summaryrefslogtreecommitdiff
path: root/src/parse/properties
diff options
context:
space:
mode:
Diffstat (limited to 'src/parse/properties')
-rw-r--r--src/parse/properties/utils.c68
1 files changed, 34 insertions, 34 deletions
diff --git a/src/parse/properties/utils.c b/src/parse/properties/utils.c
index b0ca311..8047040 100644
--- a/src/parse/properties/utils.c
+++ b/src/parse/properties/utils.c
@@ -48,7 +48,7 @@ css_error parse_colour_specifier(css_language *c,
token->type != CSS_TOKEN_FUNCTION)) {
if (c->sheet->quirks_allowed == false ||
(token != NULL &&
- token->type != CSS_TOKEN_NUMBER &&
+ token->type != CSS_TOKEN_NUMBER &&
token->type != CSS_TOKEN_DIMENSION))
goto invalid;
}
@@ -71,7 +71,7 @@ css_error parse_colour_specifier(css_language *c,
*ctx = orig_ctx;
return error;
- } else if (c->sheet->quirks_allowed &&
+ } else if (c->sheet->quirks_allowed &&
token->type == CSS_TOKEN_NUMBER) {
error = parse_hash_colour(token->idata, result);
if (error == CSS_OK)
@@ -80,7 +80,7 @@ css_error parse_colour_specifier(css_language *c,
*ctx = orig_ctx;
return error;
- } else if (c->sheet->quirks_allowed &&
+ } else if (c->sheet->quirks_allowed &&
token->type == CSS_TOKEN_DIMENSION) {
error = parse_hash_colour(token->idata, result);
if (error == CSS_OK)
@@ -97,16 +97,16 @@ css_error parse_colour_specifier(css_language *c,
for (i = 0; i < 3; i++) {
css_fixed num;
size_t consumed = 0;
- uint8_t *component = i == 0 ? &r
- : i == 1 ? &b : &g;
+ uint8_t *component = i == 0 ? &r
+ : i == 1 ? &g : &b;
int32_t intval;
consumeWhitespace(vector, ctx);
token = parserutils_vector_peek(vector, *ctx);
- if (token == NULL || (token->type !=
- CSS_TOKEN_NUMBER &&
- token->type !=
+ if (token == NULL || (token->type !=
+ CSS_TOKEN_NUMBER &&
+ token->type !=
CSS_TOKEN_PERCENTAGE))
goto invalid;
@@ -116,7 +116,7 @@ css_error parse_colour_specifier(css_language *c,
goto invalid;
num = number_from_lwc_string(token->idata,
- valid == CSS_TOKEN_NUMBER,
+ valid == CSS_TOKEN_NUMBER,
&consumed);
if (consumed != lwc_string_length(token->idata))
goto invalid;
@@ -173,7 +173,7 @@ invalid:
* \return CSS_OK on success,
* CSS_INVALID if the colour name is unknown
*/
-css_error parse_named_colour(css_language *c, lwc_string *data,
+css_error parse_named_colour(css_language *c, lwc_string *data,
uint32_t *result)
{
static const uint32_t colourmap[LAST_COLOUR + 1 - FIRST_COLOUR] = {
@@ -425,7 +425,7 @@ css_error parse_unit_specifier(css_language *c,
size_t len = lwc_string_length(token->idata);
const char *data = lwc_string_data(token->idata);
- error = parse_unit_keyword(data + consumed, len - consumed,
+ error = parse_unit_keyword(data + consumed, len - consumed,
unit);
if (error != CSS_OK) {
*ctx = orig_ctx;
@@ -445,7 +445,7 @@ css_error parse_unit_specifier(css_language *c,
*unit = default_unit;
if (c->sheet->quirks_allowed) {
- /* Also, in quirks mode, we need to cater for
+ /* Also, in quirks mode, we need to cater for
* dimensions separated from their units by whitespace
* (e.g. "0 px")
*/
@@ -488,7 +488,7 @@ css_error parse_unit_specifier(css_language *c,
* \param ptr Pointer to keyword string
* \param len Length, in bytes, of string
* \param unit Pointer to location to receive computed unit
- * \return CSS_OK on success,
+ * \return CSS_OK on success,
* CSS_INVALID on encountering an unknown keyword
*/
css_error parse_unit_keyword(const char *ptr, size_t len, css_unit *unit)
@@ -558,7 +558,7 @@ css_error parse_unit_keyword(const char *ptr, size_t len, css_unit *unit)
*/
css_error comma_list_length(css_language *c,
const parserutils_vector *vector, int *ctx,
- const css_token *token,
+ const css_token *token,
bool (*reserved)(css_language *c, const css_token *ident),
uint32_t *size)
{
@@ -579,17 +579,17 @@ css_error comma_list_length(css_language *c,
while (token != NULL) {
token = parserutils_vector_peek(
vector, *ctx);
- if (token != NULL &&
- token->type !=
+ if (token != NULL &&
+ token->type !=
CSS_TOKEN_IDENT &&
- token->type !=
+ token->type !=
CSS_TOKEN_S) {
break;
}
/* Idents must not be reserved */
- if (token != NULL && token->type ==
- CSS_TOKEN_IDENT &&
+ if (token != NULL && token->type ==
+ CSS_TOKEN_IDENT &&
reserved(c, token)) {
*ctx = orig_ctx;
return CSS_INVALID;
@@ -622,7 +622,7 @@ css_error comma_list_length(css_language *c,
/* Ensure that a valid token follows */
token = parserutils_vector_peek(vector, *ctx);
- if (token == NULL || (token->type != CSS_TOKEN_IDENT &&
+ if (token == NULL || (token->type != CSS_TOKEN_IDENT &&
token->type != CSS_TOKEN_STRING)) {
*ctx = orig_ctx;
return CSS_INVALID;
@@ -666,7 +666,7 @@ css_error comma_list_length(css_language *c,
*/
css_error comma_list_to_bytecode(css_language *c,
const parserutils_vector *vector, int *ctx,
- const css_token *token,
+ const css_token *token,
bool (*reserved)(css_language *c, const css_token *ident),
uint16_t (*get_value)(css_language *c, const css_token *token),
uint8_t **bytecode)
@@ -703,16 +703,16 @@ css_error comma_list_to_bytecode(css_language *c,
while (temp_token != NULL) {
temp_token = parserutils_vector_peek(
vector, temp_ctx);
- if (temp_token != NULL &&
- temp_token->type !=
+ if (temp_token != NULL &&
+ temp_token->type !=
CSS_TOKEN_IDENT &&
- temp_token->type !=
+ temp_token->type !=
CSS_TOKEN_S) {
break;
}
- if (temp_token != NULL &&
- temp_token->type ==
+ if (temp_token != NULL &&
+ temp_token->type ==
CSS_TOKEN_IDENT) {
len += lwc_string_length(
temp_token->idata);
@@ -739,21 +739,21 @@ css_error comma_list_to_bytecode(css_language *c,
p = buf;
/* Populate buffer with string data */
- memcpy(p, lwc_string_data(token->idata),
+ memcpy(p, lwc_string_data(token->idata),
lwc_string_length(token->idata));
p += lwc_string_length(token->idata);
while (token != NULL) {
token = parserutils_vector_peek(
vector, *ctx);
- if (token != NULL && token->type !=
+ if (token != NULL && token->type !=
CSS_TOKEN_IDENT &&
token->type !=
CSS_TOKEN_S) {
break;
}
- if (token != NULL && token->type ==
+ if (token != NULL && token->type ==
CSS_TOKEN_IDENT) {
memcpy(p, lwc_string_data(
token->idata),
@@ -777,20 +777,20 @@ css_error comma_list_to_bytecode(css_language *c,
* from the name we already have */
lerror = lwc_context_intern(
c->sheet->dictionary,
- (char *) buf, p - buf,
+ (char *) buf, p - buf,
&newname);
if (lerror != lwc_error_ok) {
error = css_error_from_lwc_error(
lerror);
goto cleanup;
}
-
+
if (newname == name) {
lwc_context_string_unref(
c->sheet->dictionary,
newname);
}
-
+
name = newname;
/* Only ref 'name' again if the token owns it,
@@ -812,8 +812,8 @@ css_error comma_list_to_bytecode(css_language *c,
memcpy(ptr, &opv, sizeof(opv));
ptr += sizeof(opv);
}
-
- lwc_context_string_ref(c->sheet->dictionary,
+
+ lwc_context_string_ref(c->sheet->dictionary,
token->idata);
memcpy(ptr, &token->idata, sizeof(token->idata));