {
+ while (optind != argc)
+ convert_from_file (argv[optind++]);
+ }
+ else
+ convert_from_stream (stdin, "stdin");
+
+ xexit (errors_printed ? 2 : 0);
+ return 0; /* Avoid bogus warnings. */
+}
+
+/* Hacking tokens and strings. */
+
+/* Return the next token as a string pointer. We cons the string. This
+ `token' means simply a command name. */
+
+/* = is so @alias works. ^ and _ are so macros can be used in math mode
+ without a space following. Possibly we should simply allow alpha, to
+ be compatible with TeX. */
+#define COMMAND_CHAR(c) (!cr_or_whitespace(c) \
+ && (c) != '{' \
+ && (c) != '}' \
+ && (c) != '=' \
+ && (c) != '_' \
+ && (c) != '^' \
+ )
+
+static char *
+read_token (void)
+{
+ int i, character;
+ char *result;
+
+ /* If the first character to be read is self-delimiting, then that
+ is the command itself. */
+ character = curchar ();
+ if (self_delimiting (character))
+ {
+ input_text_offset++;
+
+ if (character == '\n')
+ line_number++;
+
+ result = xstrdup (" ");
+ *result = character;
+ return result;
+ }
+
+ for (i = 0; ((input_text_offset != input_text_length)
+ && (character = curchar ())
+ && COMMAND_CHAR (character));
+ i++, input_text_offset++);
+ result = xmalloc (i + 1);
+ memcpy (result, &input_text[input_text_offset - i], i);
+ result[i] = 0;
+ return result;
+}
+
+/* Return nonzero if CHARACTER is self-delimiting. */
+int
+self_delimiting (int character)
+{
+ /* @; and @\ are not Texinfo commands, but they are listed here
+ anyway. I don't know why. --karl, 10aug96. */
+ return strchr ("~{|}`^\\@?=;:./-,*\'\" !\n\t", character) != NULL;
+}
+
+/* Clear whitespace from the front and end of string. */
+void
+canon_white (char *string)
+{
+ char *p = string;
+ unsigned len;
+
+ if (!*p)
+ return;
+
+ do
+ {
+ if (!cr_or_whitespace (*p))
+ break;
+ ++p;
+ }
+ while (*p);
+
+ len = strlen (p);
+ while (len && cr_or_whitespace (p[len-1]))
+ --len;
+
+ if (p != string)
+ memmove (string, p, len);
+
+ string[len] = 0;
+}
+
+/* Bash STRING, replacing all whitespace with just one space. */
+void
+fix_whitespace (char *string)
+{
+ char *temp = xmalloc (strlen (string) + 1);
+ int string_index = 0;
+ int temp_index = 0;
+ int c;
+
+ canon_white (string);
+
+ while (string[string_index])
+ {
+ c = temp[temp_index++] = string[string_index++];
+
+ if (c == ' ' || c == '\n' || c == '\t')
+ {
+ temp[temp_index - 1] = ' ';
+ while ((c = string[string_index]) && (c == ' ' ||
+ c == '\t' ||
+ c == '\n'))
+ string_index++;
+ }
+ }
+ temp[temp_index] = 0;
+ strcpy (string, temp);
+ free (temp);
+}
+
+/* Discard text until the desired string is found. The string is
+ included in the discarded text. */
+void
+discard_until (char *string)
+{
+ int temp = search_forward (string, input_text_offset);
+
+ int tt = (temp < 0) ? input_text_length : temp + strlen (string);
+ int from = input_text_offset;
+
+ /* Find out what line we are on. */
+ while (from != tt)
+ if (input_text[from++] == '\n')
+ line_number++;
+
+ if (temp < 0)
+ {
+ /* not found, move current position to end of string */
+ input_text_offset = input_text_length;
+ if (strcmp (string, "\n") != 0)
+ { /* Give a more descriptive feedback, if we are looking for ``@end ''
+ during macro execution. That means someone used a multiline
+ command as an argument to, say, @section ... style commands. */
+ char *end_block = xmalloc (8);
+ sprintf (end_block, "\n%cend ", COMMAND_PREFIX);
+ if (executing_string && strstr (string, end_block))
+ line_error (_("Multiline command %c%s used improperly"),
+ COMMAND_PREFIX, command);
+ else
+ line_error (_("Expected `%s'"), string);
+ free (end_block);
+ return;
+ }
+ }
+ else
+ /* found, move current position to after the found string */
+ input_text_offset = temp + strlen (string);
+}
+
+/* Read characters from the file until we are at MATCH.
+ Place the characters read into STRING.
+ On exit input_text_offset is after the match string.
+ Return the offset where the string starts. */
+int
+get_until (char *match, char **string)
+{
+ int len, current_point, x, new_point, tem;
+
+ current_point = x = input_text_offset;
+ new_point = search_forward (match, input_text_offset);
+
+ if (new_point < 0)
+ new_point = input_text_length;
+ len = new_point - current_point;
+
+ /* Keep track of which line number we are at. */
+ tem = new_point + (strlen (match) - 1);
+ while (x != tem)
+ if (input_text[x++] == '\n')
+ line_number++;
+
+ *string = xmalloc (len + 1);
+
+ memcpy (*string, &input_text[current_point], len);
+ (*string)[len] = 0;
+
+ /* Now leave input_text_offset in a consistent state. */
+ input_text_offset = tem;
+
+ if (input_text_offset > input_text_length)
+ input_text_offset = input_text_length;
+
+ return new_point;
+}
+
+/* Replace input_text[FROM .. TO] with its expansion. */
+void
+replace_with_expansion (int from, int *to)
+{
+ char *xp;
+ unsigned xp_len, new_len;
+ char *old_input = input_text;
+ unsigned raw_len = *to - from;
+ char *str;
+
+ /* The rest of the code here moves large buffers, so let's
+ not waste time if the input cannot possibly expand
+ into anything. Unfortunately, we cannot avoid expansion
+ when we see things like @code etc., even if they only
+ asked for expansion of macros, since any Texinfo command
+ can be potentially redefined with a macro. */
+ if (only_macro_expansion &&
+ memchr (input_text + from, COMMAND_PREFIX, raw_len) == 0)
+ return;
+
+ /* Get original string from input. */
+ str = xmalloc (raw_len + 1);
+ memcpy (str, input_text + from, raw_len);
+ str[raw_len] = 0;
+
+ /* We are going to relocate input_text, so we had better output
+ pending portion of input_text now, before the pointer changes. */
+ if (macro_expansion_output_stream && !executing_string
+ && !me_inhibit_expansion)
+ append_to_expansion_output (from);
+
+ /* Expand it. */
+ xp = expansion (str, 0);
+ xp_len = strlen (xp);
+ free (str);
+
+ /* Plunk the expansion into the middle of `input_text' --
+ which is terminated by a newline, not a null. Avoid
+ expensive move of the rest of the input if the expansion
+ has the same length as the original string. */
+ if (xp_len != raw_len)
+ {
+ new_len = from + xp_len + input_text_length - *to + 1;
+ if (executing_string)
+ { /* If we are in execute_string, we might need to update
+ the relevant element in the execution_strings[] array,
+ since it could have to be relocated from under our
+ feet. (input_text is reallocated here as well, if needed.) */
+ maybe_update_execution_strings (&input_text, new_len);
+ }
+ else if (new_len > input_text_length + 1)
+ /* Don't bother to realloc if we have enough space. */
+ input_text = xrealloc (input_text, new_len);
+
+ memmove (input_text + from + xp_len,
+ input_text + *to, input_text_length - *to + 1);
+
+ *to += xp_len - raw_len;
+ /* Since we change input_text_length here, the comparison above
+ isn't really valid, but it seems the worst that might happen is
+ an extra xrealloc or two, so let's not worry. */
+ input_text_length += xp_len - raw_len;
+ }
+ memcpy (input_text + from, xp, xp_len);
+ free (xp);
+
+ /* Synchronize the macro-expansion pointers with our new input_text. */
+ if (input_text != old_input)
+ forget_itext (old_input);
+ if (macro_expansion_output_stream && !executing_string)
+ remember_itext (input_text, from);
+}
+
+/* Read characters from the file until we are at MATCH or end of line.
+ Place the characters read into STRING. If EXPAND is nonzero,
+ expand the text before looking for MATCH for those cases where
+ MATCH might be produced by some macro. */
+void
+get_until_in_line (int expand, char *match, char **string)
+{
+ int real_bottom = input_text_length;
+ int limit = search_forward ("\n", input_text_offset);
+ if (limit < 0)
+ limit = input_text_length;
+
+ /* Replace input_text[input_text_offset .. limit-1] with its expansion.
+ This allows the node names and menu entries themselves to be
+ constructed via a macro, as in:
+ @macro foo{p, q}
+ Together: \p\ & \q\.
+ @end macro
+
+ @node @foo{A,B}, next, prev, top
+
+ Otherwise, the `,' separating the macro args A and B is taken as
+ the node argument separator, so the node name is `@foo{A'. This
+ expansion is only necessary on the first call, since we expand the
+ whole line then. */
+ if (expand)
+ {
+ replace_with_expansion (input_text_offset, &limit);
+ }
+
+ real_bottom = input_text_length;
+ input_text_length = limit;
+ get_until (match, string);
+ input_text_length = real_bottom;
+}
+
+void
+get_rest_of_line (int expand, char **string)
+{
+ xml_no_para ++;
+ if (expand)
+ {
+ char *tem;
+
+ /* Don't expand non-macros in input, since we want them
+ intact in the macro-expanded output. */
+ only_macro_expansion++;
+ get_until_in_line (1, "\n", &tem);
+ only_macro_expansion--;
+ *string = expansion (tem, 0);
+ free (tem);
+ }
+ else
+ get_until_in_line (0, "\n", string);
+
+ canon_white (*string);
+
+ if (curchar () == '\n') /* as opposed to the end of the file... */
+ {
+ line_number++;
+ input_text_offset++;
+ }
+ xml_no_para --;
+}
+
+/* Backup the input pointer to the previous character, keeping track
+ of the current line number. */
+void
+backup_input_pointer (void)
+{
+ if (input_text_offset)
+ {
+ input_text_offset--;
+ if (curchar () == '\n')
+ line_number--;
+ }
+}
+
+/* Read characters from the file until we are at MATCH or closing brace.
+ Place the characters read into STRING. */
+void
+get_until_in_braces (char *match, char **string)
+{
+ char *temp;
+ int i, brace = 0;
+ int match_len = strlen (match);
+
+ for (i = input_text_offset; i < input_text_length; i++)
+ {
+ if (i < input_text_length - 1 && input_text[i] == '@')
+ {
+ i++; /* skip commands like @, and @{ */
+ continue;
+ }
+ else if (input_text[i] == '{')
+ brace++;
+ else if (input_text[i] == '}')
+ {
+ brace--;
+ /* If looking for a brace, don't stop at the interior brace,
+ like after "baz" in "@foo{something @bar{baz} more}". */
+ if (brace == 0)
+ continue;
+ }
+ else if (input_text[i] == '\n')
+ line_number++;
+
+ if (brace < 0 ||
+ (brace == 0 && strncmp (input_text + i, match, match_len) == 0))
+ break;
+ }
+
+ match_len = i - input_text_offset;
+ temp = xmalloc (2 + match_len);
+ memcpy (temp, input_text + input_text_offset, match_len);
+ temp[match_len] = 0;
+ input_text_offset = i;
+ *string = temp;
+}
+
+
+
+/* Converting a file. */
+
+/* Convert the file named by NAME. The output is saved on the file
+ named as the argument to the @setfilename command. */
+static char *suffixes[] = {
+ /* ".txi" is checked first so that on 8+3 DOS filesystems, if they
+ have "texinfo.txi" and "texinfo.tex" in the same directory, the
+ former is used rather than the latter, due to file name truncation. */
+ ".txi",
+ ".texinfo",
+ ".texi",
+ ".txinfo",
+ "",
+ NULL
+};
+
+static void
+initialize_conversion (void)
+{
+ init_tag_table ();
+ init_indices ();
+ init_internals ();
+ init_paragraph ();
+
+ /* This is used for splitting the output file and for doing section
+ headings. It was previously initialized in `init_paragraph', but its
+ use there loses with the `init_paragraph' calls done by the
+ multitable code; the tag indices get reset to zero. */
+ output_position = 0;
+}
+
+/* Reverse the chain of structures in LIST. Output the new head
+ of the chain. You should always assign the output value of this
+ function to something, or you will lose the chain. */
+GENERIC_LIST *
+reverse_list (GENERIC_LIST *list)
+{
+ GENERIC_LIST *next;
+ GENERIC_LIST *prev = NULL;
+
+ while (list)
+ {
+ next = list->next;
+ list->next = prev;
+ prev = list;
+ list = next;
+ }
+ return prev;
+}
+
+/* We read in multiples of 4k, simply because it is a typical pipe size
+ on unix systems. */
+#define READ_BUFFER_GROWTH (4 * 4096)
+
+/* Convert the Texinfo file coming from the open stream STREAM. Assume the
+ source of the stream is named NAME. */
+static void
+convert_from_stream (FILE *stream, char *name)
+{
+ char *buffer = NULL;
+ int buffer_offset = 0, buffer_size = 0;
+
+ initialize_conversion ();
+
+ /* Read until the end of the stream. This isn't strictly correct, since
+ the texinfo input may end before the stream ends, but it is a quick
+ working hueristic. */
+ while (!feof (stream))
+ {
+ int count;
+
+ if (buffer_offset + (READ_BUFFER_GROWTH + 1) >= buffer_size)
+ buffer = (char *)
+ xrealloc (buffer, (buffer_size += READ_BUFFER_GROWTH));
+
+ count = fread (buffer + buffer_offset, 1, READ_BUFFER_GROWTH, stream);
+
+ if (count < 0)
+ {
+ perror (name);
+ xexit (1);
+ }
+
+ buffer_offset += count;
+ if (count == 0)
+ break;
+ }
+
+ /* Set the globals to the new file. */
+ input_text = buffer;
+ input_text_length = buffer_offset;
+ input_filename = xstrdup (name);
+ node_filename = xstrdup (name);
+ input_text_offset = 0;
+ line_number = 1;
+
+ /* Not strictly necessary. This magic prevents read_token () from doing
+ extra unnecessary work each time it is called (that is a lot of times).
+ The INPUT_TEXT_LENGTH is one past the actual end of the text. */
+ input_text[input_text_length] = '\n';
+
+ convert_from_loaded_file (name);
+}
+
+static void
+convert_from_file (char *name)
+{
+ int i;
+ char *filename = xmalloc (strlen (name) + 50);
+
+ /* Prepend file directory to the search path, so relative links work. */
+ prepend_to_include_path (pathname_part (name));
+
+ initialize_conversion ();
+
+ /* Try to load the file specified by NAME, concatenated with our
+ various suffixes. Prefer files like `makeinfo.texi' to
+ `makeinfo'. */
+ for (i = 0; suffixes[i]; i++)
+ {
+ strcpy (filename, name);
+ strcat (filename, suffixes[i]);
+
+ if (find_and_load (filename, 1))
+ break;
+
+ if (!suffixes[i][0] && strrchr (filename, '.'))
+ {
+ fs_error (filename);
+ free (filename);
+ return;
+ }
+ }
+
+ if (!suffixes[i])
+ {
+ fs_error (name);
+ free (filename);
+ return;
+ }
+
+ input_filename = filename;
+
+ convert_from_loaded_file (name);
+
+ /* Pop the prepended path, so multiple filenames in the
+ command line do not screw each others include paths. */
+ pop_path_from_include_path ();
+}
+
+static int
+create_html_directory (char *dir, int can_remove_file)
+{
+ struct stat st;
+
+ /* Already exists. */
+ if (stat (dir, &st) == 0)
+ {
+ /* And it's a directory, so silently reuse it. */
+ if (S_ISDIR (st.st_mode))
+ return 1;
+ /* Not a directory, so move it out of the way if we are allowed. */
+ else if (can_remove_file)
+ {
+ if (unlink (dir) != 0)
+ return 0;
+ }
+ else
+ return 0;
+ }
+
+ if (mkdir (dir, 0777) == 0)
+ /* Success! */
+ return 1;
+ else
+ return 0;
+}
+
+/* Given OUTPUT_FILENAME == ``/foo/bar/baz.html'', return
+ "/foo/bar/baz/baz.html". This routine is called only if html && splitting.
+
+ Split html output goes into the subdirectory of the toplevel
+ filename, without extension. For example:
+ @setfilename foo.info
+ produces output in files foo/index.html, foo/second-node.html, ...
+
+ But if the user said -o foo.whatever on the cmd line, then use
+ foo.whatever unchanged. */
+
+static char *
+insert_toplevel_subdirectory (char *output_filename)
+{
+ static const char index_name[] = "index.html";
+ char *dir, *subdir, *base, *basename, *p;
+ char buf[PATH_MAX];
+ const int index_len = sizeof (index_name) - 1;
+
+ strcpy (buf, output_filename);
+ dir = pathname_part (buf); /* directory of output_filename */
+ base = filename_part (buf); /* strips suffix, too */
+ basename = xstrdup (base); /* remember real @setfilename name */
+ p = dir + strlen (dir) - 1;
+ if (p > dir && IS_SLASH (*p))
+ *p = 0;
+ p = strrchr (base, '.');
+ if (p)
+ *p = 0;
+
+ /* Split html output goes into subdirectory of toplevel name. */
+ if (save_command_output_filename
+ && STREQ (output_filename, save_command_output_filename))
+ subdir = basename; /* from user, use unchanged */
+ else
+ subdir = base; /* implicit, omit suffix */
+
+ free (output_filename);
+ output_filename = xmalloc (strlen (dir) + 1
+ + strlen (basename) + 1
+ + index_len
+ + 1);
+ strcpy (output_filename, dir);
+ if (strlen (dir))
+ strcat (output_filename, "/");
+ strcat (output_filename, subdir);
+
+ /* First try, do not remove existing file. */
+ if (!create_html_directory (output_filename, 0))
+ {
+ /* That failed, try subdir name with .html.
+ Remove it if it exists. */
+ strcpy (output_filename, dir);
+ if (strlen (dir))
+ strcat (output_filename, "/");
+ strcat (output_filename, basename);
+
+ if (!create_html_directory (output_filename, 1))
+ {
+ /* Last try failed too :-\ */
+ line_error (_("Can't create directory `%s': %s"),
+ output_filename, strerror (errno));
+ xexit (1);
+ }
+ }
+
+ strcat (output_filename, "/");
+ strcat (output_filename, index_name);
+ return output_filename;
+}
+
+/* FIXME: this is way too hairy */
+static void
+convert_from_loaded_file (char *name)
+{
+ char *real_output_filename = NULL;
+
+ remember_itext (input_text, 0);
+
+ input_text_offset = 0;
+
+ /* Avoid the `\input texinfo' line in HTML output (assuming it starts
+ the file). */
+ if (looking_at ("\\input"))
+ discard_until ("\n");
+
+ /* Search this file looking for the special string which starts conversion.
+ Once found, we may truly begin. */
+ while (input_text_offset >= 0)
+ {
+ input_text_offset =
+ search_forward (setfilename_search, input_text_offset);
+
+ if (input_text_offset == 0
+ || (input_text_offset > 0
+ && input_text[input_text_offset -1] == '\n'))
+ break;
+ else if (input_text_offset > 0)
+ input_text_offset++;
+ }
+
+ if (input_text_offset < 0)
+ {
+ if (!command_output_filename)
+ {
+#if defined (REQUIRE_SETFILENAME)
+ error (_("No `%s' found in `%s'"), setfilename_search, name);
+ goto finished;
+#else
+ command_output_filename = output_name_from_input_name (name);
+#endif /* !REQUIRE_SETFILENAME */
+ }
+
+ {
+ int i, end_of_first_line;
+
+ /* Find the end of the first line in the file. */
+ for (i = 0; i < input_text_length - 1; i++)
+ if (input_text[i] == '\n')
+ break;
+
+ end_of_first_line = i + 1;
+
+ for (i = 0; i < end_of_first_line; i++)
+ {
+ if ((input_text[i] == '\\') &&
+ (strncmp (input_text + i + 1, "input", 5) == 0))
+ {
+ input_text_offset = i;
+ break;
+ }
+ }
+ }
+ }
+ else
+ input_text_offset += strlen (setfilename_search);
+
+ if (!command_output_filename)
+ {
+ get_until ("\n", &output_filename); /* read rest of line */
+ if (html || xml)
+ { /* Change any extension to .html or .xml. */
+ char *html_name, *directory_part, *basename_part, *temp;
+
+ canon_white (output_filename);
+ directory_part = pathname_part (output_filename);
+
+ basename_part = filename_part (output_filename);
+
+ /* Zap any existing extension. */
+ temp = strrchr (basename_part, '.');
+ if (temp)
+ *temp = 0;
+
+ /* Construct new filename. */
+ html_name = xmalloc (strlen (directory_part)
+ + strlen (basename_part) + 6);
+ strcpy (html_name, directory_part);
+ strcat (html_name, basename_part);
+ strcat (html_name, html ? ".html" : ".xml");
+
+ /* Replace name from @setfilename with the html name. */
+ free (output_filename);
+ output_filename = html_name;
+ }
+ }
+ else
+ {
+ if (input_text_offset != -1)
+ discard_until ("\n");
+ else
+ input_text_offset = 0;
+
+ real_output_filename = output_filename = command_output_filename;
+ command_output_filename = NULL; /* for included files or whatever */
+ }
+
+ canon_white (output_filename);
+ toplevel_output_filename = xstrdup (output_filename);
+
+ if (real_output_filename && strcmp (real_output_filename, "-") == 0)
+ {
+ if (macro_expansion_filename
+ && strcmp (macro_expansion_filename, "-") == 0)
+ {
+ fprintf (stderr,
+ _("%s: Skipping macro expansion to stdout as Info output is going there.\n"),
+ progname);
+ macro_expansion_output_stream = NULL;
+ }
+ real_output_filename = xstrdup (real_output_filename);
+ output_stream = stdout;
+ splitting = 0; /* Cannot split when writing to stdout. */
+ }
+ else
+ {
+ if (html && splitting)
+ {
+ if (FILENAME_CMP (output_filename, NULL_DEVICE) == 0
+ || FILENAME_CMP (output_filename, ALSO_NULL_DEVICE) == 0)
+ splitting = 0;
+ else
+ output_filename = insert_toplevel_subdirectory (output_filename);
+ real_output_filename = xstrdup (output_filename);
+ }
+ else if (!real_output_filename)
+ real_output_filename = expand_filename (output_filename, name);
+ else
+ real_output_filename = xstrdup (real_output_filename);
+
+ output_stream = fopen (real_output_filename, "w");
+ }
+
+ set_current_output_filename (real_output_filename);
+
+ if (xml && !docbook)
+ xml_begin_document (filename_part (output_filename));
+
+ if (verbose_mode)
+ printf (_("Making %s file `%s' from `%s'.\n"),
+ no_headers ? "text"
+ : html ? "HTML"
+ : xml ? "XML"
+ : "info",
+ output_filename, input_filename);
+
+ if (output_stream == NULL)
+ {
+ fs_error (real_output_filename);
+ goto finished;
+ }
+
+ /* Make the displayable filename from output_filename. Only the base
+ portion of the filename need be displayed. */
+ flush_output (); /* in case there was no @bye */
+ if (output_stream != stdout)
+ pretty_output_filename = filename_part (output_filename);
+ else
+ pretty_output_filename = xstrdup ("stdout");
+
+ /* For this file only, count the number of newlines from the top of
+ the file to here. This way, we keep track of line numbers for
+ error reporting. Line_number starts at 1, since the user isn't
+ zero-based. */
+ {
+ int temp = 0;
+ line_number = 1;
+ while (temp != input_text_offset)
+ if (input_text[temp++] == '\n')
+ line_number++;
+ }
+
+ /* html fixxme: should output this as trailer on first page. */
+ if (!no_headers && !html && !xml)
+ add_word_args (_("This is %s, produced by makeinfo version %s from %s.\n"),
+ output_filename, VERSION, input_filename);
+
+ close_paragraph ();
+
+ if (xml && !docbook)
+ {
+ /* Just before the real main loop, let's handle the defines. */
+ COMMAND_LINE_DEFINE *temp;
+
+ for (temp = command_line_defines; temp; temp = temp->next)
+ {
+ handle_variable_internal (temp->action, temp->define);
+ free(temp->define);
+ }
+ }
+
+ reader_loop ();
+ if (xml)
+ xml_end_document ();
+
+
+finished:
+ discard_insertions (0);
+ close_paragraph ();
+ flush_file_stack ();
+
+ if (macro_expansion_output_stream)
+ {
+ fclose (macro_expansion_output_stream);
+ if (errors_printed && !force
+ && strcmp (macro_expansion_filename, "-") != 0
+ && FILENAME_CMP (macro_expansion_filename, NULL_DEVICE) != 0
+ && FILENAME_CMP (macro_expansion_filename, ALSO_NULL_DEVICE) != 0)
+ {
+ fprintf (stderr,
+_("%s: Removing macro output file `%s' due to errors; use --force to preserve.\n"),
+ progname, macro_expansion_filename);
+ if (unlink (macro_expansion_filename) < 0)
+ perror (macro_expansion_filename);
+ }
+ }
+
+ if (output_stream)
+ {
+ output_pending_notes ();
+
+ if (html)
+ {
+ no_indent = 1;
+ start_paragraph ();
+ add_word ("</body></html>\n");
+ close_paragraph ();
+ }
+
+ /* maybe we want local variables in info output. */
+ {
+ char *trailer = info_trailer ();
+ if (!xml && !docbook && trailer)
+ {
+ if (html)
+ insert_string ("<!--");
+ insert_string (trailer);
+ free (trailer);
+ if (html)
+ insert_string ("\n-->\n");
+ }
+ }
+
+ /* Write stuff makeinfo generates after @bye, ie. info_trailer. */
+ flush_output ();
+
+ if (output_stream != stdout)
+ fclose (output_stream);
+
+ /* If validating, then validate the entire file right now. */
+ if (validating)
+ validate_file (tag_table);
+
+ handle_delayed_writes ();
+
+ if (tag_table)
+ {
+ tag_table = (TAG_ENTRY *) reverse_list ((GENERIC_LIST *) tag_table);
+ if (!no_headers && !html && !STREQ (current_output_filename, "-"))
+ write_tag_table (real_output_filename);
+ }
+
+ if (splitting && !html && (!errors_printed || force))
+ {
+ clean_old_split_files (real_output_filename);
+ split_file (real_output_filename, split_size);
+ }
+ else if (errors_printed
+ && !force
+ && strcmp (real_output_filename, "-") != 0
+ && FILENAME_CMP (real_output_filename, NULL_DEVICE) != 0
+ && FILENAME_CMP (real_output_filename, ALSO_NULL_DEVICE) != 0)
+ { /* If there were errors, and no --force, remove the output. */
+ fprintf (stderr,
+ _("%s: Removing output file `%s' due to errors; use --force to preserve.\n"),
+ progname, real_output_filename);
+ if (unlink (real_output_filename) < 0)
+ perror (real_output_filename);
+ }
+ }
+ free (real_output_filename);
+}
+
+/* If enable_encoding is set and @documentencoding is used, return a
+ Local Variables section (as a malloc-ed string) so that Emacs'
+ locale features can work. Else return NULL. */
+char *
+info_trailer (void)
+{
+ char *encoding;
+
+ if (!enable_encoding)
+ return NULL;
+
+ encoding = current_document_encoding ();
+
+ if (encoding && *encoding)
+ {
+#define LV_FMT "\n\037\nLocal Variables:\ncoding: %s\nEnd:\n"
+ char *lv = xmalloc (sizeof (LV_FMT) + strlen (encoding));
+ sprintf (lv, LV_FMT, encoding);
+ free (encoding);
+ return lv;
+ }
+
+ free (encoding);
+ return NULL;
+}
+
+void
+free_and_clear (char **pointer)
+{
+ if (*pointer)
+ {
+ free (*pointer);
+ *pointer = NULL;
+ }
+}
+
+ /* Initialize some state. */
+static void
+init_internals (void)
+{
+ free_and_clear (&output_filename);
+ free_and_clear (&command);
+ free_and_clear (&input_filename);
+ free_node_references ();
+ free_node_node_references ();
+ toc_free ();
+ init_insertion_stack ();
+ init_brace_stack ();
+ current_node = NULL; /* sometimes already freed */
+ command_index = 0;
+ in_menu = 0;
+ in_detailmenu = 0;
+ top_node_seen = 0;
+ non_top_node_seen = 0;
+ node_number = -1;
+}
+
+void
+init_paragraph (void)
+{
+ free (output_paragraph);
+ output_paragraph = xmalloc (paragraph_buffer_len);
+ output_paragraph[0] = 0;
+ output_paragraph_offset = 0;
+ output_column = 0;
+ paragraph_is_open = 0;
+ current_indent = 0;
+ meta_char_pos = 0;
+}
+
+/* This is called from `reader_loop' when we are at the * beginning a
+ menu line. */
+
+static void
+handle_menu_entry (void)
+{
+ char *tem;
+
+ /* Ugh, glean_node_from_menu wants to read the * itself. */
+ input_text_offset--;
+
+ /* Find node name in menu entry and save it in references list for
+ later validation. Use followed_reference type for detailmenu
+ references since we don't want to use them for default node pointers. */
+ tem = glean_node_from_menu (1, in_detailmenu
+ ? followed_reference : menu_reference);
+
+ if (html && tem)
+ { /* Start a menu item with the cleaned-up line. Put an anchor
+ around the start text (before `:' or the node name). */
+ char *string;
+
+ discard_until ("* ");
+
+ /* The line number was already incremented in reader_loop when we
+ saw the newline, and discard_until has now incremented again. */
+ line_number--;
+
+ if (had_menu_commentary)
+ {
+ add_html_block_elt ("<ul class=\"menu\">\n");
+ had_menu_commentary = 0;
+ in_paragraph = 0;
+ }
+
+ if (in_paragraph)
+ {
+ add_html_block_elt ("</p>\n");
+ add_html_block_elt ("<ul class=\"menu\">\n");
+ in_paragraph = 0;
+ }
+
+ in_menu_item = 1;
+
+ add_html_block_elt ("<li><a");
+ if (next_menu_item_number <= 9)
+ {
+ add_word(" accesskey=");
+ add_word_args("\"%d\"", next_menu_item_number);
+ next_menu_item_number++;
+ }
+ add_word (" href=\"");
+ string = expansion (tem, 0);
+ add_anchor_name (string, 1);
+ add_word ("\">");
+ free (string);
+
+ /* The menu item may use macros, so expand them now. */
+ only_macro_expansion++;
+ get_until_in_line (1, ":", &string);
+ only_macro_expansion--;
+ execute_string ("%s", string); /* get escaping done */
+ free (string);
+
+ add_word ("</a>");
+
+ if (looking_at ("::"))
+ discard_until (":");
+ else
+ { /* discard the node name */
+ get_until_in_line (0, ".", &string);
+ free (string);
+ }
+ input_text_offset++; /* discard the second colon or the period */
+
+ /* Insert a colon only if there is a description of this menu item. */
+ {
+ int save_input_text_offset = input_text_offset;
+ int save_line_number = line_number;
+ char *test_string;
+ get_rest_of_line (0, &test_string);
+ if (strlen (test_string) > 0)
+ add_word (": ");
+ input_text_offset = save_input_text_offset;
+ line_number = save_line_number;
+ }
+ }
+ else if (xml && tem)
+ {
+ xml_start_menu_entry (tem);
+ }
+ else if (tem)
+ { /* For Info output, we can just use the input and the main case in
+ reader_loop where we output what comes in. Just move off the *
+ so the next time through reader_loop we don't end up back here. */
+ add_char ('*');
+ input_text_offset += 2; /* undo the pointer back-up above. */
+ }
+
+ if (tem)
+ free (tem);
+}
+
+/* Find the command corresponding to STRING. If the command is found,
+ return a pointer to the data structure. Otherwise return -1. */
+static COMMAND *
+get_command_entry (char *string)
+{
+ int i;
+
+ for (i = 0; command_table[i].name; i++)
+ if (strcmp (command_table[i].name, string) == 0)
+ return &command_table[i];
+
+ /* This command is not in our predefined command table. Perhaps
+ it is a user defined command. */
+ for (i = 0; i < user_command_array_len; i++)
+ if (user_command_array[i] &&
+ (strcmp (user_command_array[i]->name, string) == 0))
+ return user_command_array[i];
+
+ /* We never heard of this command. */
+ return (COMMAND *) -1;
+}
+
+/* input_text_offset is right at the command prefix character.
+ Read the next token to determine what to do. Return zero
+ if there's no known command or macro after the prefix character. */
+static int
+read_command (void)
+{
+ COMMAND *entry;
+ int old_text_offset = input_text_offset++;
+
+ free_and_clear (&command);
+ command = read_token ();
+
+ /* Check to see if this command is a macro. If so, execute it here. */
+ {
+ MACRO_DEF *def;
+
+ def = find_macro (command);
+
+ if (def)
+ {
+ /* We disallow recursive use of a macro call. Inhibit the expansion
+ of this macro during the life of its execution. */
+ if (!(def->flags & ME_RECURSE))
+ def->inhibited = 1;
+
+ executing_macro++;
+ execute_macro (def);
+ executing_macro--;
+
+ if (!(def->flags & ME_RECURSE))
+ def->inhibited = 0;
+
+ return 1;
+ }
+ }
+
+ if (only_macro_expansion)
+ {
+ /* Back up to the place where we were called, so the
+ caller will have a chance to process this non-macro. */
+ input_text_offset = old_text_offset;
+ return 0;
+ }
+
+ /* Perform alias expansion */
+ command = alias_expand (command);
+
+ if (enclosure_command (command))
+ {
+ remember_brace (enclosure_expand);
+ enclosure_expand (START, output_paragraph_offset, 0);
+ return 0;
+ }
+
+ entry = get_command_entry (command);
+ if (entry == (COMMAND *)-1)
+ {
+ line_error (_("Unknown command `%s'"), command);
+ return 0;
+ }
+
+ if (entry->argument_in_braces == BRACE_ARGS)
+ remember_brace (entry->proc);
+ else if (entry->argument_in_braces == MAYBE_BRACE_ARGS)
+ {
+ if (curchar () == '{')
+ remember_brace (entry->proc);
+ else
+ { /* No braces, so arg is next char. */
+ int ch;
+ int saved_offset = output_paragraph_offset;
+ (*(entry->proc)) (START, output_paragraph_offset, 0);
+
+ /* Possibilities left for the next character: @ (error), }
+ (error), whitespace (skip) anything else (normal char). */
+ skip_whitespace ();
+ ch = curchar ();
+ if (ch == '@')
+ {
+ line_error (_("Use braces to give a command as an argument to @%s"),
+ entry->name);
+ return 0;
+ }
+ else if (ch == '}')
+ {
+ /* Our caller will give the error message, because this }
+ won't match anything. */
+ return 0;
+ }
+
+ add_char (ch);
+ input_text_offset++;
+ (*(entry->proc)) (END, saved_offset, output_paragraph_offset);
+ return 1;
+ }
+ }
+
+ /* Get here if we have BRACE_ARGS, NO_BRACE_ARGS, or MAYBE_BRACE_ARGS
+ with braces. */
+ (*(entry->proc)) (START, output_paragraph_offset, 0);
+ return 1;
+}
+
+/* Okay, we are ready to start the conversion. Call the reader on
+ some text, and fill the text as it is output. Handle commands by
+ remembering things like open braces and the current file position on a
+ stack, and when the corresponding close brace is found, you can call
+ the function with the proper arguments. Although the filling isn't
+ necessary for HTML, it should do no harm. */
+void
+reader_loop (void)
+{
+ int character;
+ int done = 0;
+
+ while (!done)
+ {
+ if (input_text_offset >= input_text_length)
+ break;
+
+ character = curchar ();
+
+ /* If only_macro_expansion, only handle macros and leave
+ everything else intact. */
+ if (!only_macro_expansion && !in_fixed_width_font
+ && ((!html && !xml) || escape_html)
+ && (character == '\'' || character == '`')
+ && input_text[input_text_offset + 1] == character)
+ {
+ if (html)
+ {
+ input_text_offset += 2;
+ add_word (character == '`' ? "&ldquo;" : "&rdquo;");
+ continue;
+ }
+ else if (xml)
+ {
+ input_text_offset += 2;
+ xml_insert_entity (character == '`' ? "ldquo" : "rdquo");
+ continue;
+ }
+ else
+ {
+ input_text_offset++;
+ character = '"';
+ }
+ }
+
+ /* Convert --- to --. */
+ if (!only_macro_expansion && character == '-' && !in_fixed_width_font
+ && ((!html && !xml) || escape_html))
+ {
+ int dash_count = 0;
+
+ /* Get the number of consequtive dashes. */
+ while (input_text[input_text_offset] == '-')
+ {
+ dash_count++;
+ input_text_offset++;
+ }
+
+ /* Eat one dash. */
+ dash_count--;
+
+ if (html || xml)
+ {
+ if (dash_count == 0)
+ add_char ('-');
+ else
+ while (dash_count > 0)
+ {
+ if (dash_count >= 2)
+ {
+ if (html)
+ add_word ("&mdash;");
+ else
+ xml_insert_entity ("mdash");
+ dash_count -= 2;
+ }
+ else if (dash_count >= 1)
+ {
+ if (html)
+ add_word ("&ndash;");
+ else
+ xml_insert_entity ("ndash");
+ dash_count--;
+ }
+ }
+ }
+ else
+ {
+ add_char ('-');
+ while (--dash_count > 0)
+ add_char ('-');
+ }
+
+ continue;
+ }
+
+ /* If this is a whitespace character, then check to see if the line
+ is blank. If so, advance to the carriage return. */
+ if (!only_macro_expansion && whitespace (character))
+ {
+ int i = input_text_offset + 1;
+
+ while (i < input_text_length && whitespace (input_text[i]))
+ i++;
+
+ if (i == input_text_length || input_text[i] == '\n')
+ {
+ if (i == input_text_length)
+ i--;
+
+ input_text_offset = i;
+ character = curchar ();
+ }
+ }
+
+ if (character == '\n')
+ line_number++;
+
+ switch (character)
+ {
+ case '*': /* perhaps we are at a menu */
+ /* We used to check for this in the \n case but an @c in a
+ menu swallows its newline, so check here instead. */
+ if (!only_macro_expansion && in_menu
+ && input_text_offset + 1 < input_text_length
+ && input_text[input_text_offset-1] == '\n')
+ handle_menu_entry ();
+ else
+ { /* Duplicate code from below, but not worth twisting the
+ fallthroughs to get down there. */
+ add_char (character);
+ input_text_offset++;
+ }
+ break;
+
+ /* Escapes for HTML unless we're outputting raw HTML. Do
+ this always, even if SGML rules don't require it since
+ that's easier and safer for non-conforming browsers. */
+ case '&':
+ if (html && escape_html)
+ add_word ("&amp;");
+ else
+ add_char (character);
+ input_text_offset++;
+ break;
+
+ case '<':
+ if (html && escape_html)
+ add_word ("&lt;");
+ else if (xml && escape_html)
+ xml_insert_entity ("lt");
+ else
+ add_char (character);
+ input_text_offset++;
+ break;
+
+ case '>':
+ if (html && escape_html)
+ add_word ("&gt;");
+ else if (xml && escape_html)
+ xml_insert_entity ("gt");
+ else
+ add_char (character);
+ input_text_offset++;
+ break;
+
+ case COMMAND_PREFIX: /* @ */
+ if (read_command () || !only_macro_expansion)
+ break;
+
+ /* FALLTHROUGH (usually) */
+ case '{':
+ /* Special case. We're not supposed to see this character by itself.
+ If we do, it means there is a syntax error in the input text.
+ Report the error here, but remember this brace on the stack so
+ we can ignore its partner. */
+ if (!only_macro_expansion)
+ {
+ if (command && !STREQ (command, "math"))
+ {
+ line_error (_("Misplaced %c"), '{');
+ remember_brace (misplaced_brace);
+ }
+ else
+ /* We don't mind `extra' braces inside @math. */
+ remember_brace (cm_no_op);
+ /* remember_brace advances input_text_offset. */
+ break;
+ }
+
+ /* FALLTHROUGH (usually) */
+ case '}':
+ if (!only_macro_expansion)
+ {
+ pop_and_call_brace ();
+ input_text_offset++;
+ break;
+ }
+
+ /* FALLTHROUGH (usually) */
+ default:
+ add_char (character);
+ input_text_offset++;
+ }
+ }
+ if (macro_expansion_output_stream && !only_macro_expansion)
+ maybe_write_itext (input_text, input_text_offset);
+}
+
+static void
+init_brace_stack (void)
+{
+ brace_stack = NULL;
+}
+
+/* Remember the current output position here. Save PROC
+ along with it so you can call it later. */
+static void
+remember_brace_1 (COMMAND_FUNCTION (*proc), int position)
+{
+ BRACE_ELEMENT *new = xmalloc (sizeof (BRACE_ELEMENT));
+ new->next = brace_stack;
+ new->proc = proc;
+ new->command = command ? xstrdup (command) : "";
+ new->pos = position;
+ new->line = line_number;
+ new->in_fixed_width_font = in_fixed_width_font;
+ brace_stack = new;
+}
+
+static void
+remember_brace (COMMAND_FUNCTION (*proc))
+{
+ if (curchar () != '{')
+ line_error (_("%c%s expected braces"), COMMAND_PREFIX, command);
+ else
+ input_text_offset++;
+ remember_brace_1 (proc, output_paragraph_offset);
+}
+
+/* Pop the top of the brace stack, and call the associated function
+ with the args END and POS. */
+static void
+pop_and_call_brace (void)
+{
+ if (brace_stack == NULL)
+ {
+ line_error (_("Unmatched }"));
+ return;
+ }
+
+ {
+ BRACE_ELEMENT *temp;
+
+ int pos = brace_stack->pos;
+ COMMAND_FUNCTION *proc = brace_stack->proc;
+ in_fixed_width_font = brace_stack->in_fixed_width_font;
+
+ /* Reset current command, so the proc can know who it is. This is
+ used in cm_accent. */
+ command = brace_stack->command;
+
+ temp = brace_stack->next;
+ free (brace_stack);
+ brace_stack = temp;
+
+ (*proc) (END, pos, output_paragraph_offset);
+ }
+}
+
+/* Shift all of the markers in `brace_stack' by AMOUNT. */
+static void
+adjust_braces_following (int here, int amount)
+{
+ BRACE_ELEMENT *stack = brace_stack;
+
+ while (stack)
+ {
+ if (stack->pos >= here)
+ stack->pos += amount;
+ stack = stack->next;
+ }
+}
+
+/* Return the string which invokes PROC; a pointer to a function.
+ Always returns the first function in the command table if more than
+ one matches PROC. */
+static const char *
+find_proc_name (COMMAND_FUNCTION (*proc))
+{
+ int i;
+
+ for (i = 0; command_table[i].name; i++)
+ if (proc == command_table[i].proc)
+ return command_table[i].name;
+ return _("NO_NAME!");
+}
+
+/* You call discard_braces () when you shouldn't have any braces on the stack.
+ I used to think that this happens for commands that don't take arguments
+ in braces, but that was wrong because of things like @code{foo @@}. So now
+ I only detect it at the beginning of nodes. */
+void
+discard_braces (void)
+{
+ if (!brace_stack)
+ return;
+
+ while (brace_stack)
+ {
+ if (brace_stack->proc != misplaced_brace)
+ {
+ const char *proc_name;
+
+ proc_name = find_proc_name (brace_stack->proc);
+ file_line_error (input_filename, brace_stack->line,
+ _("%c%s missing close brace"), COMMAND_PREFIX,
+ proc_name);
+ pop_and_call_brace ();
+ }
+ else
+ {
+ BRACE_ELEMENT *temp;
+ temp = brace_stack->next;
+ free (brace_stack);
+ brace_stack = temp;
+ }
+ }
+}
+
+static int
+get_char_len (int character)
+{
+ /* Return the printed length of the character. */
+ int len;
+
+ switch (character)
+ {
+ case '\t':
+ len = (output_column + 8) & 0xf7;
+ if (len > fill_column)
+ len = fill_column - output_column;
+ else
+ len = len - output_column;
+ break;
+
+ case '\n':
+ len = fill_column - output_column;
+ break;
+
+ default:
+ /* ASCII control characters appear as two characters in the output
+ (e.g., ^A). But characters with the high bit set are just one
+ on suitable terminals, so don't count them as two for line
+ breaking purposes. */
+ if (0 <= character && character < ' ')
+ len = 2;
+ else
+ len = 1;
+ }
+ return len;
+}
+
+void
+#if defined (VA_FPRINTF) && __STDC__
+add_word_args (const char *format, ...)
+#else
+add_word_args (format, va_alist)
+ const char *format;
+ va_dcl
+#endif
+{
+ char buffer[2000]; /* xx no fixed limits */
+#ifdef VA_FPRINTF
+ va_list ap;
+#endif
+
+ VA_START (ap, format);
+#ifdef VA_SPRINTF
+ VA_SPRINTF (buffer, format, ap);
+#else
+ sprintf (buffer, format, a1, a2, a3, a4, a5, a6, a7, a8);
+#endif /* not VA_SPRINTF */
+ va_end (ap);
+ add_word (buffer);
+}
+
+/* Add STRING to output_paragraph. */
+void
+add_word (char *string)
+{
+ while (*string)
+ add_char (*string++);
+}
+
+/* Like add_word, but inhibits conversion of whitespace into &nbsp;.
+ Use this to output HTML directives with embedded blanks, to make
+ them @w-safe. */
+void
+add_html_elt (char *string)
+{
+ in_html_elt++;
+ add_word (string);
+ in_html_elt--;
+}
+
+/* These two functions below, add_html_block_elt and add_html_block_elt_args,
+ are mixtures of add_html_elt and add_word_args. They inform makeinfo that
+ the current HTML element being inserted should not be enclosed in a <p>
+ element. */
+void
+add_html_block_elt (char *string)
+{
+ in_html_block_level_elt++;
+ add_word (string);
+ in_html_block_level_elt--;
+}
+
+void
+#if defined (VA_FPRINTF) && __STDC__
+add_html_block_elt_args (const char *format, ...)
+#else
+add_html_block_elt_args (format, va_alist)
+ const char *format;
+ va_dcl
+#endif
+{
+ char buffer[2000]; /* xx no fixed limits */
+#ifdef VA_FPRINTF
+ va_list ap;
+#endif
+
+ VA_START (ap, format);
+#ifdef VA_SPRINTF
+ VA_SPRINTF (buffer, format, ap);
+#else
+ sprintf (buffer, format, a1, a2, a3, a4, a5, a6, a7, a8);
+#endif /* not VA_SPRINTF */
+ va_end (ap);
+ add_html_block_elt (buffer);
+}
+
+/* Here is another awful kludge, used in add_char. Ordinarily, macro
+ expansions take place in the body of the document, and therefore we
+ should html_output_head when we see one. But there's an exception: a
+ macro call might take place within @copying, and that does not start
+ the real output, even though we fully expand the copying text.
+
+ So we need to be able to check if we are defining the @copying text.
+ We do this by looking back through the insertion stack. */
+static int
+defining_copying (void)
+{
+ INSERTION_ELT *i;
+ for (i = insertion_stack; i; i = i->next)
+ {
+ if (i->insertion == copying)
+ return 1;
+ }
+ return 0;
+}
+
+
+/* Add the character to the current paragraph. If filling_enabled is
+ nonzero, then do filling as well. */
+void
+add_char (int character)
+{
+ if (xml)
+ {
+ xml_add_char (character);
+ return;
+ }
+
+ /* If we are avoiding outputting headers, and we are currently
+ in a menu, then simply return. But if we're only expanding macros,
+ then we're being called from glean_node_from_menu to try to
+ remember a menu reference, and we need that so we can do defaulting. */
+ if (no_headers && !only_macro_expansion && (in_menu || in_detailmenu))
+ return;
+
+ /* If we are adding a character now, then we don't have to
+ ignore close_paragraph () calls any more. */
+ if (must_start_paragraph && character != '\n')
+ {
+ must_start_paragraph = 0;
+ line_already_broken = 0; /* The line is no longer broken. */
+ if (current_indent > output_column)
+ {
+ indent (current_indent - output_column);
+ output_column = current_indent;
+ }
+ }
+
+ if (non_splitting_words
+ && !(html && in_html_elt)
+ && strchr (" \t\n", character))
+ {
+ if (html || docbook)
+ { /* Seems cleaner to use &nbsp; than an 8-bit char. */
+ int saved_escape_html = escape_html;
+ escape_html = 0;
+ add_word ("&nbsp");
+ escape_html = saved_escape_html;
+ character = ';';
+ }
+ else
+ character = META (' '); /* unmeta-d in flush_output */
+ }
+
+ insertion_paragraph_closed = 0;
+
+ switch (character)
+ {
+ case '\n':
+ if (!filling_enabled && !(html && (in_menu || in_detailmenu)))
+ {
+ insert ('\n');
+
+ if (force_flush_right)
+ {
+ close_paragraph ();
+ /* Hack to force single blank lines out in this mode. */
+ flush_output ();
+ }
+
+ output_column = 0;
+
+ if (!no_indent && paragraph_is_open)
+ indent (output_column = current_indent);
+ break;
+ }
+ else if (end_of_sentence_p ())
+ /* CHARACTER is newline, and filling is enabled. */
+ {
+ insert (' ');
+ output_column++;
+ last_inserted_character = character;
+ }
+
+ if (last_char_was_newline)
+ {
+ if (html)
+ last_char_was_newline++;
+ close_paragraph ();
+ pending_indent = 0;
+ }
+ else
+ {
+ last_char_was_newline = 1;
+ if (html)
+ insert ('\n');
+ else
+ insert (' ');
+ output_column++;
+ }
+ break;
+
+ default: /* not at newline */
+ {
+ int len = get_char_len (character);
+ int suppress_insert = 0;
+
+ if ((character == ' ') && (last_char_was_newline))
+ {
+ if (!paragraph_is_open)
+ {
+ pending_indent++;
+ return;
+ }
+ }
+
+ /* This is sad, but it seems desirable to not force any
+ particular order on the front matter commands. This way,
+ the document can do @settitle, @documentlanguage, etc, in
+ any order and with any omissions, and we'll still output
+ the html <head> `just in time'. */
+ if ((executing_macro || !executing_string)
+ && !only_macro_expansion
+ && html && !html_output_head_p && !defining_copying ())
+ html_output_head ();
+
+ if (!paragraph_is_open)
+ {
+ start_paragraph ();
+ /* If the paragraph is supposed to be indented a certain
+ way, then discard all of the pending whitespace.
+ Otherwise, we let the whitespace stay. */
+ if (!paragraph_start_indent)
+ indent (pending_indent);
+ pending_indent = 0;
+
+ /* This check for in_html_block_level_elt prevents <p> from being
+ inserted when we already have html markup starting a paragraph,
+ as with <ul> and <h1> and the like. */
+ if (html && !in_html_block_level_elt)
+ {
+ if ((in_menu || in_detailmenu) && in_menu_item)
+ {
+ insert_string ("</li></ul>\n");
+ in_menu_item = 0;
+ }
+ insert_string ("<p>");
+ in_paragraph = 1;
+ adjust_braces_following (0, 3); /* adjust for <p> */
+ }
+ }
+
+ output_column += len;
+ if (output_column > fill_column)
+ {
+ if (filling_enabled && !html)
+ {
+ int temp = output_paragraph_offset;
+ while (--temp > 0 && output_paragraph[temp] != '\n')
+ {
+ /* If we have found a space, we have the place to break
+ the line. */
+ if (output_paragraph[temp] == ' ')
+ {
+ /* Remove trailing whitespace from output. */
+ while (temp && whitespace (output_paragraph[temp - 1]))
+ temp--;
+
+ /* If we went back all the way to the newline of the
+ preceding line, it probably means that the word we
+ are adding is itself wider than the space that the
+ indentation and the fill_column let us use. In
+ that case, do NOT insert another newline, since it
+ won't help. Just indent to current_indent and
+ leave it alone, since that's the most we can do. */
+ if (temp && output_paragraph[temp - 1] != '\n')
+ output_paragraph[temp++] = '\n';
+
+ /* We have correctly broken the line where we want
+ to. What we don't want is spaces following where
+ we have decided to break the line. We get rid of
+ them. */
+ {
+ int t1 = temp;
+
+ for (;; t1++)
+ {
+ if (t1 == output_paragraph_offset)
+ {
+ if (whitespace (character))
+ suppress_insert = 1;
+ break;
+ }
+ if (!whitespace (output_paragraph[t1]))
+ break;
+ }
+
+ if (t1 != temp)
+ {
+ adjust_braces_following (temp, (- (t1 - temp)));
+ memmove (&output_paragraph[temp],
+ &output_paragraph[t1],
+ output_paragraph_offset - t1);
+ output_paragraph_offset -= (t1 - temp);
+ }
+ }
+
+ /* Filled, but now indent if that is right. */
+ if (indented_fill && current_indent > 0)
+ {
+ int buffer_len = ((output_paragraph_offset - temp)
+ + current_indent);
+ char *temp_buffer = xmalloc (buffer_len);
+ int indentation = 0;
+
+ /* We have to shift any markers that are in
+ front of the wrap point. */
+ adjust_braces_following (temp, current_indent);
+
+ while (current_indent > 0 &&
+ indentation != current_indent)
+ temp_buffer[indentation++] = ' ';
+
+ memcpy ((char *) &temp_buffer[current_indent],
+ (char *) &output_paragraph[temp],
+ buffer_len - current_indent);
+
+ if (output_paragraph_offset + buffer_len
+ >= paragraph_buffer_len)
+ {
+ unsigned char *tt = xrealloc
+ (output_paragraph,
+ (paragraph_buffer_len += buffer_len));
+ output_paragraph = tt;
+ }
+ memcpy ((char *) &output_paragraph[temp],
+ temp_buffer, buffer_len);
+ output_paragraph_offset += current_indent;
+ free (temp_buffer);
+ }
+ output_column = 0;
+ while (temp < output_paragraph_offset)
+ output_column +=
+ get_char_len (output_paragraph[temp++]);
+ output_column += len;
+ break;
+ }
+ }
+ }
+ }
+
+ if (!suppress_insert)
+ {
+ insert (character);
+ last_inserted_character = character;
+ }
+ last_char_was_newline = 0;
+ line_already_broken = 0;
+ }
+ }
+}
+
+/* Add a character and store its position in meta_char_pos. */
+void
+add_meta_char (int character)
+{
+ meta_char_pos = output_paragraph_offset;
+ add_char (character);
+}
+
+/* Insert CHARACTER into `output_paragraph'. */
+void
+insert (int character)
+{
+ /* We don't want to strip trailing whitespace in multitables. Otherwise
+ horizontal separators confuse the font locking in Info mode in Emacs,
+ because it looks like a @subsection. Adding a trailing space to those
+ lines fixes it. */
+ if (character == '\n' && !html && !xml && !multitable_active)
+ {
+ while (output_paragraph_offset
+ && whitespace (output_paragraph[output_paragraph_offset-1]))
+ output_paragraph_offset--;
+ }
+
+ output_paragraph[output_paragraph_offset++] = character;
+ if (output_paragraph_offset == paragraph_buffer_len)
+ {
+ output_paragraph =
+ xrealloc (output_paragraph, (paragraph_buffer_len += 100));
+ }
+}
+
+/* Insert the null-terminated string STRING into `output_paragraph'. */
+void
+insert_string (const char *string)
+{
+ while (*string)
+ insert (*string++);
+}
+
+
+/* Sentences might have these characters after the period (or whatever). */
+#define POST_SENTENCE(c) ((c) == ')' || (c) == '\'' || (c) == '"' \
+ || (c) == ']')
+
+/* Return true if at an end-of-sentence character, possibly followed by
+ post-sentence punctuation to ignore. */
+static int
+end_of_sentence_p (void)
+{
+ int loc = output_paragraph_offset - 1;
+
+ /* If nothing has been output, don't check output_paragraph[-1]. */
+ if (loc < 0)
+ return 0;
+
+ /* A post-sentence character that is at meta_char_pos is not really
+ a post-sentence character; it was produced by a markup such as
+ @samp. We don't want the period inside @samp to be treated as a
+ sentence ender. */
+ while (loc > 0
+ && loc != meta_char_pos && POST_SENTENCE (output_paragraph[loc]))
+ loc--;
+ return loc != meta_char_pos && sentence_ender (output_paragraph[loc]);
+}
+
+
+/* Remove upto COUNT characters of whitespace from the
+ the current output line. If COUNT is less than zero,
+ then remove until none left. */
+void
+kill_self_indent (int count)
+{
+ /* Handle infinite case first. */
+ if (count < 0)
+ {
+ output_column = 0;
+ while (output_paragraph_offset)
+ {
+ if (whitespace (output_paragraph[output_paragraph_offset - 1]))
+ output_paragraph_offset--;
+ else
+ break;
+ }
+ }
+ else
+ {
+ while (output_paragraph_offset && count--)
+ if (whitespace (output_paragraph[output_paragraph_offset - 1]))
+ output_paragraph_offset--;
+ else
+ break;
+ }
+}
+
+/* Nonzero means do not honor calls to flush_output (). */
+static int flushing_ignored = 0;
+
+/* Prevent calls to flush_output () from having any effect. */
+void
+inhibit_output_flushing (void)
+{
+ flushing_ignored++;
+}
+
+/* Allow calls to flush_output () to write the paragraph data. */
+void
+uninhibit_output_flushing (void)
+{
+ flushing_ignored--;
+}
+
+void
+flush_output (void)
+{
+ int i;
+
+ if (!output_paragraph_offset || flushing_ignored)
+ return;
+
+ for (i = 0; i < output_paragraph_offset; i++)
+ {
+ if (output_paragraph[i] == '\n')
+ {
+ output_line_number++;
+ node_line_number++;
+ }
+
+ /* If we turned on the 8th bit for a space inside @w, turn it
+ back off for output. This might be problematic, since the
+ 0x80 character may be used in 8-bit character sets. Sigh.
+ In any case, don't do this for HTML, since the nbsp character
+ is valid input and must be passed along to the browser. */
+ if (!html && (output_paragraph[i] & meta_character_bit))
+ {
+ int temp = UNMETA (output_paragraph[i]);
+ if (temp == ' ')
+ output_paragraph[i] &= 0x7f;
+ }
+ }
+
+ fwrite (output_paragraph, 1, output_paragraph_offset, output_stream);
+
+ output_position += output_paragraph_offset;
+ output_paragraph_offset = 0;
+ meta_char_pos = 0;
+}
+
+/* How to close a paragraph controlling the number of lines between
+ this one and the last one. */
+
+/* Paragraph spacing is controlled by this variable. It is the number of
+ blank lines that you wish to appear between paragraphs. A value of
+ 1 creates a single blank line between paragraphs. */
+int paragraph_spacing = DEFAULT_PARAGRAPH_SPACING;
+
+static void
+close_paragraph_with_lines (int lines)
+{
+ int old_spacing = paragraph_spacing;
+ paragraph_spacing = lines;
+ close_paragraph ();
+ paragraph_spacing = old_spacing;
+}
+
+/* Close the current paragraph, leaving no blank lines between them. */
+void
+close_single_paragraph (void)
+{
+ close_paragraph_with_lines (0);
+}
+
+/* Close a paragraph after an insertion has ended. */
+void
+close_insertion_paragraph (void)
+{
+ if (!insertion_paragraph_closed)
+ {
+ /* Close the current paragraph, breaking the line. */
+ close_single_paragraph ();
+
+ /* Start a new paragraph, with the correct indentation for the now
+ current insertion level (one above the one that we are ending). */
+ start_paragraph ();
+
+ /* Tell `close_paragraph' that the previous line has already been
+ broken, so it should insert one less newline. */
+ line_already_broken = 1;
+
+ /* Tell functions such as `add_char' we've already found a newline. */
+ ignore_blank_line ();
+ }
+ else
+ {
+ /* If the insertion paragraph is closed already, then we are seeing
+ two `@end' commands in a row. Note that the first one we saw was
+ handled in the first part of this if-then-else clause, and at that
+ time `start_paragraph' was called, partially to handle the proper
+ indentation of the current line. However, the indentation level
+ may have just changed again, so we may have to outdent the current
+ line to the new indentation level. */
+ if (current_indent < output_column)
+ kill_self_indent (output_column - current_indent);
+ }
+
+ insertion_paragraph_closed = 1;
+}
+
+/* Close the currently open paragraph. */
+void
+close_paragraph (void)
+{
+ int i;
+
+ /* We don't need these newlines in XML and Docbook outputs for
+ paragraph seperation. We have <para> element for that. */
+ if (xml)
+ return;
+
+ /* The insertion paragraph is no longer closed. */
+ insertion_paragraph_closed = 0;
+
+ if (paragraph_is_open && !must_start_paragraph)
+ {
+ int tindex = output_paragraph_offset;
+
+ /* Back up to last non-newline/space character, forcing all such
+ subsequent characters to be newlines. This isn't strictly
+ necessary, but a couple of functions use the presence of a newline
+ to make decisions. */
+ for (tindex = output_paragraph_offset - 1; tindex >= 0; --tindex)
+ {
+ int c = output_paragraph[tindex];
+
+ if (c == ' '|| c == '\n')
+ output_paragraph[tindex] = '\n';
+ else
+ break;
+ }
+
+ /* All trailing whitespace is ignored. */
+ output_paragraph_offset = ++tindex;
+
+ /* Break the line if that is appropriate. */
+ if (paragraph_spacing >= 0)
+ insert ('\n');
+
+ /* Add as many blank lines as is specified in `paragraph_spacing'. */
+ if (!force_flush_right)
+ {
+ for (i = 0; i < (paragraph_spacing - line_already_broken); i++)
+ {
+ insert ('\n');
+ /* Don't need anything extra for HTML in usual case of no
+ extra paragraph spacing. */
+ if (html && i > 0)
+ insert_string ("<br>");
+ }
+ }
+
+ /* If we are doing flush right indentation, then do it now
+ on the paragraph (really a single line). */
+ if (force_flush_right)
+ do_flush_right_indentation ();
+
+ flush_output ();
+ paragraph_is_open = 0;
+ no_indent = 0;
+ output_column = 0;
+ }
+
+ ignore_blank_line ();
+}
+
+/* Make the last line just read look as if it were only a newline. */
+void
+ignore_blank_line (void)
+{
+ last_inserted_character = '\n';
+ last_char_was_newline = 1;
+}
+
+/* Align the end of the text in output_paragraph with fill_column. */
+static void
+do_flush_right_indentation (void)
+{
+ char *temp;
+ int temp_len;
+
+ kill_self_indent (-1);
+
+ if (output_paragraph[0] != '\n')
+ {
+ output_paragraph[output_paragraph_offset] = 0;
+
+ if (output_paragraph_offset < fill_column)
+ {
+ int i;
+
+ if (fill_column >= paragraph_buffer_len)
+ output_paragraph =
+ xrealloc (output_paragraph,
+ (paragraph_buffer_len += fill_column));
+
+ temp_len = strlen ((char *)output_paragraph);
+ temp = xmalloc (temp_len + 1);
+ memcpy (temp, (char *)output_paragraph, temp_len);
+
+ for (i = 0; i < fill_column - output_paragraph_offset; i++)
+ output_paragraph[i] = ' ';
+
+ memcpy ((char *)output_paragraph + i, temp, temp_len);
+ free (temp);
+ output_paragraph_offset = fill_column;
+ adjust_braces_following (0, i);
+ }
+ }
+}
+
+/* Begin a new paragraph. */
+void
+start_paragraph (void)
+{
+ /* First close existing one. */
+ if (paragraph_is_open)
+ close_paragraph ();
+
+ /* In either case, the insertion paragraph is no longer closed. */
+ insertion_paragraph_closed = 0;
+
+ /* However, the paragraph is open! */
+ paragraph_is_open = 1;
+
+ /* If we MUST_START_PARAGRAPH, that simply means that start_paragraph ()
+ had to be called before we would allow any other paragraph operations
+ to have an effect. */
+ if (!must_start_paragraph)
+ {
+ int amount_to_indent = 0;
+
+ /* If doing indentation, then insert the appropriate amount. */
+ if (!no_indent)
+ {
+ if (inhibit_paragraph_indentation)
+ {
+ amount_to_indent = current_indent;
+ if (inhibit_paragraph_indentation < 0)
+ inhibit_paragraph_indentation++;
+ }
+ else if (paragraph_start_indent < 0)
+ amount_to_indent = current_indent;
+ else
+ amount_to_indent = current_indent + paragraph_start_indent;
+
+ if (amount_to_indent >= output_column)
+ {
+ amount_to_indent -= output_column;
+ indent (amount_to_indent);
+ output_column += amount_to_indent;
+ }
+ }
+ }
+ else
+ must_start_paragraph = 0;
+}
+
+/* Insert the indentation specified by AMOUNT. */
+void
+indent (int amount)
+{
+ /* For every START_POS saved within the brace stack which will be affected
+ by this indentation, bump that start pos forward. */
+ adjust_braces_following (output_paragraph_offset, amount);
+
+ while (--amount >= 0)
+ insert (' ');
+}
+
+/* Search forward for STRING in input_text.
+ FROM says where where to start. */
+int
+search_forward (char *string, int from)
+{
+ int len = strlen (string);
+
+ while (from < input_text_length)
+ {
+ if (strncmp (input_text + from, string, len) == 0)
+ return from;
+ from++;
+ }
+ return -1;
+}
+
+/* search_forward until n characters. */
+int
+search_forward_until_pos (char *string, int from, int end_pos)
+{
+ int save_input_text_length = input_text_length;
+ input_text_length = end_pos;
+
+ from = search_forward (string, from);
+
+ input_text_length = save_input_text_length;
+
+ return from;
+}
+
+/* Return next non-whitespace and non-cr character. */
+int
+next_nonwhitespace_character (void)
+{
+ /* First check the current input_text. Start from the next char because
+ we already have input_text[input_text_offset] in ``current''. */
+ int pos = input_text_offset + 1;
+
+ while (pos < input_text_length)
+ {
+ if (!cr_or_whitespace(input_text[pos]))
+ return input_text[pos];
+ pos++;
+ }
+
+ { /* Can't find a valid character, so go through filestack
+ in case we are doing @include or expanding a macro. */
+ FSTACK *tos = filestack;
+
+ while (tos)
+ {
+ int tmp_input_text_length = filestack->size;
+ int tmp_input_text_offset = filestack->offset;
+ char *tmp_input_text = filestack->text;
+
+ while (tmp_input_text_offset < tmp_input_text_length)
+ {
+ if (!cr_or_whitespace(tmp_input_text[tmp_input_text_offset]))
+ return tmp_input_text[tmp_input_text_offset];
+ tmp_input_text_offset++;
+ }
+
+ tos = tos->next;
+ }
+ }
+
+ return -1;
+}
+
+/* An external image is a reference, kind of. The parsing is (not
+ coincidentally) similar, anyway. */
+void
+cm_image (int arg)
+{
+ char *name_arg, *w_arg, *h_arg, *alt_arg, *ext_arg;
+
+ if (arg == END)
+ return;
+
+ name_arg = get_xref_token (1); /* expands all macros in image */
+ w_arg = get_xref_token (0);
+ h_arg = get_xref_token (0);
+ alt_arg = get_xref_token (1); /* expands all macros in alt text */
+ ext_arg = get_xref_token (0);
+
+ if (*name_arg)
+ {
+ struct stat file_info;
+ char *pathname = NULL;
+ char *fullname = xmalloc (strlen (name_arg)
+ + (ext_arg && *ext_arg ? strlen (ext_arg) + 1: 4) + 1);
+
+ if (ext_arg && *ext_arg)
+ {
+ sprintf (fullname, "%s%s", name_arg, ext_arg);
+ if (access (fullname, R_OK) != 0)
+ pathname = get_file_info_in_path (fullname, include_files_path,
+ &file_info);
+
+ if (pathname == NULL)
+ {
+ /* Backwards compatibility (4.6 <= version < 4.7):
+ try prefixing @image's EXTENSION parameter with a period. */
+ sprintf (fullname, "%s.%s", name_arg, ext_arg);
+ if (access (fullname, R_OK) != 0)
+ pathname = get_file_info_in_path (fullname, include_files_path,
+ &file_info);
+ }
+ }
+ else
+ {
+ sprintf (fullname, "%s.png", name_arg);
+ if (access (fullname, R_OK) != 0) {
+ pathname = get_file_info_in_path (fullname,
+ include_files_path, &file_info);
+ if (pathname == NULL) {
+ sprintf (fullname, "%s.jpg", name_arg);
+ if (access (fullname, R_OK) != 0) {
+ sprintf (fullname, "%s.gif", name_arg);
+ if (access (fullname, R_OK) != 0) {
+ pathname = get_file_info_in_path (fullname,
+ include_files_path, &file_info);
+ }
+ }
+ }
+ }
+ }
+
+ if (html)
+ {
+ int image_in_div = 0;
+
+ if (pathname == NULL && access (fullname, R_OK) != 0)
+ {
+ line_error(_("@image file `%s' (for HTML) not readable: %s"),
+ fullname, strerror (errno));
+ return;
+ }
+ if (pathname != NULL && access (pathname, R_OK) != 0)
+ {
+ line_error (_("No such file `%s'"),
+ fullname);
+ return;
+ }
+
+ if (!paragraph_is_open)
+ {
+ add_html_block_elt ("<div class=\"block-image\">");
+ image_in_div = 1;
+ }
+
+ add_html_elt ("<img src=");
+ add_word_args ("\"%s\"", fullname);
+ add_html_elt (" alt=");
+ add_word_args ("\"%s\">",
+ escape_string (*alt_arg ? text_expansion (alt_arg) : fullname));
+
+ if (image_in_div)
+ add_html_block_elt ("</div>");
+ }
+ else if (xml && docbook)
+ xml_insert_docbook_image (name_arg);
+ else if (xml)
+ {
+ extern int xml_in_para;
+ extern int xml_no_para;
+ int elt = xml_in_para ? INLINEIMAGE : IMAGE;
+
+ if (!xml_in_para)
+ xml_no_para++;
+
+ xml_insert_element_with_attribute (elt,
+ START, "width=\"%s\" height=\"%s\" name=\"%s\" extension=\"%s\"",
+ w_arg, h_arg, name_arg, ext_arg);
+ xml_insert_element (IMAGEALTTEXT, START);
+ execute_string ("%s", alt_arg);
+ xml_insert_element (IMAGEALTTEXT, END);
+ xml_insert_element (elt, END);
+
+ if (!xml_in_para)
+ xml_no_para--;
+ }
+ else
+ { /* Try to open foo.EXT or foo.txt. */
+ FILE *image_file;
+ char *txtpath = NULL;
+ char *txtname = xmalloc (strlen (name_arg)
+ + (ext_arg && *ext_arg
+ ? strlen (ext_arg) : 4) + 1);
+ strcpy (txtname, name_arg);
+ strcat (txtname, ".txt");
+ image_file = fopen (txtname, "r");
+ if (image_file == NULL)
+ {
+ txtpath = get_file_info_in_path (txtname,
+ include_files_path, &file_info);
+ if (txtpath != NULL)
+ image_file = fopen (txtpath, "r");
+ }
+
+ if (image_file != NULL
+ || access (fullname, R_OK) == 0
+ || (pathname != NULL && access (pathname, R_OK) == 0))
+ {
+ int ch;
+ int save_inhibit_indentation = inhibit_paragraph_indentation;
+ int save_filling_enabled = filling_enabled;
+ int image_in_brackets = paragraph_is_open;
+
+ /* Write magic ^@^H[image ...^@^H] cookie in the info file, if
+ there's an accompanying bitmap. Otherwise just include the
+ text image. In the plaintext output, always include the text
+ image without the magic cookie. */
+ int use_magic_cookie = !no_headers
+ && access (fullname, R_OK) == 0 && !STREQ (fullname, txtname);
+
+ inhibit_paragraph_indentation = 1;
+ filling_enabled = 0;
+ last_char_was_newline = 0;
+
+ if (use_magic_cookie)
+ {
+ add_char ('\0');
+ add_word ("\010[image");
+
+ if (access (fullname, R_OK) == 0
+ || (pathname != NULL && access (pathname, R_OK) == 0))
+ add_word_args (" src=\"%s\"", fullname);
+
+ if (*alt_arg)
+ add_word_args (" alt=\"%s\"", alt_arg);
+ }
+
+ if (image_file != NULL)
+ {
+ if (use_magic_cookie)
+ add_word (" text=\"");
+
+ if (image_in_brackets)
+ add_char ('[');
+
+ /* Maybe we need to remove the final newline if the image
+ file is only one line to allow in-line images. On the
+ other hand, they could just make the file without a
+ final newline. */
+ while ((ch = getc (image_file)) != EOF)
+ {
+ if (use_magic_cookie && (ch == '"' || ch == '\\'))
+ add_char ('\\');
+ add_char (ch);
+ }
+
+ if (image_in_brackets)
+ add_char (']');
+
+ if (use_magic_cookie)
+ add_char ('"');
+
+ if (fclose (image_file) != 0)
+ perror (txtname);
+ }
+
+ if (use_magic_cookie)
+ {
+ add_char ('\0');
+ add_word ("\010]");
+ }
+
+ inhibit_paragraph_indentation = save_inhibit_indentation;
+ filling_enabled = save_filling_enabled;
+ }
+ else
+ warning (_("@image file `%s' (for text) unreadable: %s"),
+ txtname, strerror (errno));
+ }
+
+ free (fullname);
+ if (pathname)
+ free (pathname);
+ }
+ else
+ line_error (_("@image missing filename argument"));
+
+ if (name_arg)
+ free (name_arg);
+ if (w_arg)
+ free (w_arg);
+ if (h_arg)
+ free (h_arg);
+ if (alt_arg)
+ free (alt_arg);
+ if (ext_arg)
+ free (ext_arg);
+}
+
+/* Conditionals. */
+
+/* A structure which contains `defined' variables. */
+typedef struct defines {
+ struct defines *next;
+ char *name;
+ char *value;
+} DEFINE;
+
+/* The linked list of `set' defines. */
+DEFINE *defines = NULL;
+
+/* Add NAME to the list of `set' defines. */
+static void
+set (char *name, char *value)
+{
+ DEFINE *temp;
+
+ for (temp = defines; temp; temp = temp->next)
+ if (strcmp (name, temp->name) == 0)
+ {
+ free (temp->value);
+ temp->value = xstrdup (value);
+ return;
+ }
+
+ temp = xmalloc (sizeof (DEFINE));
+ temp->next = defines;
+ temp->name = xstrdup (name);
+ temp->value = xstrdup (value);
+ defines = temp;
+
+ if (xml && !docbook)
+ {
+ xml_insert_element_with_attribute (SETVALUE, START, "name=\"%s\"", name);
+ execute_string ("%s", value);
+ xml_insert_element (SETVALUE, END);
+ }
+}
+
+/* Remove NAME from the list of `set' defines. */
+static void
+clear (char *name)
+{
+ DEFINE *temp, *last;
+
+ last = NULL;
+ temp = defines;
+
+ while (temp)
+ {
+ if (strcmp (temp->name, name) == 0)
+ {
+ if (last)
+ last->next = temp->next;
+ else
+ defines = temp->next;
+
+ free (temp->name);
+ free (temp->value);
+ free (temp);
+ break;
+ }
+ last = temp;
+ temp = temp->next;
+ }
+
+ if (xml && !docbook)
+ {
+ xml_insert_element_with_attribute (CLEARVALUE, START, "name=\"%s\"", name);
+ xml_insert_element (CLEARVALUE, END);
+ }
+}
+
+/* Return the value of NAME. The return value is NULL if NAME is unset. */
+static char *
+set_p (char *name)
+{
+ DEFINE *temp;
+
+ for (temp = defines; temp; temp = temp->next)
+ if (strcmp (temp->name, name) == 0)
+ return temp->value;
+
+ return NULL;
+}
+
+/* Create a variable whose name appears as the first word on this line. */
+void
+cm_set (void)
+{
+ handle_variable (SET);
+}
+
+/* Remove a variable whose name appears as the first word on this line. */
+void
+cm_clear (void)
+{
+ handle_variable (CLEAR);
+}
+
+void
+cm_ifset (void)
+{
+ handle_variable (IFSET);
+}
+
+void
+cm_ifclear (void)
+{
+ handle_variable (IFCLEAR);
+}
+
+/* This command takes braces, but we parse the contents specially, so we
+ don't use the standard brace popping code.
+
+ The syntax @ifeq{arg1, arg2, texinfo-commands} performs texinfo-commands
+ if ARG1 and ARG2 caselessly string compare to the same string, otherwise,
+ it produces no output. */
+void
+cm_ifeq (void)
+{
+ char **arglist;
+
+ arglist = get_brace_args (0);
+
+ if (arglist)
+ {
+ if (array_len (arglist) > 1)
+ {
+ if ((strcasecmp (arglist[0], arglist[1]) == 0) &&
+ (arglist[2]))
+ execute_string ("%s\n", arglist[2]);
+ }
+
+ free_array (arglist);
+ }
+}
+
+void
+cm_value (int arg, int start_pos, int end_pos)
+{
+ static int value_level = 0, saved_meta_pos = -1;
+
+ /* xml_add_char() skips any content inside menus when output format is
+ Docbook, so @value{} is no use there. Also start_pos and end_pos does not
+ get updated, causing name to be empty string. So just return. */
+ if (docbook && in_menu)
+ return;
+
+ /* All the text after @value{ upto the matching } will eventually
+ disappear from output_paragraph, when this function is called
+ with ARG == END. If the text produced until then sets
+ meta_char_pos, we will need to restore it to the value it had
+ before @value was seen. So we need to save the previous value
+ of meta_char_pos here. */
+ if (arg == START)
+ {
+ /* If we are already inside some outer @value, don't overwrite
+ the value saved in saved_meta_pos. */
+ if (!value_level)
+ saved_meta_pos = meta_char_pos;
+ value_level++;
+ /* While the argument of @value is processed, we need to inhibit
+ textual transformations like "--" into "-", since @set didn't
+ do that when it grabbed the name of the variable. */
+ in_fixed_width_font++;
+ }
+ else
+ {
+ char *name = (char *) &output_paragraph[start_pos];
+ char *value;
+ output_paragraph[end_pos] = 0;
+ name = xstrdup (name);
+ value = set_p (name);
+ output_column -= end_pos - start_pos;
+ output_paragraph_offset = start_pos;
+
+ /* Restore the previous value of meta_char_pos if the stuff
+ inside this @value{} moved it. */
+ if (saved_meta_pos == -1) /* can't happen inside @value{} */
+ abort ();
+ if (value_level == 1
+ && meta_char_pos >= start_pos && meta_char_pos < end_pos)
+ {
+ meta_char_pos = saved_meta_pos;
+ saved_meta_pos = -1;
+ }
+ value_level--;
+ /* No need to decrement in_fixed_width_font, since before
+ we are called with arg == END, the reader loop already
+ popped the brace stack, which restored in_fixed_width_font,
+ among other things. */
+
+ if (value)
+ {
+ /* We need to get past the closing brace since the value may
+ expand to a context-sensitive macro (e.g. @xref) and produce
+ spurious warnings */
+ input_text_offset++;
+ execute_string ("%s", value);
+ input_text_offset--;
+ }
+ else
+ {
+ warning (_("undefined flag: %s"), name);
+ add_word_args (_("{No value for `%s'}"), name);
+ }
+
+ free (name);
+ }
+}
+
+/* Set, clear, or conditionalize based on ACTION. */
+static void
+handle_variable (int action)
+{
+ char *name;
+
+ get_rest_of_line (0, &name);
+ /* If we hit the end of text in get_rest_of_line, backing up
+ input pointer will cause the last character of the last line
+ be pushed back onto the input, which is wrong. */
+ if (input_text_offset < input_text_length)
+ backup_input_pointer ();
+ handle_variable_internal (action, name);
+ free (name);
+}
+
+static void
+handle_variable_internal (int action, char *name)
+{
+ char *temp;
+ int delimiter, additional_text_present = 0;
+
+ /* Only the first word of NAME is a valid tag. */
+ temp = name;
+ delimiter = 0;
+ while (*temp && (delimiter || !whitespace (*temp)))
+ {
+/* #if defined (SET_WITH_EQUAL) */
+ if (*temp == '"' || *temp == '\'')
+ {
+ if (*temp == delimiter)
+ delimiter = 0;
+ else
+ delimiter = *temp;
+ }
+/* #endif SET_WITH_EQUAL */
+ temp++;
+ }
+
+ if (*temp)
+ additional_text_present++;
+
+ *temp = 0;
+
+ if (!*name)
+ line_error (_("%c%s requires a name"), COMMAND_PREFIX, command);
+ else
+ {
+ switch (action)
+ {
+ case SET:
+ {
+ char *value;
+
+#if defined (SET_WITH_EQUAL)
+ /* Allow a value to be saved along with a variable. The value is
+ the text following an `=' sign in NAME, if any is present. */
+
+ for (value = name; *value && *value != '='; value++);
+
+ if (*value)
+ *value++ = 0;
+
+ if (*value == '"' || *value == '\'')
+ {
+ value++;
+ value[strlen (value) - 1] = 0;
+ }
+
+#else /* !SET_WITH_EQUAL */
+ /* The VALUE of NAME is the remainder of the line sans
+ whitespace. */
+ if (additional_text_present)
+ {
+ value = temp + 1;
+ canon_white (value);
+ }
+ else
+ value = "";
+#endif /* !SET_WITH_VALUE */
+
+ set (name, value);
+ }
+ break;
+
+ case CLEAR:
+ clear (name);
+ break;
+
+ case IFSET:
+ case IFCLEAR:
+ /* If IFSET and NAME is not set, or if IFCLEAR and NAME is set,
+ read lines from the the file until we reach a matching
+ "@end CONDITION". This means that we only take note of
+ "@ifset/clear" and "@end" commands. */
+ {
+ char condition[8];
+ int condition_len;
+ int orig_line_number = line_number;
+
+ if (action == IFSET)
+ strcpy (condition, "ifset");
+ else
+ strcpy (condition, "ifclear");
+
+ condition_len = strlen (condition);
+
+ if ((action == IFSET && !set_p (name))
+ || (action == IFCLEAR && set_p (name)))
+ {
+ int level = 0, done = 0;
+
+ while (!done && input_text_offset < input_text_length)
+ {
+ char *freeable_line, *line;
+
+ get_rest_of_line (0, &freeable_line);
+
+ for (line = freeable_line; whitespace (*line); line++);
+
+ if (*line == COMMAND_PREFIX &&
+ (strncmp (line + 1, condition, condition_len) == 0))
+ level++;
+ else if (strncmp (line, "@end", 4) == 0)
+ {
+ char *cname = line + 4;
+ char *temp;
+
+ while (*cname && whitespace (*cname))
+ cname++;
+ temp = cname;
+
+ while (*temp && !whitespace (*temp))
+ temp++;
+ *temp = 0;
+
+ if (strcmp (cname, condition) == 0)
+ {
+ if (!level)
+ {
+ done = 1;
+ }
+ else
+ level--;
+ }
+ }
+ free (freeable_line);
+ }
+
+ if (!done)
+ file_line_error (input_filename, orig_line_number,
+ _("Reached eof before matching @end %s"),
+ condition);
+
+ /* We found the end of a false @ifset/ifclear. If we are
+ in a menu, back up over the newline that ends the ifset,
+ since that newline may also begin the next menu entry. */
+ break;
+ }
+ else
+ {
+ if (action == IFSET)
+ begin_insertion (ifset);
+ else
+ begin_insertion (ifclear);
+ }
+ }
+ break;
+ }
+ }
+}
+
+/* Execution of random text not in file. */
+typedef struct {
+ char *string; /* The string buffer. */
+ int size; /* The size of the buffer. */
+ int in_use; /* Nonzero means string currently in use. */
+} EXECUTION_STRING;
+
+static EXECUTION_STRING **execution_strings = NULL;
+static int execution_strings_index = 0;
+static int execution_strings_slots = 0;
+
+static EXECUTION_STRING *
+get_execution_string (int initial_size)
+{
+ int i = 0;
+ EXECUTION_STRING *es = NULL;
+
+ if (execution_strings)
+ {
+ for (i = 0; i < execution_strings_index; i++)
+ if (execution_strings[i] && (execution_strings[i]->in_use == 0))
+ {
+ es = execution_strings[i];
+ break;
+ }
+ }
+
+ if (!es)
+ {
+ if (execution_strings_index + 1 >= execution_strings_slots)
+ {
+ execution_strings = xrealloc
+ (execution_strings,
+ (execution_strings_slots += 3) * sizeof (EXECUTION_STRING *));
+ for (; i < execution_strings_slots; i++)
+ execution_strings[i] = NULL;
+ }
+
+ execution_strings[execution_strings_index] =
+ xmalloc (sizeof (EXECUTION_STRING));
+ es = execution_strings[execution_strings_index];
+ execution_strings_index++;
+
+ es->size = 0;
+ es->string = NULL;
+ es->in_use = 0;
+ }
+
+ if (initial_size > es->size)
+ {
+ es->string = xrealloc (es->string, initial_size);
+ es->size = initial_size;
+ }
+ return es;
+}
+
+/* Given a pointer to TEXT and its desired length NEW_LEN, find TEXT's
+ entry in the execution_strings[] array and change the .STRING and
+ .SIZE members of that entry as appropriate. */
+void
+maybe_update_execution_strings (char **text, unsigned int new_len)
+{
+ int i = 0;
+
+ if (execution_strings)
+ {
+ for (i = 0; i < execution_strings_index; i++)
+ if (execution_strings[i] && (execution_strings[i]->in_use == 1) &&
+ execution_strings[i]->string == *text)
+ {
+ /* Don't ever shrink the string storage in execution_strings[]!
+ execute_string assumes that it is always big enough to store
+ every possible execution_string, and will break if that's
+ not true. So we only enlarge the string storage if the
+ current size isn't big enough. */
+ if (execution_strings[i]->size < new_len)
+ {
+ execution_strings[i]->string =
+ *text = xrealloc (*text, new_len + 1);
+ execution_strings[i]->size = new_len + 1;
+ }
+ return;
+ }
+ }
+ /* We should *never* end up here, since if we are inside
+ execute_string, TEXT is always in execution_strings[]. */
+ abort ();
+}
+
+/* FIXME: this is an arbitrary limit. */
+#define EXECUTE_STRING_MAX 16*1024
+
+/* Execute the string produced by formatting the ARGs with FORMAT. This
+ is like submitting a new file with @include. */
+void
+#if defined (VA_FPRINTF) && __STDC__
+execute_string (char *format, ...)
+#else
+execute_string (format, va_alist)
+ char *format;
+ va_dcl
+#endif
+{
+ EXECUTION_STRING *es;
+ char *temp_string, *temp_input_filename;
+#ifdef VA_FPRINTF
+ va_list ap;
+#endif
+ int insertion_level_at_start = insertion_level;
+
+ es = get_execution_string (EXECUTE_STRING_MAX);
+ temp_string = es->string;
+ es->in_use = 1;
+
+ VA_START (ap, format);
+#ifdef VA_SPRINTF
+ VA_SPRINTF (temp_string, format, ap);
+#else
+ sprintf (temp_string, format, a1, a2, a3, a4, a5, a6, a7, a8);
+#endif /* not VA_SPRINTF */
+ va_end (ap);
+
+ pushfile ();
+ input_text_offset = 0;
+ input_text = temp_string;
+ input_text_length = strlen (temp_string);
+ input_filename = xstrdup (input_filename);
+ temp_input_filename = input_filename;
+
+ executing_string++;
+ reader_loop ();
+
+ /* If insertion stack level changes during execution, that means a multiline
+ command is used inside braces or @section ... kind of commands. */
+ if (insertion_level_at_start != insertion_level && !executing_macro)
+ {
+ line_error (_("Multiline command %c%s used improperly"),
+ COMMAND_PREFIX,
+ command);
+ /* We also need to keep insertion_level intact to make sure warnings are
+ issued for @end ... command. */
+ while (insertion_level > insertion_level_at_start)
+ pop_insertion ();
+ }
+
+ popfile ();
+ executing_string--;
+ es->in_use = 0;
+ free (temp_input_filename);
+}
+
+
+/* Return what would be output for STR (in newly-malloced memory), i.e.,
+ expand Texinfo commands according to the current output format. If
+ IMPLICIT_CODE is set, expand @code{STR}. This is generally used for
+ short texts; filling, indentation, and html escapes are disabled. */
+
+char *
+expansion (char *str, int implicit_code)
+{
+ return maybe_escaped_expansion (str, implicit_code, 0);
+}
+
+
+/* Do HTML escapes according to DO_HTML_ESCAPE. Needed in
+ cm_printindex, q.v. */
+
+char *
+maybe_escaped_expansion (char *str, int implicit_code, int do_html_escape)
+{
+ char *result;
+
+ /* Inhibit indentation and filling, so that extra newlines
+ are not added to the expansion. (This is undesirable if
+ we write the expanded text to macro_expansion_output_stream.) */
+ int saved_filling_enabled = filling_enabled;
+ int saved_indented_fill = indented_fill;
+ int saved_no_indent = no_indent;
+ int saved_escape_html = escape_html;
+
+ filling_enabled = 0;
+ indented_fill = 0;
+ no_indent = 1;
+ escape_html = do_html_escape;
+
+ result = full_expansion (str, implicit_code);
+
+ filling_enabled = saved_filling_enabled;
+ indented_fill = saved_indented_fill;
+ no_indent = saved_no_indent;
+ escape_html = saved_escape_html;
+
+ return result;
+}
+
+
+/* Expand STR (or @code{STR} if IMPLICIT_CODE is nonzero). No change to
+ any formatting parameters -- filling, indentation, html escapes,
+ etc., are not reset. Always returned in new memory. */
+
+char *
+full_expansion (char *str, int implicit_code)
+{
+ int length;
+ char *result;
+
+ /* Inhibit any real output. */
+ int start = output_paragraph_offset;
+ int saved_paragraph_is_open = paragraph_is_open;
+ int saved_output_column = output_column;
+
+ /* More output state to save. */
+ int saved_meta_pos = meta_char_pos;
+ int saved_last_char = last_inserted_character;
+ int saved_last_nl = last_char_was_newline;
+
+ /* If we are called in the middle of processing a command, we need
+ to dup and save the global variable `command' (which holds the
+ name of this command), since the recursive reader loop will free
+ it from under our feet if it finds any macros in STR. */
+ char *saved_command = command ? xstrdup (command) : NULL;
+
+ inhibit_output_flushing ();
+ paragraph_is_open = 1;
+ if (strlen (str) > (implicit_code
+ ? EXECUTE_STRING_MAX - 1 - sizeof("@code{}")
+ : EXECUTE_STRING_MAX - 1))
+ line_error (_("`%.40s...' is too long for expansion; not expanded"), str);
+ else
+ execute_string (implicit_code ? "@code{%s}" : "%s", str);
+ uninhibit_output_flushing ();
+
+ /* Copy the expansion from the buffer. */
+ length = output_paragraph_offset - start;
+ result = xmalloc (1 + length);
+ memcpy (result, (char *) (output_paragraph + start), length);
+ result[length] = 0;
+
+ /* Pretend it never happened. */
+ free_and_clear (&command);
+ command = saved_command;
+
+ output_paragraph_offset = start;
+ paragraph_is_open = saved_paragraph_is_open;
+ output_column = saved_output_column;
+
+ meta_char_pos = saved_meta_pos;
+ last_inserted_character = saved_last_char;
+ last_char_was_newline = saved_last_nl;
+
+ return result;
+}
+
+
+/* Return text (info) expansion of STR no matter what the current output
+ format is. */
+
+char *
+text_expansion (char *str)
+{
+ char *ret;
+ int save_html = html;
+ int save_xml = xml;
+ int save_docbook = docbook;
+
+ html = 0;
+ xml = 0;
+ docbook = 0;
+ ret = expansion (str, 0);
+ html = save_html;
+ xml = save_xml;
+ docbook = save_docbook;
+
+ return ret;
+}
+
+
+/* Set the paragraph indentation variable to the value specified in STRING.
+ Values can be:
+ `asis': Don't change existing indentation.
+ `none': Remove existing indentation.
+ NUM: Indent NUM spaces at the starts of paragraphs.
+ If NUM is zero, we assume `none'.
+ Returns 0 if successful, or nonzero if STRING isn't one of the above. */
+int
+set_paragraph_indent (char *string)
+{
+ if (strcmp (string, "asis") == 0 || strcmp (string, _("asis")) == 0)
+ paragraph_start_indent = 0;
+ else if (strcmp (string, "none") == 0 || strcmp (string, _("none")) == 0)
+ paragraph_start_indent = -1;
+ else
+ {
+ if (sscanf (string, "%d", &paragraph_start_indent) != 1)
+ return -1;
+ else
+ {
+ if (paragraph_start_indent == 0)
+ paragraph_start_indent = -1;
+ }
+ }
+ return 0;
+}