aboutsummaryrefslogtreecommitdiff
path: root/web/webqtl/showTrait/exportPage.py
diff options
context:
space:
mode:
Diffstat (limited to 'web/webqtl/showTrait/exportPage.py')
-rwxr-xr-xweb/webqtl/showTrait/exportPage.py141
1 files changed, 0 insertions, 141 deletions
diff --git a/web/webqtl/showTrait/exportPage.py b/web/webqtl/showTrait/exportPage.py
deleted file mode 100755
index ff3f12a1..00000000
--- a/web/webqtl/showTrait/exportPage.py
+++ /dev/null
@@ -1,141 +0,0 @@
-import string
-import os
-import re
-import cPickle
-import pyXLWriter as xl
-
-from base import webqtlConfig
-from utility import webqtlUtil
-from base.templatePage import templatePage
-
-class ExportPage(templatePage):
-
- def __init__(self, fd):
-
- templatePage.__init__(self, fd)
-
- filename = webqtlUtil.genRandStr("Export_")
- workbook = xl.Writer('%s.xls' % (webqtlConfig.TMPDIR+filename))
- style_formats = [] #Array with Excel style formats - Zach 9/2/2011
- heading = workbook.add_format(align = 'center', bold = 1, border = 1, size=13, fg_color = 0x1E, color="white") #Style for the header cells
- right = workbook.add_format(align = 'right') #Style to align cell contents to the right
- style_formats.append(heading)
- style_formats.append(right)
- worksheet = workbook.add_worksheet()
-
- primaryStrainNames = fd.formdata.getvalue('strainNames', '').split(',')
- primaryVals = fd.formdata.getvalue('strainVals', '').split(',')
- primaryVars = fd.formdata.getvalue('strainVars', '').split(',')
- otherStrainNames = fd.formdata.getvalue('otherStrainNames', '').split(',')
- otherVals = fd.formdata.getvalue('otherStrainVals', '').split(',')
- otherVars = fd.formdata.getvalue('otherStrainVars', '').split(',')
- attributeData = fd.formdata.getvalue('extra_attributes', '')
- otherAttributeData = fd.formdata.getvalue('other_extra_attributes', '')
-
- #ZS: This section is to parse the attribute formdata string
- attributeTypes = attributeData.split('/')
- otherAttributeTypes = otherAttributeData.split('/')
-
- attributeNames = []
- attributeVals = []
- for i in range(len(attributeTypes)):
- if i < len(attributeTypes) - 1:
- attributeNames.append(attributeTypes[i].split(':')[0])
- attributeVals.append(attributeTypes[i].split(':')[1].split(','))
- else:
- break
-
- otherAttributeNames = []
- otherAttributeVals = []
- for i in range(len(otherAttributeTypes)):
- if i < len(otherAttributeTypes) - 1:
- otherAttributeNames.append(otherAttributeTypes[i].split(':')[0])
- otherAttributeVals.append(otherAttributeTypes[i].split(':')[1].split(','))
- else:
- break
-
- varsExist = 0 #ZS: Even if there are no variances "primaryVars" would still be populated with empty values, so we need to check if there really are any
- for i in range(len(primaryVars)):
- if primaryVars[i] != '':
- varsExist = 1
- break
-
- otherStrainsExist = 0 #ZS: Same as above; checking to see if there's a set of "other" (non-primary) strains
- for i in range(len(otherStrainNames)):
- if otherStrainNames[i] != '':
- otherStrainsExist = 1
- break
-
- if varsExist == 1:
- column_headers = ["Sample", "Value", " SE "] #ZS: Names of the header for each column in the excel worksheet
- else:
- column_headers = ["Sample", "Value"]
-
-
- for attr_name in attributeNames:
- column_headers.append(attr_name)
-
- start_line = 0 #Gets last line of "primary" strain values to define a start-point for "other" strain values
- for ncol, item in enumerate(column_headers):
- worksheet.write([start_line, ncol], item, style_formats[0])
- worksheet.set_column([ncol, ncol], 2*len(item))
-
- start_line += 1
- last_line = start_line
-
- for i in range(len(primaryStrainNames)):
- ncol = 0
- if varsExist == 1:
- for ncol, item in enumerate([primaryStrainNames[i], primaryVals[i], primaryVars[i]]):
- worksheet.write([start_line + i, ncol], item, style_formats[1])
- ncol += 1
- else:
- for ncol, item in enumerate([primaryStrainNames[i], primaryVals[i]]):
- worksheet.write([start_line + i, ncol], item, style_formats[1])
- ncol += 1
-
- for attribute_type in attributeVals:
- worksheet.write([start_line + i, ncol], attribute_type[i], style_formats[1])
- ncol += 1
-
- last_line += 1
-
- if otherStrainsExist == 1:
- start_line = last_line + 2
-
- for ncol, item in enumerate(column_headers):
- worksheet.write([start_line, ncol], item, style_formats[0])
- worksheet.set_column([ncol, ncol], 2*len(item))
- start_line += 1
-
- for i in range(len(otherStrainNames)):
- ncol = 0
- if varsExist == 1:
- for ncol, item in enumerate([otherStrainNames[i], otherVals[i], otherVars[i]]):
- worksheet.write([start_line + i, ncol], item, style_formats[1])
- ncol += 1
- else:
- for ncol, item in enumerate([otherStrainNames[i], otherVals[i]]):
- worksheet.write([start_line + i, ncol], item, style_formats[1])
-
- for attribute_type in otherAttributeVals:
- worksheet.write([start_line + i, ncol], attribute_type[i], style_formats[1])
- ncol += 1
-
- workbook.close()
-
- full_filename = os.path.join(webqtlConfig.TMPDIR, '%s.xls' % filename)
- fp = open(full_filename, 'rb')
- text = fp.read()
- fp.close()
-
- self.content_type = 'application/xls'
- self.content_disposition = 'attachment; filename=%s' % ('%s.xls' % filename)
- self.attachment = text
-
-
-
-
-
-
-