|
|
|
@ -20,6 +20,8 @@ public class CourseTableParser implements Parser<List<Course>> { |
|
|
|
|
|
|
|
|
|
private static final String COURSE_SPLIT_STR = "---------------------"; |
|
|
|
|
|
|
|
|
|
private static final String COURSE_SPLIT_TAG_STR = "</div><div>"; |
|
|
|
|
|
|
|
|
|
private static final Pattern WEEK_RANGE_REGEX = Pattern.compile("(?<startWeek>\\d+)-(?<endWeek>\\d+)\\(周\\)"); |
|
|
|
|
|
|
|
|
|
private static final Pattern SINGLE_WEEK_REGEX = Pattern.compile("(?<week>\\d+)\\(周\\)"); |
|
|
|
@ -38,48 +40,39 @@ public class CourseTableParser implements Parser<List<Course>> { |
|
|
|
|
|
|
|
|
|
List<Course> courses = new ArrayList<>(girds.size()); |
|
|
|
|
|
|
|
|
|
// 解析每个格子
|
|
|
|
|
int girdCount = 0; |
|
|
|
|
for (Element gird : girds) { |
|
|
|
|
girdCount++; |
|
|
|
|
|
|
|
|
|
// 直接获取格子里所有课程的关键字段,每个下表对应格子里相应的课程
|
|
|
|
|
Elements classElements = gird.getElementsByAttributeValue("title", "课堂名称"); |
|
|
|
|
Elements teacherElements = gird.getElementsByAttributeValue("title", "老师"); |
|
|
|
|
Elements timeElements = gird.getElementsByAttributeValue("title", "周次(节次)"); |
|
|
|
|
Elements classroomElements = gird.getElementsByAttributeValue("title", "教室"); |
|
|
|
|
|
|
|
|
|
// 如果一个格子有好几节课,用" --------------------- "切开,提取字段的时候用i指定对应字段的下表
|
|
|
|
|
String[] courseNames = gird.ownText().split(COURSE_SPLIT_STR); |
|
|
|
|
// 解析格子里的所有课
|
|
|
|
|
for (int i = 0; i < courseNames.length; i++) { |
|
|
|
|
// 将分隔符替换成标签,方便解析
|
|
|
|
|
String girdHtml = gird.outerHtml().replace(COURSE_SPLIT_STR, COURSE_SPLIT_TAG_STR); |
|
|
|
|
Elements courseElements = Jsoup.parse(girdHtml).getElementsByTag("div"); |
|
|
|
|
for (Element courseElement : courseElements) { |
|
|
|
|
String courseName = courseElement.ownText(); |
|
|
|
|
|
|
|
|
|
// 格子文本为空,说明这个格子没课,直接跳过这个格子就行了
|
|
|
|
|
if ("".equals(courseNames[i])) { |
|
|
|
|
if ("".equals(courseName)) { |
|
|
|
|
continue; |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
// 直接获取格子里所有课程的关键字段,每个下表对应格子里相应的课程
|
|
|
|
|
Elements classElements = courseElement.getElementsByAttributeValue("title", "课堂名称"); |
|
|
|
|
Elements teacherElements = courseElement.getElementsByAttributeValue("title", "老师"); |
|
|
|
|
Elements timeElements = courseElement.getElementsByAttributeValue("title", "周次(节次)"); |
|
|
|
|
Elements classroomElements = courseElement.getElementsByAttributeValue("title", "教室"); |
|
|
|
|
|
|
|
|
|
Course course = new Course(); |
|
|
|
|
|
|
|
|
|
course.setName(courseNames[i]); |
|
|
|
|
course.setTeachClass(classElements.isEmpty() ? "" : classElements.get(i).text()); |
|
|
|
|
course.setTeacher(teacherElements.isEmpty() ? "" : teacherElements.get(i).text()); |
|
|
|
|
course.setName(courseName); |
|
|
|
|
course.setTeachClass(classElements.isEmpty() ? "" : classElements.get(0).text()); |
|
|
|
|
course.setTeacher(teacherElements.isEmpty() ? "" : teacherElements.get(0).text()); |
|
|
|
|
|
|
|
|
|
ClassRoom classRoom = new ClassRoom(); |
|
|
|
|
classRoom.setRoom(classroomElements.isEmpty() ? "" : classroomElements.get(i).text()); |
|
|
|
|
classRoom.setRoom(classroomElements.isEmpty() ? "" : classroomElements.get(0).text()); |
|
|
|
|
course.setClassroom(classRoom); |
|
|
|
|
|
|
|
|
|
// 靠行位置来确定节次,而不是靠time字段的节次数据确定(因为太不好处理了)
|
|
|
|
|
// 具体算法就是行索引x2 + 1就是开始的节次(索引从0开始)
|
|
|
|
|
// 对于只有一个小节的课程,这类课程多数是在线课程,实际选课的时候照样会和其他课冲突,因此这里一律按照两小节大课处理
|
|
|
|
|
int lineIndex = (girdCount - 1) / 7; |
|
|
|
|
course.setStartSection(lineIndex * 2 + 1); |
|
|
|
|
course.setEndSection(lineIndex * 2 + 2); |
|
|
|
|
|
|
|
|
|
int weekDay = girdCount % 7; |
|
|
|
|
course.setWeekDay(weekDay == 0 ? 7 : weekDay); |
|
|
|
|
|
|
|
|
|
// 提取周次信息,根据老项目,可能会有用","分成两段周次信息,但根据实际测试没有发现类似的课程
|
|
|
|
|
String time = timeElements.isEmpty() ? "" : timeElements.get(i).text(); |
|
|
|
|
// 提取周次信息
|
|
|
|
|
String time = timeElements.isEmpty() ? "" : timeElements.get(0).text(); |
|
|
|
|
Matcher matcher = WEEK_RANGE_REGEX.matcher(time); |
|
|
|
|
if (matcher.find()) { |
|
|
|
|
course.setStartWeek(Integer.parseInt(matcher.group("startWeek"))); |
|
|
|
@ -93,6 +86,15 @@ public class CourseTableParser implements Parser<List<Course>> { |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|
|
|
|
|
|
// 靠行位置来确定节次,而不是靠time字段的节次数据确定(因为太不好处理了)
|
|
|
|
|
// 具体算法就是行索引x2 + 1就是开始的节次(索引从0开始)
|
|
|
|
|
int lineIndex = (int) (girdCount * 0.142); |
|
|
|
|
course.setStartSection(lineIndex * 2 + 1); |
|
|
|
|
course.setEndSection(lineIndex * 2 + 2); |
|
|
|
|
|
|
|
|
|
int weekDay = girdCount % 7; |
|
|
|
|
course.setWeekDay(weekDay == 0 ? 7 : weekDay); |
|
|
|
|
|
|
|
|
|
courses.add(course); |
|
|
|
|
} |
|
|
|
|
} |
|
|
|
|