1
|
|
|
# -*- coding: utf-8 -*- |
2
|
|
|
# |
3
|
|
|
# This file is part of SENAITE.CORE |
4
|
|
|
# |
5
|
|
|
# Copyright 2018 by it's authors. |
6
|
|
|
# Some rights reserved. See LICENSE.rst, CONTRIBUTORS.rst. |
7
|
|
|
|
8
|
|
|
""" Omnia Axios XRF |
9
|
|
|
""" |
10
|
|
|
from datetime import datetime |
11
|
|
|
from bika.lims.utils import to_unicode |
12
|
|
|
from bika.lims import bikaMessageFactory as _ |
13
|
|
|
from bika.lims.exportimport.instruments.resultsimport import \ |
14
|
|
|
AnalysisResultsImporter, InstrumentCSVResultsFileParser |
15
|
|
|
|
16
|
|
|
class AxiosXrfCSVMultiParser(InstrumentCSVResultsFileParser): |
17
|
|
|
|
18
|
|
|
def __init__(self, csv): |
19
|
|
|
InstrumentCSVResultsFileParser.__init__(self, csv) |
20
|
|
|
self._end_header = False |
21
|
|
|
self._columns = [] |
22
|
|
|
self.columns_name = False #To know if the next line contains |
23
|
|
|
#analytic's columns name |
24
|
|
|
|
25
|
|
|
|
26
|
|
|
def _parseline(self, line): |
27
|
|
|
# Process the line differenly if it pertains at header or results block |
28
|
|
|
if self._end_header == False: |
29
|
|
|
sline = line.strip(',') |
30
|
|
|
return self.parse_headerline(sline) |
31
|
|
|
else: |
32
|
|
|
return self.parse_resultline(line) |
33
|
|
|
|
34
|
|
|
def splitLine(self, line): |
35
|
|
|
# If pertains at header it split the line by ':' and then remove ',' |
36
|
|
|
# Else split by ',' and remove blank spaces |
37
|
|
|
if self._end_header == False: |
38
|
|
|
sline = line.split(':') |
39
|
|
|
return [token.strip(',') for token in sline] |
40
|
|
|
|
41
|
|
|
return [token.strip() for token in line.split(',')] |
42
|
|
|
|
43
|
|
|
def csvDate2BikaDate(self,DateTime): |
44
|
|
|
#11/03/2014 14:46:46 --> %d/%m/%Y %H:%M %p |
45
|
|
|
dtobj = datetime.strptime(DateTime,"%d/%m/%Y %H:%M:%S") |
46
|
|
|
return dtobj.strftime("%Y%m%d %H:%M:%S") |
47
|
|
|
|
48
|
|
|
def parse_headerline(self, line): |
49
|
|
|
#Process incoming header line |
50
|
|
|
"""11/03/2014 14:46:46 |
51
|
|
|
PANalytical |
52
|
|
|
Results quantitative - Omnian 2013, |
53
|
|
|
|
54
|
|
|
Selected archive:,Omnian 2013 |
55
|
|
|
Number of results selected:,4 |
56
|
|
|
""" |
57
|
|
|
|
58
|
|
|
# Save each header field (that we know) and its own value in the dict |
59
|
|
|
if line.startswith('Results quantitative'): |
60
|
|
|
line = to_unicode(line) |
61
|
|
|
if len(self._header) == 0: |
62
|
|
|
self.err("Unexpected header format", numline=self._numline) |
63
|
|
|
return -1 |
64
|
|
|
|
65
|
|
|
line = line.replace(',', "") |
66
|
|
|
splitted = line.split(' - ') |
67
|
|
|
self._header['Quantitative'] = splitted[1] |
68
|
|
|
return 1 |
69
|
|
|
|
70
|
|
View Code Duplication |
if line.startswith('Selected archive'): |
|
|
|
|
71
|
|
|
if len(self._header) == 0: |
72
|
|
|
self.err("No header found", numline=self._numline) |
73
|
|
|
return -1 |
74
|
|
|
|
75
|
|
|
splitted = self.splitLine(line) |
76
|
|
|
if len(splitted) > 1: |
77
|
|
|
self._header['Archive'] = splitted[1].replace('"', '').strip() |
78
|
|
|
else: |
79
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
80
|
|
|
return 0 |
81
|
|
|
|
82
|
|
View Code Duplication |
if line.startswith('Number of'): |
|
|
|
|
83
|
|
|
if len(self._header) == 0: |
84
|
|
|
self.err("No header found", numline=self._numline) |
85
|
|
|
return -1 |
86
|
|
|
|
87
|
|
|
splitted = self.splitLine(line) |
88
|
|
|
if len(splitted) > 1: |
89
|
|
|
self._header['NumResults'] = splitted[1].replace('"', '').strip() |
90
|
|
|
else: |
91
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
92
|
|
|
return 0 |
93
|
|
|
|
94
|
|
|
if line.startswith('Seq.'): |
95
|
|
|
if len(self._header) == 0: |
96
|
|
|
self.err("No header found", numline=self._numline) |
97
|
|
|
return -1 |
98
|
|
|
#Grab column names |
99
|
|
|
self._columns = line.split(',') |
100
|
|
|
self._end_header = True |
101
|
|
|
return 1 |
102
|
|
|
|
103
|
|
|
else: |
104
|
|
|
self._header['Date'] = line |
105
|
|
|
return 1 |
106
|
|
|
|
107
|
|
|
|
108
|
|
|
|
109
|
|
|
def parse_resultline(self, line): |
110
|
|
|
# Process incoming results line |
111
|
|
|
if not line.strip(): |
112
|
|
|
return 0 |
113
|
|
|
if line.startswith(',,'): |
114
|
|
|
return 0 |
115
|
|
|
|
116
|
|
|
rawdict = {} |
117
|
|
|
# Split by "," |
118
|
|
|
splitted = self.splitLine(line.strip(";")) |
119
|
|
|
|
120
|
|
|
errors = '' |
121
|
|
|
|
122
|
|
|
# Adjunt separated values from split by ',' |
123
|
|
|
for idx, result in enumerate(splitted): |
124
|
|
|
if result.startswith('"'): |
125
|
|
|
# It means that is the value's firts part |
126
|
|
|
# Consequently we take second part and append both |
127
|
|
|
result = (splitted[idx].strip('"') + "," + splitted[idx+1].strip('"')) |
128
|
|
|
splitted[idx] = result |
129
|
|
|
splitted.remove(splitted[idx+1]) |
130
|
|
|
|
131
|
|
|
result_type = '' |
132
|
|
|
result_sum = '' |
133
|
|
|
for idx, result in enumerate(splitted): |
134
|
|
|
if self._columns[idx] == 'Result type': |
135
|
|
|
result_type = result |
136
|
|
|
elif self._columns[idx].startswith('Sample name'): |
137
|
|
|
rid = result |
138
|
|
|
elif self._columns[idx].startswith('Seq.'): |
139
|
|
|
pass |
140
|
|
|
elif self._columns[idx] == 'Sum': |
141
|
|
|
result_sum = result |
142
|
|
|
else: |
143
|
|
|
rawdict[self._columns[idx]] = {'DefaultResult':result_type, |
144
|
|
|
# Replace to obtain UK values from default |
145
|
|
|
'Concentration':result.replace(',','.'), |
146
|
|
|
'Sum':result_sum} |
147
|
|
|
try: |
148
|
|
|
rawdict['DateTime'] = {'DateTime':self.csvDate2BikaDate(self._header['Date']), |
149
|
|
|
'DefaultValue':'DateTime'} |
150
|
|
|
except: |
151
|
|
|
pass |
152
|
|
|
if not rid: |
|
|
|
|
153
|
|
|
self.err("No Sample defined", numline=self._numline) |
154
|
|
|
return 0 |
155
|
|
|
|
156
|
|
|
self._addRawResult(rid, rawdict, True) |
157
|
|
|
return 0 |
158
|
|
|
|
159
|
|
|
|
160
|
|
|
def getAttachmentFileType(self): |
161
|
|
|
return "PANalytical - Omnia Axios XRF" |
162
|
|
|
|
163
|
|
|
class AxiosXrfCSVParser(InstrumentCSVResultsFileParser): |
164
|
|
|
|
165
|
|
|
def __init__(self, csv): |
166
|
|
|
InstrumentCSVResultsFileParser.__init__(self, csv) |
167
|
|
|
self._end_header = False |
168
|
|
|
self._columns = [] |
169
|
|
|
self.columns_name = False #To know if the next line contains |
170
|
|
|
#analytic's columns name |
171
|
|
|
|
172
|
|
|
def _parseline(self, line): |
173
|
|
|
# Process the line differenly if it pertains at header or results block |
174
|
|
|
if self._end_header == False: |
175
|
|
|
sline = line.strip(',') |
176
|
|
|
return self.parse_headerline(sline) |
177
|
|
|
else: |
178
|
|
|
return self.parse_resultline(line) |
179
|
|
|
|
180
|
|
|
def csvDate2BikaDate(self,DateTime): |
181
|
|
|
#11/03/2014 14:46:46 --> %d/%m/%Y %H:%M %p |
182
|
|
|
dtobj = datetime.strptime(DateTime,"%d/%m/%Y %H:%M:%S") |
183
|
|
|
return dtobj.strftime("%Y%m%d %H:%M:%S") |
184
|
|
|
|
185
|
|
|
def splitLine(self, line): |
186
|
|
|
# If pertains at header it split the line by ':' and then remove ',' |
187
|
|
|
# Else split by ',' and remove blank spaces |
188
|
|
|
if self._end_header == False: |
189
|
|
|
sline = line.split(':') |
190
|
|
|
return [token.strip(',') for token in sline] |
191
|
|
|
|
192
|
|
|
return [token.strip() for token in line.split(',')] |
193
|
|
|
|
194
|
|
|
def parse_headerline(self, line): |
195
|
|
|
#Process incoming header line |
196
|
|
|
""" |
197
|
|
|
29/11/2013 10:15:44 |
198
|
|
|
PANalytical |
199
|
|
|
"Quantification of sample ESFERA CINZA - 1g H3BO3 - 1:0,5 - NO PPC", |
200
|
|
|
|
201
|
|
|
R.M.S.:,"0,035" |
202
|
|
|
Result status:, |
203
|
|
|
Sum before normalization:,"119,5 %" |
204
|
|
|
Normalised to:,"100,0 %" |
205
|
|
|
Sample type:,Pressed powder |
206
|
|
|
Initial sample weight (g):,"2,000" |
207
|
|
|
Weight after pressing (g):,"3,000" |
208
|
|
|
Correction applied for medium:,No |
209
|
|
|
Correction applied for film:,No |
210
|
|
|
Used Compound list:,Oxides |
211
|
|
|
Results database:,omnian 2013 |
212
|
|
|
Results database in:,c:\panalytical\superq\userdata |
213
|
|
|
""" |
214
|
|
|
|
215
|
|
|
if line.startswith('"Quantification of sample') or line.startswith('Quantification of sample'): |
216
|
|
|
line = to_unicode(line) |
217
|
|
|
if len(self._header) == 0: |
218
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
219
|
|
|
return -1 |
220
|
|
|
# Remove non important string and double comas to obtein |
221
|
|
|
# the sample name free |
222
|
|
|
line = line.replace("Quantification of sample ", "") |
223
|
|
|
line = line.replace('"', "") |
224
|
|
|
splitted = line.split(' - ') |
225
|
|
|
|
226
|
|
|
if len(splitted) > 3:# Maybe we don't need this, i could be all the sample's identifier... |
227
|
|
|
self._header['Sample'] = splitted[0].strip(' ') |
228
|
|
|
self._header['Quantity'] = splitted[1] |
229
|
|
|
self._header['????'] = splitted[2]# At present we |
230
|
|
|
# don't know what |
231
|
|
|
# is that |
232
|
|
|
self._header['PPC'] = splitted[3] |
233
|
|
|
|
234
|
|
|
elif len(splitted) == 1: |
235
|
|
|
self._header['Sample'] = splitted[0].replace('Quantification of sample','').strip(' ') |
236
|
|
|
|
237
|
|
|
else: |
238
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
239
|
|
|
return 1 |
240
|
|
|
# Save each header field (that we know) and its own value in the dict |
241
|
|
View Code Duplication |
if line.startswith('R.M.S.'): |
|
|
|
|
242
|
|
|
|
243
|
|
|
if len(self._header) == 0: |
244
|
|
|
self.err("No header found", numline=self._numline) |
245
|
|
|
return -1 |
246
|
|
|
|
247
|
|
|
splitted = self.splitLine(line) |
248
|
|
|
if len(splitted) > 1: |
249
|
|
|
self._header['R.M.S.'] = splitted[1].replace('"', '').strip() |
250
|
|
|
else: |
251
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
252
|
|
|
return 0 |
253
|
|
|
|
254
|
|
View Code Duplication |
if line.startswith('Result status'): |
|
|
|
|
255
|
|
|
if len(self._header) == 0: |
256
|
|
|
self.err("No header found", numline=self._numline) |
257
|
|
|
|
258
|
|
|
splitted = self.splitLine(line) |
259
|
|
|
if len(splitted) > 1: |
260
|
|
|
self._header['Result status'] = splitted[1].replace('"', '').strip() |
261
|
|
|
else: |
262
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
263
|
|
|
|
264
|
|
|
return 0 |
265
|
|
|
|
266
|
|
View Code Duplication |
if line.startswith('Sum before normalization'): |
|
|
|
|
267
|
|
|
if len(self._header) == 0: |
268
|
|
|
self.err("No header found", numline=self._numline) |
269
|
|
|
return -1 |
270
|
|
|
|
271
|
|
|
splitted = self.splitLine(line) |
272
|
|
|
if len(splitted) > 1: |
273
|
|
|
self._header['Sum'] = splitted[1].replace('"', '').strip() |
274
|
|
|
else: |
275
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
276
|
|
|
|
277
|
|
|
return 0 |
278
|
|
|
|
279
|
|
View Code Duplication |
if line.startswith('Normalised to'): |
|
|
|
|
280
|
|
|
if len(self._header) == 0: |
281
|
|
|
self.err("No header found", numline=self._numline) |
282
|
|
|
return -1 |
283
|
|
|
|
284
|
|
|
splitted = self.splitLine(line) |
285
|
|
|
if len(splitted) > 1: |
286
|
|
|
self._header['Normalized'] = splitted[1].replace('"', '').strip() |
287
|
|
|
else: |
288
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
289
|
|
|
|
290
|
|
|
return 0 |
291
|
|
|
|
292
|
|
View Code Duplication |
if line.startswith('Sample type'): |
|
|
|
|
293
|
|
|
if len(self._header) == 0: |
294
|
|
|
self.err("No header found", numline=self._numline) |
295
|
|
|
return -1 |
296
|
|
|
|
297
|
|
|
splitted = self.splitLine(line) |
298
|
|
|
if len(splitted) > 1: |
299
|
|
|
self._header['Sample type'] = splitted[1].strip() |
300
|
|
|
else: |
301
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
302
|
|
|
|
303
|
|
|
return 0 |
304
|
|
|
|
305
|
|
View Code Duplication |
if line.startswith('Initial sample weight (g)'): |
|
|
|
|
306
|
|
|
if len(self._header) == 0: |
307
|
|
|
self.err("No header found", numline=self._numline) |
308
|
|
|
return -1 |
309
|
|
|
|
310
|
|
|
splitted = self.splitLine(line) |
311
|
|
|
if len(splitted) > 1: |
312
|
|
|
self._header['Initial sample weight'] = splitted[1].replace('"', '').strip() |
313
|
|
|
else: |
314
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
315
|
|
|
|
316
|
|
|
return 0 |
317
|
|
|
|
318
|
|
View Code Duplication |
if line.startswith('Weight after pressing (g)'): |
|
|
|
|
319
|
|
|
if len(self._header) == 0: |
320
|
|
|
self.err("No header found", numline=self._numline) |
321
|
|
|
return -1 |
322
|
|
|
|
323
|
|
|
splitted = self.splitLine(line) |
324
|
|
|
if len(splitted) > 1: |
325
|
|
|
self._header['Weight after pressing'] = splitted[1].replace('"', '').strip() |
326
|
|
|
else: |
327
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
328
|
|
|
|
329
|
|
|
return 0 |
330
|
|
|
|
331
|
|
View Code Duplication |
if line.startswith('Correction applied for medium'): |
|
|
|
|
332
|
|
|
if len(self._header) == 0: |
333
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
334
|
|
|
return -1 |
335
|
|
|
|
336
|
|
|
splitted = self.splitLine(line) |
337
|
|
|
if len(splitted) > 1: |
338
|
|
|
self._header['Correction medium'] = splitted[1].replace('"', '').strip() |
339
|
|
|
else: |
340
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
341
|
|
|
|
342
|
|
|
return 0 |
343
|
|
|
|
344
|
|
View Code Duplication |
if line.startswith('Correction applied for film'): |
|
|
|
|
345
|
|
|
if len(self._header) == 0: |
346
|
|
|
self.err("No header found", numline=self._numline) |
347
|
|
|
return -1 |
348
|
|
|
|
349
|
|
|
splitted = self.splitLine(line) |
350
|
|
|
if len(splitted) > 1: |
351
|
|
|
self._header['Correction film'] = splitted[1].replace('"', '').strip() |
352
|
|
|
else: |
353
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
354
|
|
|
|
355
|
|
|
return 0 |
356
|
|
|
|
357
|
|
View Code Duplication |
if line.startswith('Used Compound list'): |
|
|
|
|
358
|
|
|
if len(self._header) == 0: |
359
|
|
|
self.err("No header found", numline=self._numline) |
360
|
|
|
return -1 |
361
|
|
|
|
362
|
|
|
splitted = self.splitLine(line) |
363
|
|
|
if len(splitted) > 1: |
364
|
|
|
self._header['Used compound'] = splitted[1].replace('"', '').strip() |
365
|
|
|
else: |
366
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
367
|
|
|
|
368
|
|
|
return 0 |
369
|
|
View Code Duplication |
if line.startswith('Results database:'): |
|
|
|
|
370
|
|
|
if len(self._header) == 0: |
371
|
|
|
self.err("No header found", numline=self._numline) |
372
|
|
|
return -1 |
373
|
|
|
|
374
|
|
|
splitted = self.splitLine(line) |
375
|
|
|
if len(splitted) > 1: |
376
|
|
|
self._header['Result database'] = splitted[1].replace('"', '').strip() |
377
|
|
|
else: |
378
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
379
|
|
|
|
380
|
|
|
return 0 |
381
|
|
|
|
382
|
|
|
|
383
|
|
|
if self.columns_name: |
384
|
|
|
if len(self._header) == 0: |
385
|
|
|
self.err("No header found", numline=self._numline) |
386
|
|
|
return -1 |
387
|
|
|
|
388
|
|
|
#Grab column names |
389
|
|
|
self._end_header = True |
390
|
|
|
self._columns = self.splitLine(line) |
391
|
|
|
return 1 |
392
|
|
|
|
393
|
|
|
if line.startswith('Results database in'): |
394
|
|
|
if len(self._header) == 0: |
395
|
|
|
self.err("No header found", numline=self._numline) |
396
|
|
|
return -1 |
397
|
|
|
|
398
|
|
|
splitted = self.splitLine(line) |
399
|
|
|
if len(splitted) > 1: |
400
|
|
|
self._header['Database path'] = splitted[1]+splitted[2] |
401
|
|
|
self.columns_name = True |
402
|
|
|
else: |
403
|
|
|
self.warn('Unexpected header format', numline=self._numline) |
404
|
|
|
|
405
|
|
|
return 1 |
406
|
|
|
|
407
|
|
|
else: |
408
|
|
|
self._header['Date'] = line |
409
|
|
|
return 1 |
410
|
|
|
|
411
|
|
|
def parse_resultline(self, line): |
412
|
|
|
# Process incoming results line |
413
|
|
|
if not line.strip(): |
414
|
|
|
return 0 |
415
|
|
|
|
416
|
|
|
rawdict = {} |
417
|
|
|
# Split by "," |
418
|
|
|
splitted = self.splitLine(line.strip(";")) |
419
|
|
|
# Look to know if the first value is an enumerate field |
420
|
|
|
try: |
421
|
|
|
int(splitted[0]) |
422
|
|
|
rawdict["num"] = splitted[0] |
423
|
|
|
splitted = splitted[1:] |
424
|
|
|
except ValueError: |
425
|
|
|
pass |
426
|
|
|
|
427
|
|
|
# Enumerate the list to obtain: [(0,data0),(1,data1),...] |
428
|
|
|
e_splitted = list(enumerate(splitted)) |
429
|
|
|
errors = '' |
430
|
|
|
|
431
|
|
|
com = False |
432
|
|
|
for idx, result in e_splitted: |
433
|
|
|
if result.startswith('"'): |
434
|
|
|
# It means that is the first value part |
435
|
|
|
# Consequently we take second part and append both |
436
|
|
|
result = (e_splitted[idx][1].strip('"') + "," + e_splitted[idx+1][1].strip('"')) |
437
|
|
|
e_splitted[idx] = (idx,result) |
438
|
|
|
e_splitted.remove(e_splitted[idx+1]) |
439
|
|
|
com = True |
440
|
|
|
rawdict[self._columns[idx]] = result |
441
|
|
|
conc = self._columns[idx] # Main value's name |
442
|
|
|
|
443
|
|
|
|
444
|
|
|
elif com:# We have rm the 2nd part value, consequently we |
445
|
|
|
# need to decrement idx |
446
|
|
|
if len(self._columns) <= idx-1: |
447
|
|
|
self.err("Orphan value in column ${index}", |
448
|
|
|
mapping={"index":str(idx + 1)}, |
449
|
|
|
numline=self._numline) |
450
|
|
|
break |
451
|
|
|
# We add and sync the result with its value's name |
452
|
|
|
rawdict[self._columns[idx-1]] = result |
453
|
|
|
|
454
|
|
|
else: |
455
|
|
|
if len(self._columns) <= idx: |
456
|
|
|
self.err("Orphan value in column ${index}", |
457
|
|
|
mapping={"index":str(idx + 1)}, |
458
|
|
|
numline=self._numline) |
459
|
|
|
break |
460
|
|
|
rawdict[self._columns[idx]] = result |
461
|
|
|
|
462
|
|
|
aname = rawdict[self._columns[0]]# The fisrt column is analytic name |
463
|
|
|
if not aname: |
464
|
|
|
self.err("No Analysis Name defined", numline=self._numline) |
465
|
|
|
return 0 |
466
|
|
|
elif aname == "<H>": |
467
|
|
|
# <H> maybe is data error header? We need more examples... |
468
|
|
|
errors = rawdict.get('Compound') |
469
|
|
|
notes = rawdict.get('Calibration') |
470
|
|
|
rawdict['Notes'] = notes |
471
|
|
|
|
472
|
|
|
rid = self._header['Sample'] |
473
|
|
|
if not rid: |
474
|
|
|
self.err("No Sample defined", numline=self._numline) |
475
|
|
|
return 0 |
476
|
|
|
|
477
|
|
|
notes = rawdict.get('Notes', '') |
478
|
|
|
notes = "Notes: %s" % notes if notes else '' |
479
|
|
|
rawdict['DefaultResult'] = conc |
|
|
|
|
480
|
|
|
# Replace to obtain UK values from default |
481
|
|
|
rawdict[conc] = rawdict[conc].replace(',','.') |
482
|
|
|
rawdict['Remarks'] = ' '.join([errors, notes]) |
483
|
|
|
rawres = self.getRawResults().get(rid, []) |
484
|
|
|
raw = rawres[0] if len(rawres) > 0 else {} |
485
|
|
|
raw[aname] = rawdict |
486
|
|
|
if not 'DateTime' in raw: |
487
|
|
|
try: |
488
|
|
|
raw['DateTime'] = {'DateTime':self.csvDate2BikaDate(self._header['Date']), |
489
|
|
|
'DefaultValue':'DateTime'} |
490
|
|
|
except: |
491
|
|
|
pass |
492
|
|
|
|
493
|
|
|
self._addRawResult(rid, raw, True) |
494
|
|
|
return 0 |
495
|
|
|
|
496
|
|
|
|
497
|
|
|
def getAttachmentFileType(self): |
498
|
|
|
return "PANalytical - Omnia Axios XRF" |
499
|
|
|
|
500
|
|
|
|
501
|
|
|
class AxiosXrfImporter(AnalysisResultsImporter): |
502
|
|
|
|
503
|
|
|
def __init__(self, parser, context, override, |
504
|
|
|
allowed_ar_states=None, allowed_analysis_states=None, |
505
|
|
|
instrument_uid=None): |
506
|
|
|
AnalysisResultsImporter.__init__(self, parser, context, |
507
|
|
|
override, |
508
|
|
|
allowed_ar_states, |
509
|
|
|
allowed_analysis_states, |
510
|
|
|
instrument_uid) |
511
|
|
|
|