use of org.apache.poi.hssf.usermodel.HSSFCell in project OpenClinica by OpenClinica.
the class SpreadsheetPreview method createGroupsMap.
public Map<Integer, Map<String, String>> createGroupsMap(HSSFWorkbook workbook) {
if (workbook == null || workbook.getNumberOfSheets() == 0) {
return new HashMap<Integer, Map<String, String>>();
}
HSSFSheet sheet;
HSSFRow row;
HSSFCell cell;
// static group headers for a CRF; TODO: change these so they are not
// static and hard-coded
String[] groupHeaders = { "group_label", "group_layout", "group_header", "group_sub_header", "group_repeat_number", "group_repeat_max", "group_repeat_array", "group_row_start_number" };
Map<String, String> rowCells = new HashMap<String, String>();
SortedMap<Integer, Map<String, String>> allRows = new TreeMap<Integer, Map<String, String>>();
String str = "";
for (int i = 0; i < workbook.getNumberOfSheets(); i++) {
sheet = workbook.getSheetAt(i);
str = workbook.getSheetName(i);
if (str.equalsIgnoreCase("Groups")) {
for (int j = 1; j < sheet.getPhysicalNumberOfRows(); j++) {
// time again.
if (j > 1)
rowCells = new HashMap<String, String>();
row = sheet.getRow(j);
for (int k = 0; k < groupHeaders.length; k++) {
cell = row.getCell((short) k);
if (groupHeaders[k].equalsIgnoreCase("group_header")) {
rowCells.put(groupHeaders[k], getCellValue(cell).replaceAll("<[^>]*>", ""));
} else {
}
}
allRows.put(j, rowCells);
}
// end inner for loop
}
// end if
}
// end outer for
return allRows;
}
use of org.apache.poi.hssf.usermodel.HSSFCell in project OpenClinica by OpenClinica.
the class SpreadsheetPreviewNw method createGroupsMap.
public Map<Integer, Map<String, String>> createGroupsMap(HSSFWorkbook workbook) {
if (workbook == null || workbook.getNumberOfSheets() == 0) {
return new HashMap<Integer, Map<String, String>>();
}
HSSFSheet sheet;
HSSFRow row;
HSSFCell cell;
sheet = workbook.getSheetAt(4);
cell = sheet.getRow(1).getCell((short) 0);
String version = cell.getStringCellValue();
// static group headers for a CRF; TODO: change these so they are not
// static and hard-coded
// BWP>>remove "group_borders" column
String[] groupHeaders = { "group_label", "repeating_group", "group_header", "group_repeat_number", "group_repeat_max" };
if (version.equalsIgnoreCase("Version: 2.2") || version.equalsIgnoreCase("Version: 2.5") || version.equalsIgnoreCase("Version: 3.0")) {
groupHeaders = new String[] { "group_label", "group_header", "group_repeat_number", "group_repeat_max" };
}
Map<String, String> rowCells = new HashMap<String, String>();
SortedMap<Integer, Map<String, String>> allRows = new TreeMap<Integer, Map<String, String>>();
String str;
for (int i = 0; i < workbook.getNumberOfSheets(); i++) {
sheet = workbook.getSheetAt(i);
str = workbook.getSheetName(i);
if (str.equalsIgnoreCase("Groups")) {
for (int j = 1; j < sheet.getPhysicalNumberOfRows(); j++) {
// time again.
if (j > 1)
rowCells = new HashMap<String, String>();
row = sheet.getRow(j);
for (int k = 0; k < groupHeaders.length; k++) {
cell = row.getCell((short) k);
if (groupHeaders[k].equalsIgnoreCase("group_header")) {
rowCells.put(groupHeaders[k], getCellValue(cell));
} else {
rowCells.put(groupHeaders[k], getCellValue(cell).replaceAll("<[^>]*>", ""));
}
}
allRows.put(j, rowCells);
}
// end inner for loop
}
// end if
}
// end outer for
return allRows;
}
use of org.apache.poi.hssf.usermodel.HSSFCell in project OpenClinica by OpenClinica.
the class SpreadsheetPreviewNw method createCrfMap.
/*
* This method searches for a sheet named "Sections" in an Excel Spreadsheet object, then creates a HashMap containing that sheet's data. The HashMap
* contains the sheet name as the key, and a List of cells (only the ones that contain data, not blank ones). This method was created primarly to get the
* section names for a CRF preview page. The Map does not contain data for any sections that have duplicate names; just one section per section name. This
* method does not yet validate the spreadsheet as a CRF.
* @author Bruce Perry @returns A HashMap containing CRF section names as keys. Returns an empty HashMap if the spreadsheet does not contain any sheets
* named "Sections."
*/
public Map<String, String> createCrfMap(HSSFWorkbook workbook) {
if (workbook == null || workbook.getNumberOfSheets() == 0) {
return new HashMap<String, String>();
}
HSSFSheet sheet;
HSSFRow row;
HSSFCell cell;
Map<String, String> crfInfo = new HashMap<String, String>();
String mapKey = "";
String val = "";
String str = "";
String[] crfHeaders = { "crf_name", "version", "version_description", "revision_notes" };
for (int i = 0; i < workbook.getNumberOfSheets(); i++) {
sheet = workbook.getSheetAt(i);
str = workbook.getSheetName(i);
if (str.equalsIgnoreCase("CRF")) {
row = sheet.getRow(1);
for (int k = 0; k < crfHeaders.length; k++) {
// The first cell in the row contains the header CRF_NAME
mapKey = crfHeaders[k];
cell = row.getCell((short) k);
if (cell != null) {
switch(cell.getCellType()) {
case HSSFCell.CELL_TYPE_STRING:
val = cell.getStringCellValue();
break;
case HSSFCell.CELL_TYPE_NUMERIC:
val = Double.toString(cell.getNumericCellValue());
break;
case HSSFCell.CELL_TYPE_BOOLEAN:
val = Boolean.toString(cell.getBooleanCellValue());
break;
case HSSFCell.CELL_TYPE_FORMULA:
cell.getCellFormula();
break;
}
}
crfInfo.put(mapKey, val);
}
}
// end if
}
// end outer for
return crfInfo;
}
use of org.apache.poi.hssf.usermodel.HSSFCell in project OpenClinica by OpenClinica.
the class SpreadsheetPreviewNw method isDateDatatype.
private boolean isDateDatatype(String[] headers, HSSFRow row) {
HSSFCell cell;
String currentDataType = "";
if (headers == null || headers.length == 0 || row == null) {
return false;
}
for (int k = 0; k < headers.length; k++) {
cell = row.getCell((short) k);
if ("data_type".equalsIgnoreCase(headers[k])) {
currentDataType = getCellValue(cell);
if ("date".equalsIgnoreCase(currentDataType)) {
return true;
}
}
}
return false;
}
use of org.apache.poi.hssf.usermodel.HSSFCell in project OpenClinica by OpenClinica.
the class SpreadsheetPreviewNw method createItemsOrSectionMap.
/**
* This method searches for a sheet named "Items" or "Sections" in an Excel Spreadsheet object, then creates a sorted Map whose members represent a row of
* data for each "Item" or "Section" on the sheet. This method was created primarily to get Items and section data for previewing a CRF.
*
* @return A SortedMap implementation (TreeMap) containing row numbers, each pointing to a Map. The Maps represent each Item or section row in a
* spreadsheet. The items or sections themselves are in rows 1..N. An example data value from a Section row is: 1: {page_number=1.0,
* section_label=Subject Information, section_title=SimpleSection1} Returns an empty Map if the spreadsheet does not contain any sheets named
* "sections" or "items" (case insensitive).
* @param workbook
* is an object representing a spreadsheet.
* @param itemsOrSection
* should specify "items" or "sections" or the associated static variable, i.e. SpreadsheetPreview.ITEMS
*/
public Map<Integer, Map<String, String>> createItemsOrSectionMap(HSSFWorkbook workbook, String itemsOrSection) {
if (workbook == null || workbook.getNumberOfSheets() == 0) {
return new HashMap<Integer, Map<String, String>>();
}
if (itemsOrSection == null || !itemsOrSection.equalsIgnoreCase(ITEMS) && !itemsOrSection.equalsIgnoreCase(SECTIONS)) {
return new HashMap<Integer, Map<String, String>>();
}
HSSFSheet sheet;
HSSFRow row;
HSSFCell cell;
// static item headers for a CRF; TODO: change these so they are not
// static and hard-coded
/*
* New itemHeaders String[] itemHeaders = {"item_name","description_label","left_item_text",
* "units","right_item_text","section_label","group_label","header", "subheader","parent_item","column_number","page_number",
* "question_number","response_type","response_label", "response_options_text","response_values","response_layout","default_value", "data_type",
* "validation","validation_error_message","phi","required"};
*/
String[] itemHeaders = { "item_name", "description_label", "left_item_text", "units", "right_item_text", "section_label", "group_label", "header", "subheader", "parent_item", "column_number", "page_number", "question_number", "response_type", "response_label", "response_options_text", "response_values", "response_layout", "default_value", "data_type", "width_decimal", "validation", "validation_error_message", "phi", "required" };
String[] sectionHeaders = { "section_label", "section_title", "subtitle", "instructions", "page_number", "parent_section", "borders" };
Map<String, String> rowCells = new HashMap<String, String>();
SortedMap<Integer, Map<String, String>> allRows = new TreeMap<Integer, Map<String, String>>();
String str;
String dateFormat = "";
for (int i = 0; i < workbook.getNumberOfSheets(); i++) {
sheet = workbook.getSheetAt(i);
str = workbook.getSheetName(i);
if (str.equalsIgnoreCase(itemsOrSection)) {
for (int j = 1; j < sheet.getPhysicalNumberOfRows(); j++) {
String[] headers = itemsOrSection.equalsIgnoreCase(ITEMS) ? itemHeaders : sectionHeaders;
// time again.
if (j > 1)
rowCells = new HashMap<String, String>();
row = sheet.getRow(j);
if (row == null) {
continue;
}
for (int k = 0; k < headers.length; k++) {
cell = row.getCell((short) k);
if ("default_value".equalsIgnoreCase(headers[k]) && isDateDatatype(headers, row) && !"".equalsIgnoreCase(getCellValue(cell))) {
try {
// BWP>> getDateCellValue() wll throw an
// exception if
// the value is an invalid date. Keep the date
// format the same as
// it is in the database; MM/dd/yyyy
// String pttrn = ResourceBundleProvider.getFormatBundle().getString("oc_date_format_string");
String pttrn = ApplicationConstants.getDateFormatInItemData();
dateFormat = new SimpleDateFormat(pttrn).format(cell.getDateCellValue());
rowCells.put(headers[k], dateFormat);
continue;
} catch (Exception e) {
String cellVal = getCellValue(cell);
logger.info("An invalid date format was encountered when reading a default value in the spreadsheet.");
rowCells.put(headers[k], cellVal);
continue;
}
}
if (headers[k].equalsIgnoreCase("left_item_text") || headers[k].equalsIgnoreCase("right_item_text") || headers[k].equalsIgnoreCase("header") || headers[k].equalsIgnoreCase("subheader") || headers[k].equalsIgnoreCase("question_number") || headers[k].equalsIgnoreCase("section_title") || headers[k].equalsIgnoreCase("subtitle") || headers[k].equalsIgnoreCase("instructions") || headers[k].equalsIgnoreCase("response_options_text")) {
rowCells.put(headers[k], getCellValue(cell).replaceAll("\\\\,", "\\,"));
} else {
rowCells.put(headers[k], getCellValue(cell).replaceAll("\\\\,", "\\,").replaceAll("<[^>]*>", ""));
}
// logger.warn("BADS: "+headers[k]+": "+getCellValue(cell));
}
// item_name
allRows.put(j, rowCells);
}
// end inner for loop
}
// end if
}
// end outer for
return allRows;
}
Aggregations