static void batch_process_some_linkages(Label label,
Sentence sent,
Command_Options* copts)
{
Parse_Options opts = copts->popts;
if (there_was_an_error(label, sent, opts))
{
/* If we found at least one good linkage, print it. */
if (sentence_num_valid_linkages(sent) > 0) {
Linkage linkage = NULL;
int i;
for (i=0; i<sentence_num_linkages_post_processed(sent); i++)
{
if (0 == sentence_num_violations(sent, i))
{
linkage = linkage_create(i, sent, opts);
break;
}
}
process_linkage(linkage, copts);
linkage_delete(linkage);
}
fprintf(stdout, "+++++ error %d\n", batch_errors);
}
else
{
if (test_enabled(test, "batch-print-parse-statistics"))
{
print_parse_statistics(sent, opts, copts);
}
}
}
static void wordgraph_unlink_xtmpfile(void)
{
char *fn;
if (!test_enabled("gvfile"))
{
concatfn(fn, TMPDIR, DOT_FILENAME);
if (unlink(fn) == -1)
prt_error("Warning: Cannot unlink %s: %s\n", fn, strerror(errno));
}
}
/**
* Check for the auto-next-linkage test request (for LG code development).
* It is given using the special command: test=auto-next-linkage[:display_max]
* when :display_max is an optional indication of the maximum number of
* linkages to auto-display (the default is DISPLAY_MAX).
* For example, to issue up to 20000 linkages for each batch sentence,
* the following can be used:
* link-parser -limit=30000 -test=auto-next-linkage:20000 < file.batch
*/
static int auto_next_linkage_test(const char *test_opt)
{
const char *auto_next_linkage_pos =
test_enabled(test_opt, "auto-next-linkage");
int max_display = 0;
if (auto_next_linkage_pos == NULL) return 0;
if (':' == auto_next_linkage_pos[0])
max_display = atoi(auto_next_linkage_pos + 1);
if (max_display != 0) return max_display;
return DISPLAY_MAX;
}
static bool
pushpop(GLuint bits, const char *test)
{
bool pushpop_affects = (bits & (GL_ENABLE_BIT |
GL_COLOR_BUFFER_BIT)) != 0;
printf("%s test:\n", test);
glEnable(GL_FRAMEBUFFER_SRGB);
glPushAttrib(bits);
glDisable(GL_FRAMEBUFFER_SRGB);
glPopAttrib();
if (!test_enabled(pushpop_affects))
return false;
/* Now, test the bits the other direction. */
glDisable(GL_FRAMEBUFFER_SRGB);
glPushAttrib(bits);
glEnable(GL_FRAMEBUFFER_SRGB);
glPopAttrib();
return test_enabled(!pushpop_affects);
}
/**
* Display the word-graph in the indicated mode.
* This is for debug. It is not reentrant due to the static pid and the
* possibly created fixed filenames.
* When Using X11, a "dot -Txlib" program is launched on the graph
* description file. The xlib driver refreshes the graph when the file is
* changed, displaying additional sentences in the same window. The viewer
* program exits on program end (see the comments in the code). When
* compiled with MSVC or MINGW, the system PhotoViewer is used by default,
* unless !wg=x is used (for using X11 when available).
*
* The "dot" and the "PhotoViewer" programs must be in the PATH.
*
* FIXME? "dot" may get a SEGV due to memory corruptions in it (a known
* problem - exists even in 2.38). This can be worked-around by trying it
* again until it succeeds (but the window size, if changed by the user,
* will not be preserved).
*
* modestr: a graph display mode as defined in wordgraph.h (default "ldu").
*/
void wordgraph_show(Sentence sent, const char *modestr)
{
String *wgd;
char *gvf_name;
bool generate_gvfile = test_enabled("gvfile"); /* keep it for debug */
char *wgds;
bool gvfile = false;
unsigned int mode = 0;
const char *mp;
/* No check is done for correct flags - at most "mode" will be nonsense. */
for (mp = modestr; '\0' != *mp && ',' != *mp; mp++) mode |= 1<<(*mp-'a');
/* test=wg: sets the mode to ":" (0x2000000) and thus no flags are set. */
if ((0 == mode) || (WGR_X11 == mode))
mode |= WGR_LEGEND|WGR_DBGLABEL|WGR_UNSPLIT;
wgd = wordgraph2dot(sent, mode, modestr);
wgds = string_copy(wgd);
string_delete(wgd);
#if defined(HAVE_FORK) && !defined(POPEN_DOT)
gvfile = true;
#endif
if (gvfile || generate_gvfile)
{
FILE *gvf;
bool gvf_error = false;
static bool wordgraph_unlink_xtmpfile_needed = true;
concatfn(gvf_name, TMPDIR, DOT_FILENAME);
gvf = fopen(gvf_name, "w");
if (NULL == gvf)
{
prt_error("Error: wordgraph_show: open %s failed: %s",
gvf_name, strerror(errno));
}
else
{
if (fprintf(gvf, "%s", wgds) == -1)
{
gvf_error = true;
prt_error("Error: wordgraph_show: print to %s failed: %s",
gvf_name, strerror(errno));
}
if (fclose(gvf) == EOF)
{
gvf_error = true;
prt_error("Error: wordgraph_show: close %s failed: %s",
gvf_name, strerror(errno));
}
}
if (gvf_error && gvfile) /* we need it - cannot continue */
{
free(wgds);
return;
}
if (wordgraph_unlink_xtmpfile_needed)
{
/* The filename is fixed - removal needed only once. */
wordgraph_unlink_xtmpfile_needed = false;
atexit(wordgraph_unlink_xtmpfile);
}
}
#if !defined HAVE_FORK || defined POPEN_DOT
x_popen((mode & WGR_X11)? POPEN_DOT_CMD : POPEN_DOT_CMD_WINDOWS, wgds);
#else
{
const char *const args[] = { DOT_COMMAND, DOT_DRIVER, gvf_name, NULL };
x_forkexec(args, &pid);
}
#endif
free(wgds);
}
int main(int argc, char * argv[])
{
FILE *input_fh = stdin;
Dictionary dict;
const char *language = NULL;
int num_linkages;
Label label = NO_LABEL;
Command_Options *copts;
Parse_Options opts;
bool batch_in_progress = false;
isatty_stdin = isatty(fileno(stdin));
isatty_stdout = isatty(fileno(stdout));
#ifdef _WIN32
/* If compiled with MSVC/MinGW, we still support running under Cygwin.
* This is done by checking running_under_cygwin to resolve
* incompatibilities. */
const char *ostype = getenv("OSTYPE");
if ((NULL != ostype) && (0 == strcmp(ostype, "cygwin")))
running_under_cygwin = true;
/* argv encoding is in the current locale. */
argv = argv2utf8(argc);
if (NULL == argv)
{
prt_error("Fatal error: Unable to parse command line\n");
exit(-1);
}
#ifdef _MSC_VER
_set_printf_count_output(1); /* enable %n support for display_1line_help()*/
#endif /* _MSC_VER */
win32_set_utf8_output();
#endif /* _WIN32 */
#if LATER
/* Try to catch the SIGWINCH ... except this is not working. */
struct sigaction winch_act;
winch_act.sa_handler = winch_handler;
winch_act.sa_sigaction = NULL;
sigemptyset (&winch_act.sa_mask);
winch_act.sa_flags = 0;
sigaction (SIGWINCH, &winch_act, NULL);
#endif
copts = command_options_create();
if (copts == NULL || copts->panic_opts == NULL)
{
prt_error("Fatal error: unable to create parse options\n");
exit(-1);
}
opts = copts->popts;
setup_panic_parse_options(copts->panic_opts);
copts->panic_mode = true;
parse_options_set_max_parse_time(opts, 30);
parse_options_set_linkage_limit(opts, 1000);
parse_options_set_min_null_count(opts, 0);
parse_options_set_max_null_count(opts, 0);
parse_options_set_short_length(opts, 16);
parse_options_set_islands_ok(opts, false);
parse_options_set_display_morphology(opts, false);
save_default_opts(copts); /* Options so far are the defaults */
if ((argc > 1) && (argv[1][0] != '-')) {
/* The dictionary is the first argument if it doesn't begin with "-" */
language = argv[1];
}
for (int i = 1; i < argc; i++)
{
if (strcmp("--help", argv[i]) == 0)
{
print_usage(stdout, argv[0], copts, 0);
}
if (strcmp("--version", argv[i]) == 0)
{
printf("Version: %s\n", linkgrammar_get_version());
printf("%s\n", linkgrammar_get_configuration());
exit(0);
}
}
/* Process command line variable-setting commands (only). */
for (int i = 1; i < argc; i++)
{
if (argv[i][0] == '-')
{
const char *var = argv[i] + ((argv[i][1] != '-') ? 1 : 2);
if ((var[0] != '!') && (0 > issue_special_command(var, copts, NULL)))
print_usage(stderr, argv[0], copts, -1);
}
else if (i != 1)
{
prt_error("Fatal error: Unknown argument '%s'.\n", argv[i]);
print_usage(stderr, argv[0], copts, -1);
}
}
if (language && *language)
{
dict = dictionary_create_lang(language);
if (dict == NULL)
{
prt_error("Fatal error: Unable to open dictionary.\n");
exit(-1);
}
}
else
{
dict = dictionary_create_default_lang();
if (dict == NULL)
{
prt_error("Fatal error: Unable to open default dictionary.\n");
exit(-1);
}
}
/* Process the command line '!' commands */
for (int i = 1; i < argc; i++)
{
if ((argv[i][0] == '-') && (argv[i][1] == '!'))
{
if (0 > issue_special_command(argv[i]+1, copts, dict))
print_usage(stderr, argv[0], copts, -1);
}
}
check_winsize(copts);
prt_error("Info: Dictionary version %s, locale %s\n",
linkgrammar_get_dict_version(dict),
linkgrammar_get_dict_locale(dict));
prt_error("Info: Library version %s. Enter \"!help\" for help.\n",
linkgrammar_get_version());
/* Main input loop */
while (true)
{
char *input_string;
Sentence sent = NULL;
/* Make sure stderr is shown even when MSVC binary runs under
* Cygwin/MSYS pty (in that case it is fully buffered(!)). */
fflush(stderr);
verbosity = parse_options_get_verbosity(opts);
debug = parse_options_get_debug(opts);
test = parse_options_get_test(opts);
input_string = fget_input_string(input_fh, stdout, /*check_return*/false);
check_winsize(copts);
if (NULL == input_string)
{
if (ferror(input_fh))
prt_error("Error: Read: %s\n", strerror(errno));
if (input_fh == stdin) break;
fclose (input_fh);
input_fh = stdin;
continue;
}
/* Discard whitespace characters from end of string. */
for (char *p = &input_string[strlen(input_string)-1];
(p > input_string) && strchr(WHITESPACE, *p) ; p--)
{
*p = '\0';
}
/* If the input string is just whitespace, then ignore it. */
if (strspn(input_string, WHITESPACE) == strlen(input_string))
continue;
char command = special_command(input_string, copts, dict);
if ('e' == command) break; /* It was an exit command */
if ('c' == command) continue; /* It was another command */
if (-1 == command) continue; /* It was a bad command */
/* We have to handle the !file command inline; it's too hairy
* otherwise ... */
if ('f' == command)
{
char * filename = &input_string[strcspn(input_string, WHITESPACE)] + 1;
int fnlen = strlen(filename);
if (0 == fnlen)
{
prt_error("Error: Missing file name argument\n");
continue;
}
if ('\n' == filename[fnlen-1]) filename[fnlen-1] = '\0';
struct stat statbuf;
if ((0 == stat(filename, &statbuf)) && statbuf.st_mode & S_IFDIR)
{
prt_error("Error: Cannot open %s: %s\n",
filename, strerror(EISDIR));
continue;
}
input_fh = fopen(filename, "r");
if (NULL == input_fh)
{
prt_error("Error: Cannot open %s: %s\n", filename, strerror(errno));
input_fh = stdin;
continue;
}
continue;
}
if (!copts->batch_mode) batch_in_progress = false;
if ('\0' != test[0])
{
/* In batch mode warn only once.
* In auto-next-linkage mode don't warn at all. */
if (!batch_in_progress && !auto_next_linkage_test(test))
{
fflush(stdout);
/* Remind the developer this is a test mode. */
prt_error("Warning: Tests enabled: %s\n", test);
if (copts->batch_mode) batch_in_progress = true;
}
}
if (copts->echo_on)
{
printf("%s\n", input_string);
}
if (copts->batch_mode || auto_next_linkage_test(test))
{
label = strip_off_label(input_string);
}
// Post-processing-based pruning will clip away connectors
// that we might otherwise want to examine. So disable PP
// pruning in this situation.
if (copts->display_bad)
parse_options_set_perform_pp_prune(opts, false);
else
parse_options_set_perform_pp_prune(opts, true);
sent = sentence_create(input_string, dict);
if (sentence_split(sent, opts) < 0)
{
sentence_delete(sent);
sent = NULL;
continue;
}
if (0 != copts->display_wordgraph)
{
const char *wg_display_flags = ""; /* default flags */
switch (copts->display_wordgraph)
{
case 1: /* default flags */
break;
case 2: /* subgraphs with a legend */
wg_display_flags = "sl";
break;
case 3:
{
/* Use esoteric flags from the test user variable. */
const char *s = test_enabled(test, "wg");
if ((NULL != s) && (':' == s[0])) wg_display_flags = s;
}
break;
default:
prt_error("Warning: wordgraph=%d: Unknown value, using 1\n",
copts->display_wordgraph);
copts->display_wordgraph = 1;
}
sentence_display_wordgraph(sent, wg_display_flags);
}
/* First parse with the default disjunct_cost as set by the library
* (currently 2.7). Usually parse here with no null links.
* However, if "-test=one-step-parse" is used and we are said to
* parse with null links, allow parsing here with null links too. */
bool one_step_parse = !copts->batch_mode && copts->allow_null &&
test_enabled(test, "one-step-parse");
int max_null_count = one_step_parse ? sentence_length(sent) : 0;
parse_options_set_min_null_count(opts, 0);
parse_options_set_max_null_count(opts, max_null_count);
parse_options_reset_resources(opts);
num_linkages = sentence_parse(sent, opts);
/* num_linkages is negative only on a hard-error;
* typically, due to a zero-length sentence. */
if (num_linkages < 0)
{
sentence_delete(sent);
sent = NULL;
continue;
}
#if 0
/* Try again, this time omitting the requirement for
* definite articles, etc. This should allow for the parsing
* of newspaper headlines and other clipped speech.
*
* XXX Unfortunately, this also allows for the parsing of
* all sorts of ungrammatical sentences which should not
* parse, and leads to bad parses of many other unparsable
* but otherwise grammatical sentences. Thus, this trick
* pretty much fails; we leave it here to document the
* experiment.
*/
if (num_linkages == 0)
{
parse_options_set_disjunct_cost(opts, 4.5);
num_linkages = sentence_parse(sent, opts);
if (num_linkages < 0) continue;
}
#endif /* 0 */
/* Try using a larger list of disjuncts */
/* XXX FIXME: the lg_expand_disjunct_list() routine is not
* currently a part of the public API; it should be made so,
* or this expansion idea should be abandoned... not sure which.
*/
if ((num_linkages == 0) && parse_options_get_use_cluster_disjuncts(opts))
{
int expanded;
if (verbosity > 0) fprintf(stdout, "No standard linkages, expanding disjunct set.\n");
parse_options_set_disjunct_cost(opts, 3.9);
expanded = lg_expand_disjunct_list(sent);
if (expanded)
{
num_linkages = sentence_parse(sent, opts);
}
if (0 < num_linkages) printf("Got One !!!!!!!!!!!!!!!!!\n");
}
/* If asked to show bad linkages, then show them. */
if ((num_linkages == 0) && (!copts->batch_mode))
{
if (copts->display_bad)
{
num_linkages = sentence_num_linkages_found(sent);
}
}
/* Now parse with null links */
if (!one_step_parse && num_linkages == 0 && !copts->batch_mode)
{
if (verbosity > 0) fprintf(stdout, "No complete linkages found.\n");
if (copts->allow_null)
{
/* XXX should use expanded disjunct list here too */
parse_options_set_min_null_count(opts, 1);
parse_options_set_max_null_count(opts, sentence_length(sent));
num_linkages = sentence_parse(sent, opts);
}
}
if (verbosity > 0)
{
if (parse_options_timer_expired(opts))
fprintf(stdout, "Timer is expired!\n");
if (parse_options_memory_exhausted(opts))
fprintf(stdout, "Memory is exhausted!\n");
}
if ((num_linkages == 0) &&
copts->panic_mode &&
parse_options_resources_exhausted(opts))
{
/* print_total_time(opts); */
batch_errors++;
if (verbosity > 0) fprintf(stdout, "Entering \"panic\" mode...\n");
/* If the parser used was the SAT solver, set the panic parser to
* it too.
* FIXME? Currently, the SAT solver code is not too useful in
* panic mode since it doesn't handle parsing with null words, so
* using the regular parser in that case could be beneficial.
* However, this currently causes a crash due to a memory
* management mess. */
parse_options_set_use_sat_parser(copts->panic_opts,
parse_options_get_use_sat_parser(opts));
parse_options_reset_resources(copts->panic_opts);
parse_options_set_verbosity(copts->panic_opts, verbosity);
(void)sentence_parse(sent, copts->panic_opts);
if (verbosity > 0)
{
if (parse_options_timer_expired(copts->panic_opts))
fprintf(stdout, "Panic timer is expired!\n");
}
}
if (verbosity > 1) parse_options_print_total_time(opts);
const char *rc = "";
if (copts->batch_mode)
{
batch_process_some_linkages(label, sent, copts);
}
else
{
rc = process_some_linkages(input_fh, sent, copts);
}
fflush(stdout);
sentence_delete(sent);
sent = NULL;
if ((NULL == rc) && (input_fh == stdin)) break;
}
if (copts->batch_mode)
{
/* print_time(opts, "Total"); */
fprintf(stderr,
"%d error%s.\n", batch_errors, (batch_errors==1) ? "" : "s");
}
/* Free stuff, so that mem-leak detectors don't complain. */
command_options_delete(copts);
dictionary_delete(dict);
printf ("Bye.\n");
return 0;
}
void compute_chosen_words(Sentence sent, Linkage linkage, Parse_Options opts)
{
WordIdx i; /* index of chosen_words */
WordIdx j;
Disjunct **cdjp = linkage->chosen_disjuncts;
const char **chosen_words = alloca(linkage->num_words * sizeof(*chosen_words));
int *remap = alloca(linkage->num_words * sizeof(*remap));
bool *show_word = alloca(linkage->num_words * sizeof(*show_word));
bool display_morphology = opts->display_morphology;
Gword **lwg_path = linkage->wg_path;
Gword **n_lwg_path = NULL; /* new Wordgraph path, to match chosen_words */
Gword **nullblock_start = NULL; /* start of a null block, to be put in [] */
size_t nbsize = 0; /* number of word in a null block */
Gword *sentence_word;
memset(show_word, 0, linkage->num_words * sizeof(*show_word));
if (verbosity_level(D_CCW))
print_lwg_path(lwg_path, "Linkage");
for (i = 0; i < linkage->num_words; i++)
{
Disjunct *cdj = cdjp[i];
Gword *w; /* current word */
const Gword *nw; /* next word (NULL if none) */
Gword **wgp; /* wordgraph_path traversing pointer */
const char *t = NULL; /* current word string */
bool at_nullblock_end; /* current word is at end of a nullblock */
bool join_alt = false; /* morpheme-join this alternative */
char *s;
size_t l;
size_t m;
lgdebug(D_CCW, "Loop start, word%zu: cdj %s, path %s\n",
i, cdj ? cdj->word_string : "NULL",
lwg_path[i] ? lwg_path[i]->subword : "NULL");
w = lwg_path[i];
nw = lwg_path[i+1];
wgp = &lwg_path[i];
sentence_word = wg_get_sentence_word(sent, w);
/* FIXME If the original word was capitalized in a capitalizable
* position, the displayed null word may be its downcase version. */
if (NULL == cdj) /* a null word (the chosen disjunct was NULL) */
{
chosen_words[i] = NULL;
nbsize++;
if (NULL == nullblock_start) /* it starts a new null block */
nullblock_start = wgp;
at_nullblock_end = (NULL == nw) ||
(wg_get_sentence_word(sent, nw->unsplit_word) != sentence_word);
/* Accumulate null words in this alternative */
if (!at_nullblock_end && (NULL == cdjp[i+1]) &&
((w->morpheme_type == MT_PUNC) == (nw->morpheme_type == MT_PUNC)))
{
lgdebug(D_CCW, "Skipping word%zu cdjp=NULL#%zu, path %s\n",
i, nbsize, w->subword);
chosen_words[i] = NULL;
continue;
}
if (NULL != nullblock_start)
{
/* If we are here, this null word is an end of a null block */
lgdebug(+D_CCW, "Handling %zu null words at %zu: ", nbsize, i);
if (1 == nbsize)
{
/* Case 1: A single null subword. */
lgdebug(D_CCW, "A single null subword.\n");
t = join_null_word(sent, wgp, nbsize);
gwordlist_append(&n_lwg_path, w);
}
else
{
lgdebug(D_CCW, "Combining null subwords");
/* Use alternative_id to check for start of alternative. */
if (((*nullblock_start)->alternative_id == *nullblock_start)
&& at_nullblock_end)
{
/* Case 2: A null unsplit_word (all-nulls alternative).*/
lgdebug(D_CCW, " (null alternative)\n");
t = sentence_word->subword;
gwordlist_append(&n_lwg_path, sentence_word);
}
else
{
/* Case 3: Join together >=2 null morphemes. */
Gword *wgnull;
lgdebug(D_CCW, " (null partial word)\n");
wgnull = wordgraph_null_join(sent, wgp-nbsize+1, wgp);
gwordlist_append(&n_lwg_path, wgnull);
t = wgnull->subword;
}
}
nullblock_start = NULL;
nbsize = 0;
show_word[i] = true;
if (MT_WALL != w->morpheme_type)
{
/* Put brackets around the null word. */
l = strlen(t) + 2;
s = (char *) alloca(l+1);
s[0] = NULLWORD_START;
strcpy(&s[1], t);
s[l-1] = NULLWORD_END;
s[l] = '\0';
t = string_set_add(s, sent->string_set);
lgdebug(D_CCW, " %s\n", t);
/* Null words have no links, so take care not to drop them. */
}
}
}
else
{
/* This word has a linkage. */
/* TODO: Suppress "virtual-morphemes", currently the dictcap ones. */
char *sm;
t = cdj->word_string;
/* Print the subscript, as in "dog.n" as opposed to "dog". */
if (0)
{
/* TODO */
}
else
{
/* Get rid of those ugly ".Ixx" */
if (is_idiom_word(t))
{
s = strdupa(t);
sm = strrchr(s, SUBSCRIPT_MARK); /* Possible double subscript. */
UNREACHABLE(NULL == sm); /* We know it has a subscript. */
*sm = '\0';
t = string_set_add(s, sent->string_set);
}
else if (HIDE_MORPHO)
{
/* Concatenate the word morphemes together into one word.
* Concatenate their subscripts into one subscript.
* Use subscript separator SUBSCRIPT_SEP.
* XXX Check whether we can encounter an idiom word here.
* FIXME Combining contracted words is not handled yet, because
* combining morphemes which have non-LL links to other words is
* not yet implemented.
* FIXME Move to a separate function. */
Gword **wgaltp;
size_t join_len = 0;
size_t mcnt = 0;
/* If the alternative contains morpheme subwords, mark it
* for joining... */
const Gword *unsplit_word = w->unsplit_word;
for (wgaltp = wgp, j = i; NULL != *wgaltp; wgaltp++, j++)
{
if ((*wgaltp)->unsplit_word != unsplit_word) break;
if (MT_INFRASTRUCTURE ==
(*wgaltp)->unsplit_word->morpheme_type) break;
mcnt++;
if (NULL == cdjp[j])
{
/* ... but not if it contains a null word */
join_alt = false;
break;
}
join_len += strlen(cdjp[j]->word_string) + 1;
if ((*wgaltp)->morpheme_type & IS_REG_MORPHEME)
join_alt = true;
}
if (join_alt)
{
/* Join it in two steps: 1. Base words. 2. Subscripts.
* FIXME? Can be done in one step (more efficient but maybe
* less clear).
* Put SUBSCRIPT_SEP between the subscripts.
* XXX No 1-1 correspondence between the hidden base words
* and the subscripts after the join, in case there are base
* words with and without subscripts. */
const char subscript_sep_str[] = { SUBSCRIPT_SEP, '\0'};
char *join = calloc(join_len + 1, 1); /* zeroed out */
join[0] = '\0';
/* 1. Join base words. (Could just use the unsplit_word.) */
for (wgaltp = wgp, m = 0; m < mcnt; wgaltp++, m++)
{
add_morpheme_unmarked(sent, join, cdjp[i+m]->word_string,
(*wgaltp)->morpheme_type);
}
strcat(join, subscript_mark_str()); /* tentative */
/* 2. Join subscripts. */
for (wgaltp = wgp, m = 0; m < mcnt; wgaltp++, m++)
{
/* Cannot NULLify the word - we may have links to it. */
if (m != mcnt-1) chosen_words[i+m] = "";
sm = strchr(cdjp[i+m]->word_string, SUBSCRIPT_MARK);
if (NULL != sm)
{
/* Supposing stem subscript is .=x (x optional) */
if (MT_STEM == (*wgaltp)->morpheme_type)
{
sm += 1 + STEM_MARK_L; /* sm+strlen(".=") */
if ('\0' == *sm) sm = NULL;
#if 0
if ((cnt-1) == m)
{
/* Support a prefix-stem combination. In that case
* we have just nullified the combined word, so we
* need to move it to the position of the prefix.
* FIXME: May still not be good enough. */
move_combined_word = i+m-1;
/* And the later chosen_word assignment should be:
* chosen_words[-1 != move_combined_word ?
* move_combined_word : i] = t;
*/
}
else
{
move_combined_word = -1;
}
#endif
}
}
if (NULL != sm)
{
strcat(join, sm+1);
strcat(join, subscript_sep_str);
}
}
/* Remove an extra mark, if any */
join_len = strlen(join);
if ((SUBSCRIPT_SEP == join[join_len-1]) ||
(SUBSCRIPT_MARK == join[join_len-1]))
join[join_len-1] = '\0';
gwordlist_append(&n_lwg_path, sentence_word);
t = string_set_add(join, sent->string_set);
free(join);
i += mcnt-1;
}
}
}
if (!join_alt) gwordlist_append(&n_lwg_path, *wgp);
/*
* Add guess marks in [] square brackets, if needed, at the
* end of the base word. Convert the badly-printing
* SUBSCRIPT_MARK (hex 03 or ^C) into a period.
*/
if (t)
{
s = strdupa(t);
sm = strrchr(s, SUBSCRIPT_MARK);
if (sm) *sm = SUBSCRIPT_DOT;
if ((!(w->status & WS_GUESS) && (w->status & WS_INDICT))
|| !DISPLAY_GUESS_MARKS)
{
t = string_set_add(s, sent->string_set);
}
else
{
const char *regex_name = w->regex_name;
/* 4 = 1(null) + 1(guess_mark) + 2 (sizeof "[]") */
int baselen = NULL == sm ? strlen(t) : (size_t)(sm-s);
char guess_mark = 0;
switch (w->status & WS_GUESS)
{
case WS_SPELL:
guess_mark = GM_SPELL;
break;
case WS_RUNON:
guess_mark = GM_RUNON;
break;
case WS_REGEX:
guess_mark = GM_REGEX;
break;
case 0:
guess_mark = GM_UNKNOWN;
break;
default:
assert(0, "Missing 'case: %2x'", w->status & WS_GUESS);
}
/* In the case of display_morphology==0, the guess indication of
* the last subword is used as the guess indication of the whole
* word.
* FIXME? The guess indications of other subwords are ignored in
* this mode. This implies that if a first or middle subword has
* a guess indication but the last subword doesn't have, no guess
* indication would be shown at all. */
if ((NULL == regex_name) || HIDE_MORPHO) regex_name = "";
s = alloca(strlen(t) + strlen(regex_name) + 4);
strncpy(s, t, baselen);
s[baselen] = '[';
s[baselen + 1] = guess_mark;
strcpy(s + baselen + 2, regex_name);
strcat(s, "]");
if (NULL != sm) strcat(s, sm);
t = string_set_add(s, sent->string_set);
}
}
}
assert(t != NULL, "Word %zu: NULL", i);
chosen_words[i] = t;
}
/* Conditional test removal of quotation marks and the "capdict" tokens,
* to facilitate using diff on sentence batch runs. */
if (test_enabled("removeZZZ"))
{
for (i=0, j=0; i<linkage->num_links; i++)
{
Link *lnk = &(linkage->link_array[i]);
if (0 == strcmp("ZZZ", lnk->link_name))
chosen_words[lnk->rw] = NULL;
}
}
/* If morphology printing is being suppressed, then all links
* connecting morphemes will be discarded. */
if (HIDE_MORPHO)
{
/* Discard morphology links. */
for (i=0; i<linkage->num_links; i++)
{
Link * lnk = &linkage->link_array[i];
if (is_morphology_link(lnk->link_name))
{
/* Mark link for discarding. */
lnk->link_name = NULL;
}
else
{
/* Mark word for not discarding. */
show_word[lnk->rw] = true;
show_word[lnk->lw] = true;
}
}
}
/* We alloc a little more than needed, but so what... */
linkage->word = (const char **) exalloc(linkage->num_words*sizeof(char *));
/* Copy over the chosen words, dropping the discarded words.
* However, don't discard existing words (chosen_words[i][0]).
* Note that if a word only has morphology links and is not combined with
* another word, then it will get displayed with no links at all (e.g.
* when explicitly specifying root and suffix for debug: root.= =suf */
for (i=0, j=0; i<linkage->num_words; ++i)
{
if (chosen_words[i] &&
(chosen_words[i][0] || (!HIDE_MORPHO || show_word[i])))
{
const char *cwtmp = linkage->word[j];
linkage->word[j] = chosen_words[i];
chosen_words[i] = cwtmp;
remap[i] = j;
j++;
}
else
{
remap[i] = -1;
}
}
linkage->num_words = j;
remap_linkages(linkage, remap); /* Update linkage->link_array / num_links. */
linkage->wg_path_display = n_lwg_path;
if (verbosity_level(D_CCW))
print_lwg_path(n_lwg_path, "Display");
}
