static MVMint32 NFD_and_push_collation_values (MVMThreadContext *tc, MVMCodepoint cp, collation_stack *stack, MVMCodepointIter *ci, char *name) {
MVMNormalizer norm;
MVMCodepoint cp_out;
MVMint32 ready,
result_pos = 0;
MVMCodepoint *result = MVM_malloc(sizeof(MVMCodepoint) * initial_collation_norm_buf_size);
MVMint32 result_size = initial_collation_norm_buf_size;
MVMint64 rtrn = 0;
MVM_unicode_normalizer_init(tc, &norm, MVM_NORMALIZE_NFD);
ready = MVM_unicode_normalizer_process_codepoint(tc, &norm, cp, &cp_out);
if (ready) {
if (result_size <= result_pos + ready)
result = MVM_realloc(result, sizeof(MVMCodepoint) * (result_size += initial_collation_norm_buf_size));
result[result_pos++] = cp_out;
while (0 < --ready)
result[result_pos++] = MVM_unicode_normalizer_get_codepoint(tc, &norm);
}
MVM_unicode_normalizer_eof(tc, &norm);
ready = MVM_unicode_normalizer_available(tc, &norm);
while (ready--) {
if (result_size <= result_pos + ready + 1)
result = MVM_realloc(result, sizeof(MVMCodepoint) * (result_size += initial_collation_norm_buf_size));
result[result_pos++] = MVM_unicode_normalizer_get_codepoint(tc, &norm);
}
/* If the codepoint changed or we now have more than before */
if (result[0] != cp || 1 < result_pos)
rtrn = collation_push_cp(tc, stack, ci, result, result_pos, name);
if (result)
MVM_free(result);
return rtrn;
}
/* In situations where we have hit EOF, we need to decode what's left and flush
* the normalization buffer also. */
static void reached_eof(MVMThreadContext *tc, MVMDecodeStream *ds) {
/* Decode all the things. */
if (ds->bytes_head)
run_decode(tc, ds, NULL, NULL);
/* If there's some things left in the normalization buffer, take them. */
MVM_unicode_normalizer_eof(tc, &(ds->norm));
if (MVM_unicode_normalizer_available(tc, &(ds->norm))) {
MVMint32 ready = MVM_unicode_normalizer_available(tc, &(ds->norm));
MVMGrapheme32 *buffer = MVM_malloc(ready * sizeof(MVMGrapheme32));
MVMint32 count = 0;
while (ready--)
buffer[count++] = MVM_unicode_normalizer_get_grapheme(tc, &(ds->norm));
MVM_string_decodestream_add_chars(tc, ds, buffer, count);
}
}
/* Processes a codepoint that we regard as a "normalization terminator". These
* never have a decomposition, and for all practical purposes will not have a
* combiner on them. We treat them specially so we don't, during I/O, block on
* seeing a codepoint after them, which for things like REPLs that need to see
* input right after a \n makes for problems. */
MVMint32 MVM_unicode_normalizer_process_codepoint_norm_terminator(MVMThreadContext *tc, MVMNormalizer *n, MVMCodepoint in, MVMCodepoint *out) {
/* Add the codepoint into the buffer. */
add_codepoint_to_buffer(tc, n, in);
/* Treat this as an "eof", which really means "normalize what ya got". */
MVM_unicode_normalizer_eof(tc, n);
/* Hand back a normalized codepoint, and the number available (have to
* compensate for the one we steal for *out). */
*out = MVM_unicode_normalizer_get_codepoint(tc, n);
return 1 + MVM_unicode_normalizer_available(tc, n);
}
/* Takes an NFG string and populates the array out, which must be a 32-bit
* integer array, with codepoints normalized according to the specified
* normalization form. */
void MVM_unicode_string_to_codepoints(MVMThreadContext *tc, MVMString *s, MVMNormalization form, MVMObject *out) {
MVMCodepoint *result;
MVMint64 result_pos, result_alloc;
MVMCodepointIter ci;
/* Validate output array and set up result storage. */
assert_codepoint_array(tc, out, "Normalization output must be native array of 32-bit integers");
result_alloc = s->body.num_graphs;
result = MVM_malloc(result_alloc * sizeof(MVMCodepoint));
result_pos = 0;
/* Create codepoint iterator. */
MVM_string_ci_init(tc, &ci, s);
/* If we want NFC, just iterate, since NFG is constructed out of NFC. */
if (form == MVM_NORMALIZE_NFC) {
while (MVM_string_ci_has_more(tc, &ci)) {
maybe_grow_result(&result, &result_alloc, result_pos + 1);
result[result_pos++] = MVM_string_ci_get_codepoint(tc, &ci);
}
}
/* Otherwise, need to feed it through a normalizer. */
else {
MVMNormalizer norm;
MVMint32 ready;
MVM_unicode_normalizer_init(tc, &norm, form);
while (MVM_string_ci_has_more(tc, &ci)) {
MVMCodepoint cp;
ready = MVM_unicode_normalizer_process_codepoint(tc, &norm, MVM_string_ci_get_codepoint(tc, &ci), &cp);
if (ready) {
maybe_grow_result(&result, &result_alloc, result_pos + ready);
result[result_pos++] = cp;
while (--ready > 0)
result[result_pos++] = MVM_unicode_normalizer_get_codepoint(tc, &norm);
}
}
MVM_unicode_normalizer_eof(tc, &norm);
ready = MVM_unicode_normalizer_available(tc, &norm);
maybe_grow_result(&result, &result_alloc, result_pos + ready);
while (ready--)
result[result_pos++] = MVM_unicode_normalizer_get_codepoint(tc, &norm);
MVM_unicode_normalizer_cleanup(tc, &norm);
}
/* Put result into array body. */
((MVMArray *)out)->body.slots.u32 = result;
((MVMArray *)out)->body.start = 0;
((MVMArray *)out)->body.elems = result_pos;
}
/* Takes an object, which must be of VMArray representation and holding
* 32-bit integers. Treats them as Unicode codepoints, normalizes them at
* Grapheme level, and returns the resulting NFG string. */
MVMString * MVM_unicode_codepoints_to_nfg_string(MVMThreadContext *tc, MVMObject *codes) {
MVMNormalizer norm;
MVMCodepoint *input;
MVMGrapheme32 *result;
MVMint64 input_pos, input_codes, result_pos, result_alloc;
MVMint32 ready;
MVMString *str;
/* Get input array; if it's empty, we're done already. */
assert_codepoint_array(tc, codes, "Code points to string input must be native array of 32-bit integers");
input = (MVMCodepoint *)((MVMArray *)codes)->body.slots.u32 + ((MVMArray *)codes)->body.start;
input_codes = ((MVMArray *)codes)->body.elems;
if (input_codes == 0)
return tc->instance->str_consts.empty;
/* Guess output size based on input size. */
result_alloc = input_codes;
result = MVM_malloc(result_alloc * sizeof(MVMCodepoint));
/* Perform normalization at grapheme level. */
MVM_unicode_normalizer_init(tc, &norm, MVM_NORMALIZE_NFG);
input_pos = 0;
result_pos = 0;
while (input_pos < input_codes) {
MVMGrapheme32 g;
ready = MVM_unicode_normalizer_process_codepoint_to_grapheme(tc, &norm, input[input_pos], &g);
if (ready) {
maybe_grow_result(&result, &result_alloc, result_pos + ready);
result[result_pos++] = g;
while (--ready > 0)
result[result_pos++] = MVM_unicode_normalizer_get_grapheme(tc, &norm);
}
input_pos++;
}
MVM_unicode_normalizer_eof(tc, &norm);
ready = MVM_unicode_normalizer_available(tc, &norm);
maybe_grow_result(&result, &result_alloc, result_pos + ready);
while (ready--)
result[result_pos++] = MVM_unicode_normalizer_get_grapheme(tc, &norm);
MVM_unicode_normalizer_cleanup(tc, &norm);
/* Produce an MVMString of the result. */
str = (MVMString *)MVM_repr_alloc_init(tc, tc->instance->VMString);
str->body.storage.blob_32 = result;
str->body.storage_type = MVM_STRING_GRAPHEME_32;
str->body.num_graphs = result_pos;
return str;
}
void MVM_unicode_normalize_codepoints(MVMThreadContext *tc, MVMObject *in, MVMObject *out, MVMNormalization form) {
MVMNormalizer norm;
MVMCodepoint *input;
MVMCodepoint *result;
MVMint64 input_pos, input_codes, result_pos, result_alloc;
MVMint32 ready;
/* Validate input/output array. */
assert_codepoint_array(tc, in, "Normalization input must be native array of 32-bit integers");
assert_codepoint_array(tc, out, "Normalization output must be native array of 32-bit integers");
/* Get input array; if it's empty, we're done already. */
input = (MVMCodepoint *)((MVMArray *)in)->body.slots.u32 + ((MVMArray *)in)->body.start;
input_codes = ((MVMArray *)in)->body.elems;
if (input_codes == 0)
return;
/* Guess output size based on input size. */
result_alloc = input_codes;
result = MVM_malloc(result_alloc * sizeof(MVMCodepoint));
/* Perform normalization. */
MVM_unicode_normalizer_init(tc, &norm, form);
input_pos = 0;
result_pos = 0;
while (input_pos < input_codes) {
MVMCodepoint cp;
ready = MVM_unicode_normalizer_process_codepoint(tc, &norm, input[input_pos], &cp);
if (ready) {
maybe_grow_result(&result, &result_alloc, result_pos + ready);
result[result_pos++] = cp;
while (--ready > 0)
result[result_pos++] = MVM_unicode_normalizer_get_codepoint(tc, &norm);
}
input_pos++;
}
MVM_unicode_normalizer_eof(tc, &norm);
ready = MVM_unicode_normalizer_available(tc, &norm);
maybe_grow_result(&result, &result_alloc, result_pos + ready);
while (ready--)
result[result_pos++] = MVM_unicode_normalizer_get_codepoint(tc, &norm);
MVM_unicode_normalizer_cleanup(tc, &norm);
/* Put result into array body. */
((MVMArray *)out)->body.slots.u32 = result;
((MVMArray *)out)->body.start = 0;
((MVMArray *)out)->body.elems = result_pos;
}
/* Checks if the decode stream is empty. */
MVMint32 MVM_string_decodestream_is_empty(MVMThreadContext *tc, MVMDecodeStream *ds) {
return !(ds->bytes_head || ds->chars_head || MVM_unicode_normalizer_available(tc, &(ds->norm)));
}
/* Decodes all the buffers, producing a string containing all the decoded
* characters. */
MVMString * MVM_string_decodestream_get_all(MVMThreadContext *tc, MVMDecodeStream *ds) {
MVMString *result = (MVMString *)MVM_repr_alloc_init(tc, tc->instance->VMString);
result->body.storage_type = MVM_STRING_GRAPHEME_32;
/* Decode all the things. */
run_decode(tc, ds, NULL, NULL);
/* If there's some things left in the normalization buffer, take them. */
MVM_unicode_normalizer_eof(tc, &(ds->norm));
if (MVM_unicode_normalizer_available(tc, &(ds->norm))) {
MVMint32 ready = MVM_unicode_normalizer_available(tc, &(ds->norm));
MVMGrapheme32 *buffer = MVM_malloc(ready * sizeof(MVMGrapheme32));
MVMint32 count = 0;
while (ready--)
buffer[count++] = MVM_unicode_normalizer_get_grapheme(tc, &(ds->norm));
MVM_string_decodestream_add_chars(tc, ds, buffer, count);
}
/* If there's no codepoint buffer, then return the empty string. */
if (!ds->chars_head) {
result->body.storage.blob_32 = NULL;
result->body.num_graphs = 0;
}
/* If there's exactly one resulting codepoint buffer and we swallowed none
* of it, just use it. */
else if (ds->chars_head == ds->chars_tail && ds->chars_head_pos == 0) {
/* Set up result string. */
result->body.storage.blob_32 = ds->chars_head->chars;
result->body.num_graphs = ds->chars_head->length;
/* Don't free the buffer's memory itself, just the holder, as we
* stole that for the buffer into the string above. */
MVM_free(ds->chars_head);
ds->chars_head = ds->chars_tail = NULL;
}
/* Otherwise, need to assemble all the things. */
else {
/* Calculate length. */
MVMint32 length = 0, pos = 0;
MVMDecodeStreamChars *cur_chars = ds->chars_head;
while (cur_chars) {
if (cur_chars == ds->chars_head)
length += cur_chars->length - ds->chars_head_pos;
else
length += cur_chars->length;
cur_chars = cur_chars->next;
}
/* Allocate a result buffer of the right size. */
result->body.storage.blob_32 = MVM_malloc(length * sizeof(MVMGrapheme32));
result->body.num_graphs = length;
/* Copy all the things into the target, freeing as we go. */
cur_chars = ds->chars_head;
while (cur_chars) {
if (cur_chars == ds->chars_head) {
MVMint32 to_copy = ds->chars_head->length - ds->chars_head_pos;
memcpy(result->body.storage.blob_32 + pos, cur_chars->chars + ds->chars_head_pos,
cur_chars->length * sizeof(MVMGrapheme32));
pos += to_copy;
}
else {
memcpy(result->body.storage.blob_32 + pos, cur_chars->chars,
cur_chars->length * sizeof(MVMGrapheme32));
pos += cur_chars->length;
}
cur_chars = cur_chars->next;
}
ds->chars_head = ds->chars_tail = NULL;
}
return result;
}
static void compute_case_change(MVMThreadContext *tc, MVMGrapheme32 synth_g, MVMNFGSynthetic *synth_info, MVMint32 case_) {
MVMint32 num_result_graphs;
MVMGrapheme32 *result = NULL;
const MVMCodepoint *result_cps = NULL;
/* Transform the base character. */
MVMuint32 num_result_cps = MVM_unicode_get_case_change(tc,
synth_info->codes[synth_info->base_index], case_, &result_cps);
if (num_result_cps == 0 || (num_result_cps == 1 && result_cps[0] == synth_info->codes[synth_info->base_index])) {
/* Base character does not change, so grapheme stays the same. We
* install a non-null sentinel for this case, and set the result
* grapheme count to zero, which indicates no change. */
result = CASE_UNCHANGED;
num_result_graphs = 0;
}
else {
/* We can potentially get multiple graphemes back. We may also get
* into situations where we case change the base and suddenly we
* can normalize the whole thing to a non-synthetic. So, we take
* a trip through the normalizer. We push any codepoints before the
* base in the synthetic (only happens with Prepend codepoints).
* We then push the first codepoint we get back from the case change
* then the codeponits after the base characters (generally Extend
* codepoints).
* Finally we push anything else the case change produced. This should
* do about the right thing for both case changes that produce a
* base and a combiner, and those that produce a base and a base,
* since the normalizer applies canonical combining class sorting. */
MVMNormalizer norm;
MVMint32 i;
MVM_unicode_normalizer_init(tc, &norm, MVM_NORMALIZE_NFG);
if (0 < synth_info->base_index)
MVM_unicode_normalizer_push_codepoints(tc, &norm,
synth_info->codes,
synth_info->base_index);
/* Push the first result on */
MVM_unicode_normalizer_push_codepoints(tc, &norm, result_cps, 1);
/* Push any combiners after that codepoint so the combiners attach to the
* first codepoint of the casechange not the second or more */
MVM_unicode_normalizer_push_codepoints(tc, &norm,
synth_info->codes + synth_info->base_index + 1,
synth_info->num_codes - synth_info->base_index - 1);
if (1 < num_result_cps)
MVM_unicode_normalizer_push_codepoints(tc, &norm,
result_cps + 1,
num_result_cps - 1);
MVM_unicode_normalizer_eof(tc, &norm);
num_result_graphs = MVM_unicode_normalizer_available(tc, &norm);
result = MVM_malloc(num_result_graphs * sizeof(MVMGrapheme32));
for (i = 0; i < num_result_graphs; i++)
result[i] = MVM_unicode_normalizer_get_grapheme(tc, &norm);
MVM_unicode_normalizer_cleanup(tc, &norm);
}
switch (case_) {
case MVM_unicode_case_change_type_upper:
synth_info->case_uc = result;
synth_info->case_uc_graphs = num_result_graphs;
break;
case MVM_unicode_case_change_type_lower:
synth_info->case_lc = result;
synth_info->case_lc_graphs = num_result_graphs;
break;
case MVM_unicode_case_change_type_title:
synth_info->case_tc = result;
synth_info->case_tc_graphs = num_result_graphs;
break;
case MVM_unicode_case_change_type_fold:
synth_info->case_fc = result;
synth_info->case_fc_graphs = num_result_graphs;
break;
default:
MVM_panic(1, "NFG: invalid case change %d", case_);
}
}
static void compute_case_change(MVMThreadContext *tc, MVMGrapheme32 synth, MVMNFGSynthetic *synth_info, MVMint32 case_) {
MVMGrapheme32 *result;
MVMint32 num_result_graphs;
/* Transform the base character. */
const MVMCodepoint *result_cps;
MVMuint32 num_result_cps = MVM_unicode_get_case_change(tc, synth_info->base,
case_, &result_cps);
if (num_result_cps == 0 || *result_cps == synth_info->base) {
/* Base character does not change, so grapheme stays the same. We
* install a non-null sentinel for this case, and set the result
* grapheme count to zero, which indicates no change. */
result = CASE_UNCHANGED;
num_result_graphs = 0;
}
else {
/* We can potentially get multiple graphemes back. We may also get
* into situations where we case change the base and suddenly we
* can normalize the whole thing to a non-synthetic. So, we take
* a trip through the normalizer. Note we push the first thing
* we get back from the case change, then our combiners, and
* finally anything else the case change produced. This should
* do about the right thing for both case changes that produce a
* base and a combiner, and those that produce a base and a base,
* since the normalizer applies Unicode canonical sorting. */
MVMNormalizer norm;
MVMint32 i;
MVM_unicode_normalizer_init(tc, &norm, MVM_NORMALIZE_NFG);
MVM_unicode_normalizer_push_codepoints(tc, &norm, result_cps, 1);
MVM_unicode_normalizer_push_codepoints(tc, &norm, synth_info->combs,
synth_info->num_combs);
if (num_result_cps > 1)
MVM_unicode_normalizer_push_codepoints(tc, &norm, result_cps + 1,
num_result_cps - 1);
MVM_unicode_normalizer_eof(tc, &norm);
num_result_graphs = MVM_unicode_normalizer_available(tc, &norm);
result = MVM_malloc(num_result_graphs * sizeof(MVMGrapheme32));
for (i = 0; i < num_result_graphs; i++)
result[i] = MVM_unicode_normalizer_get_grapheme(tc, &norm);
MVM_unicode_normalizer_cleanup(tc, &norm);
}
switch (case_) {
case MVM_unicode_case_change_type_upper:
synth_info->case_uc = result;
synth_info->case_uc_graphs = num_result_graphs;
break;
case MVM_unicode_case_change_type_lower:
synth_info->case_lc = result;
synth_info->case_lc_graphs = num_result_graphs;
break;
case MVM_unicode_case_change_type_title:
synth_info->case_tc = result;
synth_info->case_tc_graphs = num_result_graphs;
break;
case MVM_unicode_case_change_type_fold:
synth_info->case_fc = result;
synth_info->case_fc_graphs = num_result_graphs;
break;
default:
MVM_panic(1, "NFG: invalid case change %d", case_);
}
}