diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-15 19:44:05 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-15 19:44:05 +0000 |
commit | d318611dd6f23fcfedd50e9b9e24620b102ba96a (patch) | |
tree | 8b9eef82ca40fdd5a8deeabf07572074c236095d /src/preproc/refer/refer.cpp | |
parent | Initial commit. (diff) | |
download | groff-upstream/1.23.0.tar.xz groff-upstream/1.23.0.zip |
Adding upstream version 1.23.0.upstream/1.23.0upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to '')
-rw-r--r-- | src/preproc/refer/refer.cpp | 1267 |
1 files changed, 1267 insertions, 0 deletions
diff --git a/src/preproc/refer/refer.cpp b/src/preproc/refer/refer.cpp new file mode 100644 index 0000000..a5c291e --- /dev/null +++ b/src/preproc/refer/refer.cpp @@ -0,0 +1,1267 @@ +/* Copyright (C) 1989-2020 Free Software Foundation, Inc. + Written by James Clark (jjc@jclark.com) + +This file is part of groff. + +groff 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 3 of the License, or +(at your option) any later version. + +groff 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, see <http://www.gnu.org/licenses/>. */ + +#include "refer.h" +#include "refid.h" +#include "ref.h" +#include "token.h" +#include "search.h" +#include "command.h" + +extern "C" const char *Version_string; + +const char PRE_LABEL_MARKER = '\013'; +const char POST_LABEL_MARKER = '\014'; +const char LABEL_MARKER = '\015'; // label_type is added on + +#define FORCE_LEFT_BRACKET 04 +#define FORCE_RIGHT_BRACKET 010 + +static FILE *outfp = stdout; + +string capitalize_fields; +string reverse_fields; +string abbreviate_fields; +string period_before_last_name = ". "; +string period_before_initial = "."; +string period_before_hyphen = ""; +string period_before_other = ". "; +string sort_fields; +int annotation_field = -1; +string annotation_macro; +string discard_fields = "XYZ"; +string pre_label = "\\*([."; +string post_label = "\\*(.]"; +string sep_label = ", "; +int have_bibliography = 0; +int accumulate = 0; +int move_punctuation = 0; +int abbreviate_label_ranges = 0; +string label_range_indicator; +int label_in_text = 1; +int label_in_reference = 1; +int date_as_label = 0; +int sort_adjacent_labels = 0; +// Join exactly two authors with this. +string join_authors_exactly_two = " and "; +// When there are more than two authors join the last two with this. +string join_authors_last_two = ", and "; +// Otherwise join authors with this. +string join_authors_default = ", "; +string separate_label_second_parts = ", "; +// Use this string to represent that there are other authors. +string et_al = " et al"; +// Use et al only if it can replace at least this many authors. +int et_al_min_elide = 2; +// Use et al only if the total number of authors is at least this. +int et_al_min_total = 3; + + +int compatible_flag = 0; + +int short_label_flag = 0; + +static bool recognize_R1_R2 = true; + +search_list database_list; +int search_default = 1; +static int default_database_loaded = 0; + +static reference **citation = 0; +static int ncitations = 0; +static int citation_max = 0; + +static reference **reference_hash_table = 0; +static int hash_table_size; +static int nreferences = 0; + +static int need_syncing = 0; +string pending_line; +string pending_lf_lines; + +static void output_pending_line(); +static unsigned immediately_handle_reference(const string &); +static void immediately_output_references(); +static unsigned store_reference(const string &); +static void divert_to_temporary_file(); +static reference *make_reference(const string &, unsigned *); +static void usage(FILE *stream); +static void do_file(const char *); +static void split_punct(string &line, string &punct); +static void output_citation_group(reference **v, int n, label_type, + FILE *fp); +static void possibly_load_default_database(); + +int main(int argc, char **argv) +{ + program_name = argv[0]; + static char stderr_buf[BUFSIZ]; + setbuf(stderr, stderr_buf); + outfp = stdout; + int finished_options = 0; + int bib_flag = 0; + int done_spec = 0; + + // TODO: Migrate to getopt_long; see, e.g., src/preproc/eqn/main.cpp. + for (--argc, ++argv; + !finished_options && argc > 0 && argv[0][0] == '-' + && argv[0][1] != '\0'; + argv++, argc--) { + const char *opt = argv[0] + 1; + while (opt != 0 && *opt != '\0') { + switch (*opt) { + case 'C': + compatible_flag = 1; + opt++; + break; + case 'B': + bib_flag = 1; + label_in_reference = 0; + label_in_text = 0; + ++opt; + if (*opt == '\0') { + annotation_field = 'X'; + annotation_macro = "AP"; + } + else if (csalnum(opt[0]) && opt[1] == '.' && opt[2] != '\0') { + annotation_field = opt[0]; + annotation_macro = opt + 2; + } + opt = 0; + break; + case 'P': + move_punctuation = 1; + opt++; + break; + case 'R': + recognize_R1_R2 = false; + opt++; + break; + case 'S': + // Not a very useful spec. + set_label_spec("(A.n|Q)', '(D.y|D)"); + done_spec = 1; + pre_label = " ("; + post_label = ")"; + sep_label = "; "; + opt++; + break; + case 'V': + do_verify = true; + opt++; + break; + case 'f': + { + const char *num = 0; + if (*++opt == '\0') { + if (argc > 1) { + num = *++argv; + --argc; + } + else { + error("'f' option requires an argument"); + usage(stderr); + exit(1); + } + } + else { + num = opt; + opt = 0; + } + const char *ptr; + for (ptr = num; *ptr; ptr++) + if (!csdigit(*ptr)) { + error("invalid character '%1' in argument to 'f' option", + *ptr); + break; + } + if (*ptr == '\0') { + string spec; + spec = '%'; + spec += num; + spec += '\0'; + set_label_spec(spec.contents()); + done_spec = 1; + } + break; + } + case 'b': + label_in_text = 0; + label_in_reference = 0; + opt++; + break; + case 'e': + accumulate = 1; + opt++; + break; + case 'c': + capitalize_fields = ++opt; + opt = 0; + break; + case 'k': + { + char buf[5]; + if (csalpha(*++opt)) + buf[0] = *opt++; + else { + if (*opt != '\0') + error("invalid field name '%1' in argument to 'k' option", + *opt++); + buf[0] = 'L'; + } + buf[1] = '~'; + buf[2] = '%'; + buf[3] = 'a'; + buf[4] = '\0'; + set_label_spec(buf); + done_spec = 1; + } + break; + case 'a': + { + const char *ptr; + for (ptr = ++opt; *ptr; ptr++) + if (!csdigit(*ptr)) { + error("'a' option argument must be an integer"); + break; + } + if (*ptr == '\0') { + reverse_fields = 'A'; + reverse_fields += opt; + } + opt = 0; + } + break; + case 'i': + linear_ignore_fields = ++opt; + opt = 0; + break; + case 'l': + { + char buf[INT_DIGITS*2 + 11]; // A.n+2D.y-3%a + strcpy(buf, "A.n"); + if (*++opt != '\0' && *opt != ',') { + char *ptr; + long n = strtol(opt, &ptr, 10); + if (n == 0 && ptr == opt) { + error("invalid integer '%1' in 'l' option argument", opt); + opt = 0; + break; + } + if (n < 0) + n = 0; + opt = ptr; + sprintf(strchr(buf, '\0'), "+%ld", n); + } + strcat(buf, "D.y"); + if (*opt == ',') + opt++; + if (*opt != '\0') { + char *ptr; + long n = strtol(opt, &ptr, 10); + if (n == 0 && ptr == opt) { + error("invalid integer '%1' in 'l' option argument", opt); + opt = 0; + break; + } + if (n < 0) + n = 0; + sprintf(strchr(buf, '\0'), "-%ld", n); + opt = ptr; + if (*opt != '\0') + error("argument to 'l' option not of form 'm,n'"); + } + strcat(buf, "%a"); + if (!set_label_spec(buf)) + assert(0 == "set_label_spec() failed"); + done_spec = 1; + } + break; + case 'n': + search_default = 0; + opt++; + break; + case 'p': + { + const char *filename = 0; + if (*++opt == '\0') { + if (argc > 1) { + filename = *++argv; + argc--; + } + else { + error("option 'p' requires an argument"); + usage(stderr); + exit(1); + } + } + else { + filename = opt; + opt = 0; + } + database_list.add_file(filename); + } + break; + case 's': + if (*++opt == '\0') + sort_fields = "AD"; + else { + sort_fields = opt; + opt = 0; + } + accumulate = 1; + break; + case 't': + { + char *ptr; + long n = strtol(opt, &ptr, 10); + if (n == 0 && ptr == opt) { + error("invalid integer '%1' in 't' option argument", opt); + opt = 0; + break; + } + if (n < 1) + n = 1; + linear_truncate_len = int(n); + opt = ptr; + break; + } + case '-': + if (opt[1] == '\0') { + finished_options = 1; + opt++; + break; + } + if (strcmp(opt, "-version") == 0) { + case 'v': + printf("GNU refer (groff) version %s\n", Version_string); + exit(0); + break; + } + if (strcmp(opt, "-help") == 0) { + usage(stdout); + exit(0); + break; + } + // fall through + default: + error("unrecognized option '%1'", opt); + usage(stderr); + exit(1); + break; + } + } + } + if (!done_spec) + set_label_spec("%1"); + if (argc <= 0) { + if (bib_flag) + do_bib("-"); + else + do_file("-"); + } + else { + for (int i = 0; i < argc; i++) { + if (bib_flag) + do_bib(argv[i]); + else + do_file(argv[i]); + } + } + if (accumulate) + output_references(); + if (fflush(stdout) < 0) + fatal("output error: %1", strerror(errno)); + return 0; +} + +static void usage(FILE *stream) +{ + fprintf(stream, +"usage: %s [-bCenPRS] [-aN] [-cXYZ] [-fN] [-iXYZ] [-kX] [-lM,N]" +" [-p db-file] [-sXYZ] [-tN] [-Bl.m] [file ...]\n" +"usage: %s {-v | --version}\n" +"usage: %s --help\n", + program_name, program_name, program_name); +} + +static void possibly_load_default_database() +{ + if (search_default && !default_database_loaded) { + char *filename = getenv("REFER"); + if (filename) + database_list.add_file(filename); + else + database_list.add_file(DEFAULT_INDEX, 1); + default_database_loaded = 1; + } +} + +static bool is_list(const string &str) +{ + const char *start = str.contents(); + const char *end = start + str.length(); + while (end > start && csspace(end[-1])) + end--; + while (start < end && csspace(*start)) + start++; + return end - start == 6 && memcmp(start, "$LIST$", 6) == 0; +} + +static void do_file(const char *filename) +{ + FILE *fp; + if (strcmp(filename, "-") == 0) { + fp = stdin; + } + else { + errno = 0; + fp = fopen(filename, "r"); + if (fp == 0) { + error("can't open '%1': %2", filename, strerror(errno)); + return; + } + } + string fn(filename); + fn += '\0'; + normalize_for_lf(fn); + current_filename = fn.contents(); + fprintf(outfp, ".lf 1 %s\n", current_filename); + current_lineno = 1; + string line; + for (;;) { + line.clear(); + for (;;) { + int c = getc(fp); + if (EOF == c) { + if (line.length() > 0) + line += '\n'; + break; + } + if (is_invalid_input_char(c)) + error("invalid input character code %1", c); + else { + line += c; + if ('\n' == c) + break; + } + } + int len = line.length(); + if (len == 0) + break; + current_lineno++; + if (len >= 2 && line[0] == '.' && line[1] == '[') { + int start_lineno = current_lineno; + bool at_start_of_line = true; + string str; + string post; + string pre(line.contents() + 2, line.length() - 3); + for (;;) { + int c = getc(fp); + if (EOF == c) { + error_with_file_and_line(current_filename, start_lineno, + "missing '.]' line"); + break; + } + if (at_start_of_line) + current_lineno++; + if (at_start_of_line && '.' == c) { + int d = getc(fp); + if (d == ']') { + while ((d = getc(fp)) != '\n' && d != EOF) { + if (is_invalid_input_char(d)) + error("invalid input character code %1", d); + else + post += d; + } + break; + } + if (d != EOF) + ungetc(d, fp); + } + if (is_invalid_input_char(c)) + error("invalid input character code %1", c); + else + str += c; + at_start_of_line = ('\n' == c); + } + if (is_list(str)) { + output_pending_line(); + if (accumulate) + output_references(); + else + error("found '$LIST$' but not accumulating references"); + } + else { + unsigned flags = (accumulate + ? store_reference(str) + : immediately_handle_reference(str)); + if (label_in_text) { + if (accumulate && outfp == stdout) + divert_to_temporary_file(); + if (pending_line.length() == 0) { + warning("can't attach citation to previous line"); + } + else + pending_line.set_length(pending_line.length() - 1); + string punct; + if (move_punctuation) + split_punct(pending_line, punct); + int have_text = pre.length() > 0 || post.length() > 0; + label_type lt = label_type(flags & ~(FORCE_LEFT_BRACKET + |FORCE_RIGHT_BRACKET)); + if ((flags & FORCE_LEFT_BRACKET) || !have_text) + pending_line += PRE_LABEL_MARKER; + pending_line += pre; + char lm = LABEL_MARKER + (int)lt; + pending_line += lm; + pending_line += post; + if ((flags & FORCE_RIGHT_BRACKET) || !have_text) + pending_line += POST_LABEL_MARKER; + pending_line += punct; + pending_line += '\n'; + } + } + need_syncing = 1; + } + else if (len >= 4 + && '.' == line[0] && 'l' == line[1] && 'f' == line[2] + && (compatible_flag || '\n' == line[3] || ' ' == line[3])) + { + pending_lf_lines += line; + line += '\0'; + if (interpret_lf_args(line.contents() + 3)) + current_lineno--; + } + else if (recognize_R1_R2 + && len >= 4 + && '.' == line[0] && 'R' == line[1] && '1' == line[2] + && (compatible_flag || '\n' == line[3] || ' ' == line[3])) + { + line.clear(); + int start_lineno = current_lineno; + bool at_start_of_line = true; + for (;;) { + int c = getc(fp); + if (c != EOF && at_start_of_line) + current_lineno++; + if (at_start_of_line && '.' == c) { + c = getc(fp); + if ('R' == c) { + c = getc(fp); + if ('2' == c) { + c = getc(fp); + if (compatible_flag || ' ' == c || '\n' == c || EOF == c) + { + while (c != EOF && c != '\n') + c = getc(fp); + break; + } + else { + line += '.'; + line += 'R'; + line += '2'; + } + } + else { + line += '.'; + line += 'R'; + } + } + else + line += '.'; + } + if (EOF == c) { + error_with_file_and_line(current_filename, start_lineno, + "missing '.R2' line"); + break; + } + if (is_invalid_input_char(c)) + error_with_file_and_line(current_filename, start_lineno, + "invalid input character code %1", + c); + else { + line += c; + at_start_of_line = ('\n' == c); + } + } + output_pending_line(); + if (accumulate) + output_references(); + else + nreferences = 0; + process_commands(line, current_filename, start_lineno + 1); + need_syncing = 1; + } + else { + output_pending_line(); + pending_line = line; + } + } + need_syncing = 0; + output_pending_line(); + if (fp != stdin) + fclose(fp); +} + +class label_processing_state { + enum { + NORMAL, + PENDING_LABEL, + PENDING_LABEL_POST, + PENDING_LABEL_POST_PRE, + PENDING_POST + } state; + label_type type; // type of pending labels + int count; // number of pending labels + reference **rptr; // pointer to next reference + int rcount; // number of references left + FILE *fp; + int handle_pending(int c); +public: + label_processing_state(reference **, int, FILE *); + ~label_processing_state(); + void process(int c); +}; + +static void output_pending_line() +{ + if (label_in_text && !accumulate && ncitations > 0) { + label_processing_state state(citation, ncitations, outfp); + int len = pending_line.length(); + for (int i = 0; i < len; i++) + state.process((unsigned char)(pending_line[i])); + } + else + put_string(pending_line, outfp); + pending_line.clear(); + if (pending_lf_lines.length() > 0) { + put_string(pending_lf_lines, outfp); + pending_lf_lines.clear(); + } + if (!accumulate) + immediately_output_references(); + if (need_syncing) { + fprintf(outfp, ".lf %d %s\n", current_lineno, current_filename); + need_syncing = 0; + } +} + +static void split_punct(string &line, string &punct) +{ + const char *start = line.contents(); + const char *end = start + line.length(); + const char *ptr = start; + const char *last_token_start = 0; + for (;;) { + if (ptr >= end) + break; + last_token_start = ptr; + if (*ptr == PRE_LABEL_MARKER || *ptr == POST_LABEL_MARKER + || (*ptr >= LABEL_MARKER + && *ptr < LABEL_MARKER + N_LABEL_TYPES)) + ptr++; + else if (!get_token(&ptr, end)) + break; + } + if (last_token_start) { + const token_info *ti = lookup_token(last_token_start, end); + if (ti->is_punct()) { + punct.append(last_token_start, end - last_token_start); + line.set_length(last_token_start - start); + } + } +} + +static void divert_to_temporary_file() +{ + outfp = xtmpfile(); +} + +static void store_citation(reference *ref) +{ + if (ncitations >= citation_max) { + if (citation == 0) + citation = new reference*[citation_max = 100]; + else { + reference **old_citation = citation; + citation_max *= 2; + citation = new reference *[citation_max]; + memcpy(citation, old_citation, ncitations*sizeof(reference *)); + delete[] old_citation; + } + } + citation[ncitations++] = ref; +} + +static unsigned store_reference(const string &str) +{ + if (reference_hash_table == 0) { + reference_hash_table = new reference *[17]; + hash_table_size = 17; + for (int i = 0; i < hash_table_size; i++) + reference_hash_table[i] = 0; + } + unsigned flags; + reference *ref = make_reference(str, &flags); + ref->compute_hash_code(); + unsigned h = ref->hash(); + reference **ptr; + for (ptr = reference_hash_table + (h % hash_table_size); + *ptr != 0; + ((ptr == reference_hash_table) + ? (ptr = reference_hash_table + hash_table_size - 1) + : --ptr)) + if (same_reference(**ptr, *ref)) + break; + if (*ptr != 0) { + if (ref->is_merged()) + warning("fields ignored because reference already used"); + delete ref; + ref = *ptr; + } + else { + *ptr = ref; + ref->set_number(nreferences); + nreferences++; + ref->pre_compute_label(); + ref->compute_sort_key(); + if (nreferences*2 >= hash_table_size) { + // Rehash it. + reference **old_table = reference_hash_table; + int old_size = hash_table_size; + hash_table_size = next_size(hash_table_size); + reference_hash_table = new reference*[hash_table_size]; + int i; + for (i = 0; i < hash_table_size; i++) + reference_hash_table[i] = 0; + for (i = 0; i < old_size; i++) + if (old_table[i]) { + reference **p; + for (p = (reference_hash_table + + (old_table[i]->hash() % hash_table_size)); + *p; + ((p == reference_hash_table) + ? (p = reference_hash_table + hash_table_size - 1) + : --p)) + ; + *p = old_table[i]; + } + delete[] old_table; + } + } + if (label_in_text) + store_citation(ref); + return flags; +} + +unsigned immediately_handle_reference(const string &str) +{ + unsigned flags; + reference *ref = make_reference(str, &flags); + ref->set_number(nreferences); + if (label_in_text || label_in_reference) { + ref->pre_compute_label(); + ref->immediate_compute_label(); + } + nreferences++; + store_citation(ref); + return flags; +} + +static void immediately_output_references() +{ + for (int i = 0; i < ncitations; i++) { + reference *ref = citation[i]; + if (label_in_reference) { + fputs(".ds [F ", outfp); + const string &label = ref->get_label(NORMAL_LABEL); + if (label.length() > 0 + && (label[0] == ' ' || label[0] == '\\' || label[0] == '"')) + putc('"', outfp); + put_string(label, outfp); + putc('\n', outfp); + } + ref->output(outfp); + delete ref; + } + ncitations = 0; +} + +static void output_citation_group(reference **v, int n, label_type type, + FILE *fp) +{ + if (sort_adjacent_labels) { + // Do an insertion sort. Usually n will be very small. + for (int i = 1; i < n; i++) { + int num = v[i]->get_number(); + reference *temp = v[i]; + int j; + for (j = i - 1; j >= 0 && v[j]->get_number() > num; j--) + v[j + 1] = v[j]; + v[j + 1] = temp; + } + } + // This messes up if !accumulate. + if (accumulate && n > 1) { + // remove duplicates + int j = 1; + for (int i = 1; i < n; i++) + if (v[i]->get_label(type) != v[i - 1]->get_label(type)) + v[j++] = v[i]; + n = j; + } + string merged_label; + for (int i = 0; i < n; i++) { + int nmerged = v[i]->merge_labels(v + i + 1, n - i - 1, type, + merged_label); + if (nmerged > 0) { + put_string(merged_label, fp); + i += nmerged; + } + else + put_string(v[i]->get_label(type), fp); + if (i < n - 1) + put_string(sep_label, fp); + } +} + + +label_processing_state::label_processing_state(reference **p, int n, + FILE *f) +: state(NORMAL), count(0), rptr(p), rcount(n), fp(f) +{ +} + +label_processing_state::~label_processing_state() +{ + int handled = handle_pending(EOF); + assert(!handled); + assert(rcount == 0); +} + +int label_processing_state::handle_pending(int c) +{ + switch (state) { + case NORMAL: + break; + case PENDING_LABEL: + if (POST_LABEL_MARKER == c) { + state = PENDING_LABEL_POST; + return 1; + } + else { + output_citation_group(rptr, count, type, fp); + rptr += count ; + rcount -= count; + state = NORMAL; + } + break; + case PENDING_LABEL_POST: + if (PRE_LABEL_MARKER == c) { + state = PENDING_LABEL_POST_PRE; + return 1; + } + else { + output_citation_group(rptr, count, type, fp); + rptr += count; + rcount -= count; + put_string(post_label, fp); + state = NORMAL; + } + break; + case PENDING_LABEL_POST_PRE: + if (c >= LABEL_MARKER + && c < LABEL_MARKER + N_LABEL_TYPES + && c - LABEL_MARKER == type) { + count += 1; + state = PENDING_LABEL; + return 1; + } + else { + output_citation_group(rptr, count, type, fp); + rptr += count; + rcount -= count; + put_string(sep_label, fp); + state = NORMAL; + } + break; + case PENDING_POST: + if (PRE_LABEL_MARKER == c) { + put_string(sep_label, fp); + state = NORMAL; + return 1; + } + else { + put_string(post_label, fp); + state = NORMAL; + } + break; + } + return 0; +} + +void label_processing_state::process(int c) +{ + if (handle_pending(c)) + return; + assert(state == NORMAL); + switch (c) { + case PRE_LABEL_MARKER: + put_string(pre_label, fp); + state = NORMAL; + break; + case POST_LABEL_MARKER: + state = PENDING_POST; + break; + case LABEL_MARKER: + case LABEL_MARKER + 1: + count = 1; + state = PENDING_LABEL; + type = label_type(c - LABEL_MARKER); + break; + default: + state = NORMAL; + putc(c, fp); + break; + } +} + +extern "C" { + +int rcompare(const void *p1, const void *p2) +{ + return compare_reference(**(reference **)p1, **(reference **)p2); +} + +} + +void output_references() +{ + assert(accumulate); + if (!hash_table_size) { + if (have_bibliography) + error("nothing to reference (probably 'bibliography' before" + " 'sort')"); + accumulate = 0; + nreferences = 0; + return; + } + if (nreferences > 0) { + int j = 0; + int i; + for (i = 0; i < hash_table_size; i++) + if (reference_hash_table[i] != 0) + reference_hash_table[j++] = reference_hash_table[i]; + assert(j == nreferences); + for (; j < hash_table_size; j++) + reference_hash_table[j] = 0; + qsort(reference_hash_table, nreferences, sizeof(reference*), + rcompare); + for (i = 0; i < nreferences; i++) + reference_hash_table[i]->set_number(i); + compute_labels(reference_hash_table, nreferences); + } + if (outfp != stdout) { + rewind(outfp); + { + label_processing_state state(citation, ncitations, stdout); + int c; + while ((c = getc(outfp)) != EOF) + state.process(c); + } + ncitations = 0; + fclose(outfp); + outfp = stdout; + } + if (nreferences > 0) { + fputs(".]<\n", outfp); + for (int i = 0; i < nreferences; i++) { + if (sort_fields.length() > 0) + reference_hash_table[i]->print_sort_key_comment(outfp); + if (label_in_reference) { + fputs(".ds [F ", outfp); + const string &label + = reference_hash_table[i]->get_label(NORMAL_LABEL); + if (label.length() > 0 + && (label[0] == ' ' || label[0] == '\\' || label[0] == '"')) + putc('"', outfp); + put_string(label, outfp); + putc('\n', outfp); + } + reference_hash_table[i]->output(outfp); + delete reference_hash_table[i]; + reference_hash_table[i] = 0; + } + fputs(".]>\n", outfp); + nreferences = 0; + } + clear_labels(); +} + +static reference *find_reference(const char *query, int query_len) +{ + // This is so that error messages look better. + while (query_len > 0 && csspace(query[query_len - 1])) + query_len--; + string str; + for (int i = 0; i < query_len; i++) + str += query[i] == '\n' ? ' ' : query[i]; + str += '\0'; + possibly_load_default_database(); + search_list_iterator iter(&database_list, str.contents()); + reference_id rid; + const char *start; + int len; + if (!iter.next(&start, &len, &rid)) { + error("no matches for '%1'", str.contents()); + return 0; + } + const char *end = start + len; + while (start < end) { + if (*start == '%') + break; + while (start < end && *start++ != '\n') + ; + } + if (start >= end) { + error("found a reference for '%1' but it didn't contain any fields", + str.contents()); + return 0; + } + reference *result = new reference(start, end - start, &rid); + if (iter.next(&start, &len, &rid)) + warning("multiple matches for '%1'", str.contents()); + return result; +} + +static reference *make_reference(const string &str, unsigned *flagsp) +{ + const char *start = str.contents(); + const char *end = start + str.length(); + const char *ptr = start; + while (ptr < end) { + if (*ptr == '%') + break; + while (ptr < end && *ptr++ != '\n') + ; + } + *flagsp = 0; + for (; start < ptr; start++) { + if (*start == '#') + *flagsp = (SHORT_LABEL | (*flagsp & (FORCE_RIGHT_BRACKET + | FORCE_LEFT_BRACKET))); + else if (*start == '[') + *flagsp |= FORCE_LEFT_BRACKET; + else if (*start == ']') + *flagsp |= FORCE_RIGHT_BRACKET; + else if (!csspace(*start)) + break; + } + if (start >= end) { + error("empty reference"); + return new reference; + } + reference *database_ref = 0; + if (start < ptr) + database_ref = find_reference(start, ptr - start); + reference *inline_ref = 0; + if (ptr < end) + inline_ref = new reference(ptr, end - ptr); + if (inline_ref) { + if (database_ref) { + database_ref->merge(*inline_ref); + delete inline_ref; + return database_ref; + } + else + return inline_ref; + } + else if (database_ref) + return database_ref; + else + return new reference; +} + +static void do_ref(const string &str) +{ + if (accumulate) + (void)store_reference(str); + else { + (void)immediately_handle_reference(str); + immediately_output_references(); + } +} + +static void trim_blanks(string &str) +{ + const char *start = str.contents(); + const char *end = start + str.length(); + while (end > start && end[-1] != '\n' && csspace(end[-1])) + --end; + str.set_length(end - start); +} + +void do_bib(const char *filename) +{ + FILE *fp; + if (strcmp(filename, "-") == 0) + fp = stdin; + else { + errno = 0; + fp = fopen(filename, "r"); + if (fp == 0) { + error("can't open '%1': %2", filename, strerror(errno)); + return; + } + current_filename = filename; + } + current_lineno = 1; + enum { + START, MIDDLE, BODY, BODY_START, BODY_BLANK, BODY_DOT + } state = START; + string body; + for (;;) { + int c = getc(fp); + if (EOF == c) + break; + if (is_invalid_input_char(c)) { + error("invalid input character code %1", c); + continue; + } + switch (state) { + case START: + if ('%' == c) { + body = c; + state = BODY; + } + else if (c != '\n') + state = MIDDLE; + break; + case MIDDLE: + if ('\n' == c) + state = START; + break; + case BODY: + body += c; + if ('\n' == c) + state = BODY_START; + break; + case BODY_START: + if ('\n' == c) { + do_ref(body); + state = START; + } + else if ('.' == c) + state = BODY_DOT; + else if (csspace(c)) { + state = BODY_BLANK; + body += c; + } + else { + body += c; + state = BODY; + } + break; + case BODY_BLANK: + if ('\n' == c) { + trim_blanks(body); + do_ref(body); + state = START; + } + else if (csspace(c)) + body += c; + else { + body += c; + state = BODY; + } + break; + case BODY_DOT: + if (']' == c) { + do_ref(body); + state = MIDDLE; + } + else { + body += '.'; + body += c; + state = ('\n' == c) ? BODY_START : BODY; + } + break; + default: + assert(0 == "unhandled case while parsing bibliography file"); + } + if ('\n' == c) + current_lineno++; + } + switch (state) { + case START: + case MIDDLE: + break; + case BODY: + body += '\n'; + do_ref(body); + break; + case BODY_DOT: + case BODY_START: + do_ref(body); + break; + case BODY_BLANK: + trim_blanks(body); + do_ref(body); + break; + } + fclose(fp); +} + +// from the Dragon Book + +unsigned hash_string(const char *s, int len) +{ + const char *end = s + len; + unsigned h = 0, g; + while (s < end) { + h <<= 4; + h += *s++; + if ((g = h & 0xf0000000) != 0) { + h ^= g >> 24; + h ^= g; + } + } + return h; +} + +int next_size(int n) +{ + static const int table_sizes[] = { + 101, 503, 1009, 2003, 3001, 4001, 5003, 10007, 20011, 40009, + 80021, 160001, 500009, 1000003, 2000003, 4000037, 8000009, + 16000057, 32000011, 64000031, 128000003, 0 + }; + + const int *p; + for (p = table_sizes; *p <= n && *p != 0; p++) + ; + assert(*p != 0); + return *p; +} + +// Local Variables: +// fill-column: 72 +// mode: C++ +// End: +// vim: set cindent noexpandtab shiftwidth=2 textwidth=72: |