Files
Upsilon/apps/code/script.cpp
2022-06-17 12:12:39 +02:00

119 lines
3.5 KiB
C++

#include "script.h"
#include "script_store.h"
namespace Code {
static inline void intToText(int i, char * buffer, int bufferSize) {
// We only support integers from 0 to 99.
assert(i >= 0);
assert(i < 100);
assert(bufferSize >= 3);
if (i/10 == 0) {
buffer[0] = i+'0';
buffer[1] = 0;
return;
}
buffer[0] = i/10+'0';
buffer[1] = i-10*(i/10)+'0';
buffer[2] = 0;
}
bool Script::DefaultName(char buffer[], size_t bufferSize) {
assert(bufferSize >= k_defaultScriptNameMaxSize);
static constexpr char defaultScriptName[] = "script";
static constexpr int defaultScriptNameLength = 6;
strlcpy(buffer, defaultScriptName, bufferSize);
int currentScriptNumber = 1;
while (currentScriptNumber <= k_maxNumberOfDefaultScriptNames) {
// Change the number in the script name.
intToText(currentScriptNumber, &buffer[defaultScriptNameLength], bufferSize - defaultScriptNameLength );
if (ScriptStore::ScriptNameIsFree(buffer)) {
return true;
}
currentScriptNumber++;
}
// We did not find a new script name
return false;
}
bool Script::nameCompliant(const char * name) {
/* We allow here the empty script name ".py", because it is the name used to
* create a new empty script. When naming or renaming a script, we check
* elsewhere that the name is no longer empty.
* The name format is ([a-z_][a-z0-9_]*)*\.py
*
* We do not allow upper cases in the script names because script names are
* used in the URLs of the NumWorks workshop website and we do not want
* problems with case sensitivity. */
UTF8Decoder decoder(name);
CodePoint c = decoder.nextCodePoint();
if (c == UCodePointNull || !(c.isLatinSmallLetter() || c == '_' || c == '.')) {
/* The name cannot be empty. Its first letter must be in [a-z_] or the
* extension dot. */
return false;
}
while (c != UCodePointNull) {
if (c == '.' && strcmp(decoder.stringPosition(), ScriptStore::k_scriptExtension) == 0) {
return true;
}
if (!(c.isLatinSmallLetter() || c == '_' || c.isDecimalDigit())) {
return false;
}
c = decoder.nextCodePoint();
}
return false;
}
uint8_t * StatusFromData(Script::Data d) {
return const_cast<uint8_t *>(static_cast<const uint8_t *>(d.buffer));
}
bool Script::autoImportationStatus() const {
return getStatutBit(k_autoImportationStatusMask);
}
void Script::toggleAutoimportationStatus() {
assert(!isNull());
Data d = value();
*StatusFromData(d) ^= k_autoImportationStatusMask;
setValue(d);
}
const char * Script::content() const {
Data d = value();
return ((const char *)d.buffer) + StatusSize();
}
bool Script::fetchedFromConsole() const {
return getStatutBit(k_fetchedFromConsoleMask);
}
void Script::setFetchedFromConsole(bool fetched) {
setStatutBit(k_fetchedFromConsoleMask, k_fetchedFromConsoleOffset, fetched);
}
bool Script::fetchedForVariableBox() const {
return getStatutBit(k_fetchedForVariableBoxMask);
}
void Script::setFetchedForVariableBox(bool fetched) {
setStatutBit(k_fetchedForVariableBoxMask, k_fetchedForVariableBoxOffset, fetched);
}
bool Script::getStatutBit(uint8_t mask) const {
assert(!isNull());
Data d = value();
return ((*StatusFromData(d)) & mask) != 0;
}
void Script::setStatutBit(uint8_t mask, uint8_t offset, bool statusBit) {
assert(!isNull());
Data d = value();
uint8_t * status = StatusFromData(d);
*status = ((*status) & ~mask) | (static_cast<uint8_t>(statusBit) << offset); //TODO Create and use a bit operations library
setValue(d);
}
}