use of org.sagacity.sqltoy.config.model.SummaryGroupMeta in project sagacity-sqltoy by chenrenfei.
the class GroupSummaryTest method testSummary.
@Test
public void testSummary() {
Object[][] values = new Object[][] { { "202101", "手机", 100, 2000 }, { "202101", "电脑", 90, null }, { "202102", "手机", 80, 1700 }, { "202102", "电脑", 60, 7900 } };
List dataSet = new ArrayList();
for (int i = 0; i < values.length; i++) {
List row = new ArrayList();
for (int j = 0; j < values[i].length; j++) {
row.add(values[i][j]);
}
dataSet.add(row);
}
SummaryColMeta[] colMetas = new SummaryColMeta[2];
SummaryColMeta colMeta1 = new SummaryColMeta();
colMeta1.setColIndex(2);
colMeta1.setAveSkipNull(true);
colMeta1.setSummaryType(1);
SummaryColMeta colMeta2 = new SummaryColMeta();
colMeta2.setColIndex(3);
colMeta2.setAveSkipNull(true);
colMeta2.setSummaryType(3);
colMetas[0] = colMeta1;
colMetas[1] = colMeta2;
SummaryColMeta[] colMetas1 = new SummaryColMeta[2];
SummaryColMeta colMeta11 = new SummaryColMeta();
colMeta11.setColIndex(2);
colMeta11.setAveSkipNull(true);
colMeta11.setSummaryType(1);
SummaryColMeta colMeta21 = new SummaryColMeta();
colMeta21.setColIndex(3);
colMeta21.setAveSkipNull(true);
colMeta21.setSummaryType(3);
colMetas1[0] = colMeta11;
colMetas1[1] = colMeta21;
SummaryGroupMeta[] groupMetas = new SummaryGroupMeta[2];
SummaryGroupMeta globalMeta = new SummaryGroupMeta();
globalMeta.setLabelIndex(0);
globalMeta.setAverageTitle("平均值");
globalMeta.setSumTitle("总计");
globalMeta.setSumSite("top");
globalMeta.setRowSize(2);
globalMeta.setSummaryType(3);
globalMeta.setSummaryCols(colMetas);
groupMetas[0] = globalMeta;
SummaryGroupMeta groupMeta = new SummaryGroupMeta();
groupMeta.setGroupCols(new Integer[] { 0 });
// groupMeta.setAverageTitle("平均值");
groupMeta.setSumTitle("小计");
groupMeta.setLabelIndex(0);
globalMeta.setSummaryType(1);
groupMeta.setRowSize(1);
groupMeta.setSummaryCols(colMetas1);
groupMeta.setSumSite("left");
groupMetas[1] = groupMeta;
CollectionUtil.groupSummary(dataSet, groupMetas, false, null);
for (int i = 0; i < dataSet.size(); i++) {
System.err.println(JSON.toJSONString(dataSet.get(i)));
}
}
use of org.sagacity.sqltoy.config.model.SummaryGroupMeta in project sagacity-sqltoy by chenrenfei.
the class SqlXMLConfigParse method parseCalculator.
/**
* @todo 解析对sqltoy查询结果的计算处理逻辑定义(包含:旋转、汇总等)
* @param sqlToyConfig
* @param sqlElt
* @param local
* @throws Exception
*/
private static void parseCalculator(SqlToyConfig sqlToyConfig, Element sqlElt, String local) throws Exception {
NodeList elements = sqlElt.getChildNodes();
Element elt;
String eltName;
List resultProcessor = new ArrayList();
for (int i = 0; i < elements.getLength(); i++) {
if (elements.item(i).getNodeType() == Node.ELEMENT_NODE) {
elt = (Element) elements.item(i);
eltName = elt.getNodeName();
// 旋转(只能进行一次旋转)
if (eltName.equals(local.concat("pivot"))) {
PivotModel pivotModel = new PivotModel();
if (elt.hasAttribute("group-columns")) {
pivotModel.setGroupCols(trimParams(elt.getAttribute("group-columns").toLowerCase().split("\\,")));
}
if (elt.hasAttribute("category-columns")) {
pivotModel.setCategoryCols(trimParams(elt.getAttribute("category-columns").toLowerCase().split("\\,")));
}
if (elt.hasAttribute("category-sql")) {
pivotModel.setCategorySql(elt.getAttribute("category-sql"));
}
String[] startEndCols = new String[2];
startEndCols[0] = elt.getAttribute("start-column").toLowerCase();
if (elt.hasAttribute("end-column")) {
startEndCols[1] = elt.getAttribute("end-column").toLowerCase();
} else {
startEndCols[1] = startEndCols[0];
}
if (elt.hasAttribute("default-value")) {
String defaultValue = elt.getAttribute("default-value");
if (elt.hasAttribute("default-type")) {
String defaultType = elt.getAttribute("default-type").toLowerCase();
try {
pivotModel.setDefaultValue(BeanUtil.convertType(defaultValue, defaultType));
} catch (Exception e) {
e.printStackTrace();
}
} else {
pivotModel.setDefaultValue(defaultValue);
}
}
pivotModel.setStartEndCols(startEndCols);
resultProcessor.add(pivotModel);
} else // 列转行
if (eltName.equals(local.concat("unpivot"))) {
UnpivotModel unpivotModel = new UnpivotModel();
XMLUtil.setAttributes(elt, unpivotModel);
if (unpivotModel.getColumnsToRows().length > 1) {
resultProcessor.add(unpivotModel);
}
} else // 汇总合计
if (eltName.equals(local.concat("summary"))) {
SummaryModel summaryModel = new SummaryModel();
// 是否逆向汇总
if (elt.hasAttribute("reverse")) {
summaryModel.setReverse(Boolean.parseBoolean(elt.getAttribute("reverse")));
}
// 汇总合计涉及的列
if (elt.hasAttribute("sum-columns")) {
summaryModel.setSummaryCols(elt.getAttribute("sum-columns").toLowerCase());
} else if (elt.hasAttribute("columns")) {
summaryModel.setSummaryCols(elt.getAttribute("columns").toLowerCase());
}
// 计算平均值的列
if (elt.hasAttribute("average-columns")) {
summaryModel.setAverageCols(elt.getAttribute("average-columns").toLowerCase());
}
// 保留小数点位数(2022-2-23 扩展成数组,便于给不同平均值列设置不同的小数位)
if (elt.hasAttribute("average-radix-sizes")) {
summaryModel.setRadixSize(trimParamsToInt(elt.getAttribute("average-radix-sizes").split("\\,")));
} else if (elt.hasAttribute("radix-size")) {
summaryModel.setRadixSize(trimParamsToInt(elt.getAttribute("radix-size").split("\\,")));
}
if (elt.hasAttribute("average-rounding-modes")) {
String[] roundingModeAry = trimParams(elt.getAttribute("average-rounding-modes").toUpperCase().split("\\,"));
RoundingMode[] roudingModes = new RoundingMode[roundingModeAry.length];
String roundingMode;
RoundingMode roundMode = null;
for (int k = 0; k < roundingModeAry.length; k++) {
roundingMode = roundingModeAry[k];
if (roundingMode.equals("HALF_UP")) {
roundMode = RoundingMode.HALF_UP;
} else if (roundingMode.equals("HALF_DOWN")) {
roundMode = RoundingMode.HALF_DOWN;
} else if (roundingMode.equals("ROUND_DOWN")) {
roundMode = RoundingMode.DOWN;
} else if (roundingMode.equals("ROUND_UP")) {
roundMode = RoundingMode.UP;
} else {
roundMode = RoundingMode.HALF_UP;
}
roudingModes[k] = roundMode;
}
summaryModel.setRoundingModes(roudingModes);
}
// 汇总所在位置
if (elt.hasAttribute("sum-site")) {
summaryModel.setSumSite(elt.getAttribute("sum-site"));
}
// sum和average值左右拼接时的连接字符串
if (elt.hasAttribute("link-sign")) {
summaryModel.setLinkSign(elt.getAttribute("link-sign"));
}
// 求平均时是否过滤掉null的记录
if (elt.hasAttribute("average-skip-null")) {
summaryModel.setAverageSkipNull(Boolean.parseBoolean(elt.getAttribute("average-skip-null")));
}
NodeList nodeList = elt.getElementsByTagName(local.concat("global"));
List<SummaryGroupMeta> groupMetaList = new ArrayList<SummaryGroupMeta>();
// 全局汇总
if (nodeList.getLength() > 0) {
SummaryGroupMeta globalMeta = new SummaryGroupMeta();
Element globalSummary = (Element) nodeList.item(0);
if (globalSummary.hasAttribute("label-column")) {
globalMeta.setLabelColumn(globalSummary.getAttribute("label-column").toLowerCase());
}
if (globalSummary.hasAttribute("average-label")) {
globalMeta.setAverageTitle(globalSummary.getAttribute("average-label"));
}
// 汇总分组列
if (globalSummary.hasAttribute("group-column")) {
globalMeta.setGroupColumn(globalSummary.getAttribute("group-column").toLowerCase());
}
if (globalSummary.hasAttribute("sum-label")) {
globalMeta.setSumTitle(globalSummary.getAttribute("sum-label"));
}
if (globalSummary.hasAttribute("reverse")) {
globalMeta.setGlobalReverse(Boolean.parseBoolean(globalSummary.getAttribute("reverse")));
}
if (summaryModel.isReverse()) {
globalMeta.setGlobalReverse(false);
}
groupMetaList.add(globalMeta);
}
// 分组汇总
nodeList = elt.getElementsByTagName(local.concat("group"));
if (nodeList.getLength() > 0) {
Element groupElt;
for (int j = 0; j < nodeList.getLength(); j++) {
groupElt = (Element) nodeList.item(j);
SummaryGroupMeta groupMeta = new SummaryGroupMeta();
groupMeta.setGroupColumn(groupElt.getAttribute("group-column").toLowerCase());
if (groupElt.hasAttribute("average-label")) {
groupMeta.setAverageTitle(groupElt.getAttribute("average-label"));
}
if (groupElt.hasAttribute("sum-label")) {
groupMeta.setSumTitle(groupElt.getAttribute("sum-label"));
}
if (groupElt.hasAttribute("label-column")) {
groupMeta.setLabelColumn(groupElt.getAttribute("label-column"));
}
groupMetaList.add(groupMeta);
}
}
if (!groupMetaList.isEmpty()) {
SummaryGroupMeta[] groupMetas = new SummaryGroupMeta[groupMetaList.size()];
groupMetaList.toArray(groupMetas);
summaryModel.setGroupMeta(groupMetas);
}
resultProcessor.add(summaryModel);
} else // 列与列进行比较
if (eltName.equals(local.concat("cols-chain-relative"))) {
ColsChainRelativeModel colsRelativeModel = new ColsChainRelativeModel();
XMLUtil.setAttributes(elt, colsRelativeModel);
resultProcessor.add(colsRelativeModel);
} else // 行与行进行比较
if (eltName.equals(local.concat("rows-chain-relative"))) {
RowsChainRelativeModel rowsRelativeModel = new RowsChainRelativeModel();
XMLUtil.setAttributes(elt, rowsRelativeModel);
resultProcessor.add(rowsRelativeModel);
} else // 集合数据顺序颠倒
if (eltName.equals(local.concat("reverse"))) {
ReverseModel reverseModel = new ReverseModel();
XMLUtil.setAttributes(elt, reverseModel);
resultProcessor.add(reverseModel);
}
}
}
// 加入sqlToyConfig
sqlToyConfig.setResultProcessor(resultProcessor);
}
use of org.sagacity.sqltoy.config.model.SummaryGroupMeta in project sagacity-sqltoy by chenrenfei.
the class GroupSummary method process.
public static void process(SummaryModel summaryModel, LabelIndexModel labelIndexMap, List result) {
// 记录小于2条无需汇总计算
if (result == null || result.size() < 2) {
return;
}
// 计算的列,columns="1..result.width()-1"
int dataWidth = ((List) result.get(0)).size();
List<Integer> sumColList = parseColumns(labelIndexMap, summaryModel.getSummaryCols(), dataWidth);
List<Integer> aveColList = parseColumns(labelIndexMap, summaryModel.getAverageCols(), dataWidth);
Set<Integer> summaryColsSet = new LinkedHashSet<Integer>();
for (Integer index : sumColList) {
summaryColsSet.add(index);
}
for (Integer index : aveColList) {
summaryColsSet.add(index);
}
// 未设置分组和汇总计算列信息
if (summaryModel.getGroupMeta() == null || summaryModel.getGroupMeta().length == 0 || summaryColsSet.size() == 0) {
throw new IllegalArgumentException("summary计算未正确配置sum-columns或average-columns或group分组信息!");
}
// 全部计算列
Integer[] summaryCols = new Integer[summaryColsSet.size()];
summaryColsSet.toArray(summaryCols);
// 同时存在求和、求平均
boolean bothSumAverage = !sumColList.isEmpty() && !aveColList.isEmpty();
// 组织分组配置
String sumSite;
for (SummaryGroupMeta groupMeta : summaryModel.getGroupMeta()) {
sumSite = (summaryModel.getSumSite() == null) ? "top" : summaryModel.getSumSite().toLowerCase();
List<Integer> groupColsList = parseColumns(labelIndexMap, groupMeta.getGroupColumn(), dataWidth);
Integer[] groupCols = new Integer[groupColsList.size()];
groupColsList.toArray(groupCols);
// 分组列
groupMeta.setGroupCols(groupCols);
if (bothSumAverage) {
if (StringUtil.isNotBlank(groupMeta.getSumTitle()) && StringUtil.isNotBlank(groupMeta.getAverageTitle())) {
groupMeta.setSummaryType(3);
} else if (StringUtil.isNotBlank(groupMeta.getAverageTitle())) {
groupMeta.setSummaryType(2);
} else // summaryType默认为1即sum计算
{
groupMeta.setSummaryType(1);
}
} else if (!sumColList.isEmpty()) {
groupMeta.setSummaryType(1);
} else if (!aveColList.isEmpty()) {
groupMeta.setSummaryType(2);
}
groupMeta.setSumSite(sumSite);
// 分组的标题列
groupMeta.setLabelIndex(NumberUtil.isInteger(groupMeta.getLabelColumn()) ? Integer.parseInt(groupMeta.getLabelColumn()) : labelIndexMap.get(groupMeta.getLabelColumn().toLowerCase()));
// 汇总和求平均分两行组装,update 2022-2-28 增加每个分组是否同时有汇总标题和求平均标题,允许不同分组只有汇总或求平均
if (groupMeta.getSummaryType() == 3 && (sumSite.equals("top") || sumSite.equals("bottom"))) {
groupMeta.setRowSize(2);
}
groupMeta.setSummaryCols(createColMeta(summaryCols, summaryModel, sumColList, aveColList));
}
CollectionUtil.groupSummary(result, summaryModel.getGroupMeta(), summaryModel.isReverse(), summaryModel.getLinkSign());
}
use of org.sagacity.sqltoy.config.model.SummaryGroupMeta in project sagacity-sqltoy by chenrenfei.
the class CollectionUtil method groupSummary.
/**
* @TODO 分组汇总计算
* @param sumData
* @param groupMetas
* @param isReverse
* @param linkSign
*/
public static void groupSummary(List sumData, SummaryGroupMeta[] groupMetas, boolean isReverse, String linkSign) {
// 分组计算,数据集合少于2条没有必要计算
if (sumData == null || sumData.size() < 2 || groupMetas == null || groupMetas.length == 0) {
return;
}
// 内部子分组自动加上父级分组的列,规避重复
Set<Integer> groupCols = new LinkedHashSet<Integer>();
for (SummaryGroupMeta groupMeta : groupMetas) {
if (groupMeta.getGroupCols() != null && groupMeta.getGroupCols().length > 0) {
for (int index : groupMeta.getGroupCols()) {
groupCols.add(index);
}
}
// 子分组合并父分组的列
if (groupCols.size() > 0) {
Integer[] cols = new Integer[groupCols.size()];
groupCols.toArray(cols);
groupMeta.setGroupCols(cols);
}
// 转小写
if (groupMeta.getSumSite() != null) {
groupMeta.setSumSite(groupMeta.getSumSite().toLowerCase());
} else {
groupMeta.setSumSite("");
}
}
// 进行数据逆转,然后统一按照顺序计算,结果再进行逆向处理
if (isReverse) {
Collections.reverse(sumData);
for (SummaryGroupMeta groupMeta : groupMetas) {
if (groupMeta.getSumSite() != null) {
// sum和ave 是两行上下模式,逆序
if ("top".equals(groupMeta.getSumSite())) {
groupMeta.setSumSite("bottom");
} else if ("bottom".equals(groupMeta.getSumSite())) {
groupMeta.setSumSite("top");
}
}
}
}
summaryList(sumData, groupMetas, StringUtil.isBlank(linkSign) ? " / " : linkSign);
// 将结果反转
if (isReverse) {
Collections.reverse(sumData);
}
}
use of org.sagacity.sqltoy.config.model.SummaryGroupMeta in project sagacity-sqltoy by chenrenfei.
the class CollectionUtil method summaryList.
/**
* @TODO 进行汇总计算
* @param dataSet
* @param groupMetas
* @param linkSign
*/
private static void summaryList(List<List> dataSet, SummaryGroupMeta[] groupMetas, String linkSign) {
List<List> iterList = new ArrayList();
for (List item : dataSet) {
iterList.add(item);
}
int groupSize = groupMetas.length;
SummaryGroupMeta groupMeta;
List row;
List preRow = iterList.get(0);
int dataSize = iterList.size();
int addRows = 0;
List sumRows;
for (int i = 0; i < dataSize; i++) {
row = iterList.get(i);
// 从最明细分组开始(从里而外)
for (int j = groupSize; j > 0; j--) {
groupMeta = groupMetas[j - 1];
// 判断分组字段值是否相同
if (isEquals(row, preRow, i, groupMeta.getGroupCols())) {
// 汇总计算
calculateTotal(row, groupMeta);
} else {
sumRows = createSummaryRow(preRow, groupMeta, linkSign);
// 插入汇总行(可能存在sum、ave 两行数据)
dataSet.addAll(i + addRows, sumRows);
// 累加增加的记录行数
addRows = addRows + sumRows.size();
// 重置分组的列计算的汇总相关的值(sum、rowCount、nullRowCount)
for (SummaryColMeta colMeta : groupMeta.getSummaryCols()) {
colMeta.setNullCount(0);
colMeta.setSumValue(BigDecimal.ZERO);
colMeta.setRowCount(0);
}
calculateTotal(row, groupMeta);
}
// 最后一行
if (i == dataSize - 1) {
// 全局汇总的置顶
if (groupMeta.isGlobalReverse()) {
dataSet.add(0, createSummaryRow(row, groupMeta, linkSign));
} else {
dataSet.addAll(createSummaryRow(row, groupMeta, linkSign));
}
}
}
preRow = row;
}
}
Aggregations