# rgToolFactoryMultIn.py
# see https://bitbucket.org/fubar/galaxytoolfactory/wiki/Home
#
# copyright ross lazarus (ross stop lazarus at gmail stop com) May 2012
#
# all rights reserved
# Licensed under the LGPL
# suggestions for improvement and bug fixes welcome at https://bitbucket.org/fubar/galaxytoolfactory/wiki/Home
#
# sept 2014 added additional params from
# https://bitbucket.org/mvdbeek/dockertoolfactory/src/d4863bcf7b521532c7e8c61b6333840ba5393f73/DockerToolFactory.py?at=default
# passing them is complex
# and they are restricted to NOT contain commas or double quotes to ensure that they can be safely passed together on
# the toolfactory command line as a comma delimited double quoted string for parsing and passing to the script
# see examples on this tool form
# august 2014
# Allows arbitrary number of input files
# NOTE positional parameters are now passed to script
# and output (may be "None") is *before* arbitrary number of inputs
#
# march 2014
# had to remove dependencies because cross toolshed dependencies are not possible - can't pre-specify a toolshed url for graphicsmagick and ghostscript
# grrrrr - night before a demo
# added dependencies to a tool_dependencies.xml if html page generated so generated tool is properly portable
#
# added ghostscript and graphicsmagick as dependencies
# fixed a wierd problem where gs was trying to use the new_files_path from universe (database/tmp) as ./database/tmp
# errors ensued
#
# august 2013
# found a problem with GS if $TMP or $TEMP missing - now inject /tmp and warn
#
# july 2013
# added ability to combine images and individual log files into html output
# just make sure there's a log file foo.log and it will be output
# together with all images named like "foo_*.pdf
# otherwise old format for html
#
# January 2013
# problem pointed out by Carlos Borroto
# added escaping for <>$ - thought I did that ages ago...
#
# August 11 2012
# changed to use shell=False and cl as a sequence
# This is a Galaxy tool factory for simple scripts in python, R or whatever ails ye.
# It also serves as the wrapper for the new tool.
#
# you paste and run your script
# Only works for simple scripts that read one input from the history.
# Optionally can write one new history dataset,
# and optionally collect any number of outputs into links on an autogenerated HTML page.
# DO NOT install on a public or important site - please.
# installed generated tools are fine if the script is safe.
# They just run normally and their user cannot do anything unusually insecure
# but please, practice safe toolshed.
# Read the fucking code before you install any tool
# especially this one
# After you get the script working on some test data, you can
# optionally generate a toolshed compatible gzip file
# containing your script safely wrapped as an ordinary Galaxy script in your local toolshed for
# safe and largely automated installation in a production Galaxy.
# If you opt for an HTML output, you get all the script outputs arranged
# as a single Html history item - all output files are linked, thumbnails for all the pdfs.
# Ugly but really inexpensive.
#
# Patches appreciated please.
#
#
# long route to June 2012 product
# Behold the awesome power of Galaxy and the toolshed with the tool factory to bind them
# derived from an integrated script model
# called rgBaseScriptWrapper.py
# Note to the unwary:
# This tool allows arbitrary scripting on your Galaxy as the Galaxy user
# There is nothing stopping a malicious user doing whatever they choose
# Extremely dangerous!!
# Totally insecure. So, trusted users only
#
# preferred model is a developer using their throw away workstation instance - ie a private site.
# no real risk. The universe_wsgi.ini admin_users string is checked - only admin users are permitted to run this tool.
#
import sys
import shutil
import subprocess
import os
import time
import tempfile
import optparse
import tarfile
import re
import shutil
import math
progname = os.path.split(sys.argv[0])[1]
myversion = 'V001.1 March 2014'
verbose = False
debug = False
toolFactoryURL = 'https://bitbucket.org/fubar/galaxytoolfactory'
# if we do html we need these dependencies specified in a tool_dependencies.xml file and referred to in the generated
# tool xml
toolhtmldepskel = """
%s
"""
toolhtmldepskel = """
%s
"""
protorequirements = """ghostscriptgraphicsmagick"""
def timenow():
"""return current time as a string
"""
return time.strftime('%d/%m/%Y %H:%M:%S', time.localtime(time.time()))
html_escape_table = {
"&": "&",
">": ">",
"<": "<",
"$": "\$"
}
def html_escape(text):
"""Produce entities within text."""
return "".join(html_escape_table.get(c,c) for c in text)
def cmd_exists(cmd):
return subprocess.call("type " + cmd, shell=True,
stdout=subprocess.PIPE, stderr=subprocess.PIPE) == 0
def parse_citations(citations_text):
"""
"""
citations = [c for c in citations_text.split("**ENTRY**") if c.strip()]
citation_tuples = []
for citation in citations:
if citation.startswith("doi"):
citation_tuples.append( ("doi", citation[len("doi"):].strip() ) )
else:
citation_tuples.append( ("bibtex", citation[len("bibtex"):].strip() ) )
return citation_tuples
class ScriptRunner:
"""class is a wrapper for an arbitrary script
"""
def __init__(self,opts=None,treatbashSpecial=True):
"""
cleanup inputs, setup some outputs
"""
self.useGM = cmd_exists('gm')
self.useIM = cmd_exists('convert')
self.useGS = cmd_exists('gs')
self.temp_warned = False # we want only one warning if $TMP not set
self.treatbashSpecial = treatbashSpecial
if opts.output_dir: # simplify for the tool tarball
os.chdir(opts.output_dir)
self.thumbformat = 'png'
self.opts = opts
self.toolname = re.sub('[^a-zA-Z0-9_]+', '', opts.tool_name) # a sanitizer now does this but..
self.toolid = self.toolname
self.myname = sys.argv[0] # get our name because we write ourselves out as a tool later
self.pyfile = self.myname # crude but efficient - the cruft won't hurt much
self.xmlfile = '%s.xml' % self.toolname
s = open(self.opts.script_path,'r').readlines()
s = [x.rstrip() for x in s] # remove pesky dos line endings if needed
self.script = '\n'.join(s)
fhandle,self.sfile = tempfile.mkstemp(prefix=self.toolname,suffix=".%s" % (opts.interpreter))
tscript = open(self.sfile,'w') # use self.sfile as script source for Popen
tscript.write(self.script)
tscript.close()
self.indentedScript = '\n'.join([' %s' % html_escape(x) for x in s]) # for restructured text in help
self.escapedScript = '\n'.join([html_escape(x) for x in s])
self.elog = os.path.join(self.opts.output_dir,"%s_error.log" % self.toolname)
if opts.output_dir: # may not want these complexities
self.tlog = os.path.join(self.opts.output_dir,"%s_runner.log" % self.toolname)
art = '%s.%s' % (self.toolname,opts.interpreter)
artpath = os.path.join(self.opts.output_dir,art) # need full path
artifact = open(artpath,'w') # use self.sfile as script source for Popen
artifact.write(self.script)
artifact.close()
self.cl = []
self.html = []
self.test1Inputs = [] # now a list
a = self.cl.append
a(opts.interpreter)
if self.treatbashSpecial and opts.interpreter in ['bash','sh']:
a(self.sfile)
else:
a('-') # stdin
# if multiple inputs - positional or need to distinguish them with cl params
if opts.input_tab:
tests = []
for i,intab in enumerate(opts.input_tab): # if multiple, make tests
if intab.find(',') <> -1:
(gpath,uname) = intab.split(',')
else:
gpath = uname = intab
tests.append(os.path.basename(gpath))
self.test1Inputs = '' % (','.join(tests))
else:
self.test1Inputs = ''
# we always pass path,name pairs in using python optparse append
# but the command line has to be different
self.infile_paths = ','.join([x.split(',')[0] for x in self.opts.input_tab])
self.infile_names = ','.join([x.split(',')[1] for x in self.opts.input_tab])
if self.opts.interpreter == 'python':
# yes, this is how additional parameters are always passed in python - to the TF itself and to
# scripts to avoid having unknown parameter names (yes, they can be parsed but...) on the command line
a('--INPATHS "%s"' % (self.infile_paths))
a('--INNAMES "%s"' % (self.infile_names))
if self.opts.output_tab:
a('--OUTPATH "%s"' % self.opts.output_tab)
for p in opts.additional_parameters:
p = p.replace('"','')
psplit=p.split(',')
param = psplit[0]
value = psplit[1]
a('--additional_parameters "%s,%s"' % (param,value))
if (self.opts.interpreter == 'Rscript'):
# pass params on command line
a('INPATHS "%s"' % self.infile_paths)
a('INNAMES "%s"' % self.infile_names)
if self.opts.output_tab:
a('OUTPATH "%s"' % self.opts.output_tab)
for param in opts.additional_parameters:
param, value=param.split(',')
a('%s="%s"' % (param,value))
if (self.opts.interpreter == 'perl'):
# pass params on command line
a('%s' % self.infile_paths)
a('%s' % self.infile_names)
if self.opts.output_tab:
a('%s' % self.opts.output_tab)
for param in opts.additional_parameters:
param, value=param.split(',')
if (value.find(' ') <> -1):
a('%s="%s"' % (param,value))
else:
a('%s=%s' % (param,value))
if self.opts.interpreter == 'sh' or self.opts.interpreter == 'bash':
# more is better - now move all params into environment AND drop on to command line.
self.cl.insert(0,'env')
self.cl.insert(1,'INPATHS=%s' % (self.infile_paths))
self.cl.insert(2,'INNAMES=%s' % (self.infile_names))
if self.opts.output_tab:
self.cl.insert(3,'OUTPATH=%s' % (self.opts.output_tab))
a('OUTPATH=%s' % (self.opts.output_tab))
# sets those environment variables for the script
# additional params appear in CL - yes, it's confusing
for i,param in enumerate(opts.additional_parameters):
psplit = param.split(',')
n = psplit[0]
v = psplit[1]
if (v.find(' ') <> -1):
a('%s="%s"' % (n,v))
self.cl.insert(4+i,'%s="%s"' % (n,v))
else:
a('%s=%s' % (n,v))
self.cl.insert(4+i,'%s=%s' % (n,v))
self.outFormats = opts.output_format
self.inputFormats = opts.input_formats
self.test1Output = '%s_test1_output.xls' % self.toolname
self.test1HTML = '%s_test1_output.html' % self.toolname
def makeXML(self):
"""
Create a Galaxy xml tool wrapper for the new script as a string to write out
fixme - use templating or something less fugly than this example of what we produce
a tabular file
reverse.py --script_path "$runMe" --interpreter "python"
--tool_name "reverse" --input_tab "$input1" --output_tab "$tab_file"
**What it Does**
Reverse the columns in a tabular file
# reverse order of columns in a tabular file
import sys
inp = sys.argv[1]
outp = sys.argv[2]
i = open(inp,'r')
o = open(outp,'w')
for row in i:
rs = row.rstrip().split('\t')
rs.reverse()
o.write('\t'.join(rs))
o.write('\n')
i.close()
o.close()
"""
newXML="""
%(tooldesc)s
%(requirements)s
%(command)s
%(inputs)s
%(additionalInputs)s
%(outputs)s
%(script)s
%(tooltests)s
%(help)s
%(citations)s
10.1093/bioinformatics/bts573""" # needs a dict with toolname, toolid, interpreter, scriptname, command, inputs as a multi line string ready to write, outputs ditto, help ditto
newCommand="""
%(toolname)s.py --script_path "$runMe" --interpreter "%(interpreter)s"
--tool_name "%(toolname)s"
%(command_inputs)s
%(command_outputs)s
"""
# may NOT be an input or htmlout - appended later
tooltestsTabOnly = """
%(test1Inputs)s
%(additionalParams)s
"""
tooltestsHTMLOnly = """
%(test1Inputs)s
%(additionalParams)s
"""
tooltestsBoth = """
%(test1Inputs)s
%(additionalParams)s
"""
xdict = {}
xdict['additionalParams'] = ''
xdict['additionalInputs'] = ''
if self.opts.additional_parameters:
if self.opts.edit_additional_parameters: # add to new tool form with default value set to original value
xdict['additionalInputs'] = '\n'.join(['' % (x.split(',')[0],x.split(',')[1],x.split(',')[2],
x.split(',')[3], x.split(',')[4]) for x in self.opts.additional_parameters])
xdict['additionalParams'] = '\n'.join(['' % (x.split(',')[0],x.split(',')[1]) for x in self.opts.additional_parameters])
xdict['requirements'] = ''
if self.opts.make_HTML:
if self.opts.include_dependencies == "yes":
xdict['requirements'] = protorequirements
xdict['tool_version'] = self.opts.tool_version
xdict['test1HTML'] = self.test1HTML
xdict['test1Output'] = self.test1Output
xdict['test1Inputs'] = self.test1Inputs
if self.opts.make_HTML and self.opts.output_tab:
xdict['tooltests'] = tooltestsBoth % xdict
elif self.opts.make_HTML:
xdict['tooltests'] = tooltestsHTMLOnly % xdict
else:
xdict['tooltests'] = tooltestsTabOnly % xdict
xdict['script'] = self.escapedScript
# configfile is least painful way to embed script to avoid external dependencies
# but requires escaping of <, > and $ to avoid Mako parsing
if self.opts.help_text:
helptext = open(self.opts.help_text,'r').readlines()
helptext = [html_escape(x) for x in helptext] # must html escape here too - thanks to Marius van den Beek
xdict['help'] = ''.join([x for x in helptext])
else:
xdict['help'] = 'Please ask the tool author (%s) for help as none was supplied at tool generation\n' % (self.opts.user_email)
coda = ['**Script**','Pressing execute will run the following code over your input file and generate some outputs in your history::']
coda.append('\n')
coda.append(self.indentedScript)
coda.append('\n**Attribution**\nThis Galaxy tool was created by %s at %s\nusing the Galaxy Tool Factory.\n' % (self.opts.user_email,timenow()))
coda.append('See %s for details of that project' % (toolFactoryURL))
coda.append('Please cite: Creating re-usable tools from scripts: The Galaxy Tool Factory. Ross Lazarus; Antony Kaspi; Mark Ziemann; The Galaxy Team. ')
coda.append('Bioinformatics 2012; doi: 10.1093/bioinformatics/bts573\n')
xdict['help'] = '%s\n%s' % (xdict['help'],'\n'.join(coda))
if self.opts.tool_desc:
xdict['tooldesc'] = '%s' % self.opts.tool_desc
else:
xdict['tooldesc'] = ''
xdict['command_outputs'] = ''
xdict['outputs'] = ''
if self.opts.input_tab <> 'None':
cins = ['\n',]
cins.append('#for intab in $input1:')
cins.append('--input_tab "$intab,$intab.name"')
cins.append('#end for\n')
xdict['command_inputs'] = '\n'.join(cins)
xdict['inputs'] = ''' \n''' % (self.inputFormats,self.inputFormats)
else:
xdict['command_inputs'] = '' # assume no input - eg a random data generator
xdict['inputs'] = ''
if (len(self.opts.additional_parameters) > 0):
cins = ['\n',]
for params in self.opts.additional_parameters:
if self.opts.edit_additional_parameters:
psplit = params.split(',') # name,value...
psplit[1] = '$%s' % psplit[0] # replace with form value
cins.append('--additional_parameters "%s"' % ','.join(psplit))
else:
cins.append('--additional_parameters "%s"' % params)
xdict['command_inputs'] = '%s\n%s' % (xdict['command_inputs'],'\n'.join(cins))
xdict['inputs'] += ' \n' % self.toolname
xdict['toolname'] = self.toolname
xdict['toolid'] = self.toolid
xdict['interpreter'] = self.opts.interpreter
xdict['scriptname'] = self.sfile
if self.opts.make_HTML:
xdict['command_outputs'] += ' --output_dir "$html_file.files_path" --output_html "$html_file" --make_HTML "yes"'
xdict['outputs'] += ' \n'
else:
xdict['command_outputs'] += ' --output_dir "./"'
if self.opts.output_tab:
xdict['command_outputs'] += ' --output_tab "$tab_file"'
xdict['outputs'] += ' \n' % self.outFormats
xdict['command'] = newCommand % xdict
if self.opts.citations:
citationstext = open(self.opts.citations,'r').read()
citation_tuples = parse_citations(citationstext)
citations_xml = ""
for citation_type, citation_content in citation_tuples:
citation_xml = """%s""" % (citation_type, html_escape(citation_content))
citations_xml += citation_xml
xdict['citations'] = citations_xml
else:
xdict['citations'] = ""
xmls = newXML % xdict
xf = open(self.xmlfile,'w')
xf.write(xmls)
xf.write('\n')
xf.close()
# ready for the tarball
def makeTooltar(self):
"""
a tool is a gz tarball with eg
/toolname/tool.xml /toolname/tool.py /toolname/test-data/test1_in.foo ...
"""
retval = self.run()
if retval:
print >> sys.stderr,'## Run failed. Cannot build yet. Please fix and retry'
sys.exit(1)
tdir = self.toolname
os.mkdir(tdir)
self.makeXML()
if self.opts.make_HTML:
if self.opts.help_text:
hlp = open(self.opts.help_text,'r').read()
else:
hlp = 'Please ask the tool author for help as none was supplied at tool generation\n'
if self.opts.include_dependencies == "yes":
tooldepcontent = toolhtmldepskel % hlp
else:
tooldepcontent = emptytoolhtmldepskel % hlp
depf = open(os.path.join(tdir,'tool_dependencies.xml'),'w')
depf.write(tooldepcontent)
depf.write('\n')
depf.close()
if self.opts.input_tab <> 'None': # no reproducible test otherwise? TODO: maybe..
testdir = os.path.join(tdir,'test-data')
os.mkdir(testdir) # make tests directory
for i,intab in enumerate(self.opts.input_tab):
si = self.opts.input_tab[i]
if si.find(',') <> -1:
s = si.split(',')[0]
si = s
dest = os.path.join(testdir,os.path.basename(si))
if si <> dest:
shutil.copyfile(si,dest)
if self.opts.output_tab <> None:
shutil.copyfile(self.opts.output_tab,os.path.join(testdir,self.test1Output))
if self.opts.make_HTML:
shutil.copyfile(self.opts.output_html,os.path.join(testdir,self.test1HTML))
if self.opts.output_dir:
shutil.copyfile(self.tlog,os.path.join(testdir,'test1_out.log'))
outpif = '%s.py' % self.toolname # new name
outpiname = os.path.join(tdir,outpif) # path for the tool tarball
pyin = os.path.basename(self.pyfile) # our name - we rewrite ourselves (TM)
notes = ['# %s - a self annotated version of %s generated by running %s\n' % (outpiname,pyin,pyin),]
notes.append('# to make a new Galaxy tool called %s\n' % self.toolname)
notes.append('# User %s at %s\n' % (self.opts.user_email,timenow()))
pi = open(self.pyfile,'r').readlines() # our code becomes new tool wrapper (!) - first Galaxy worm
notes += pi
outpi = open(outpiname,'w')
outpi.write(''.join(notes))
outpi.write('\n')
outpi.close()
stname = os.path.join(tdir,self.sfile)
if not os.path.exists(stname):
shutil.copyfile(self.sfile, stname)
xtname = os.path.join(tdir,self.xmlfile)
if not os.path.exists(xtname):
shutil.copyfile(self.xmlfile,xtname)
tarpath = "%s.tar.gz" % self.toolname
tar = tarfile.open(tarpath, "w:gz")
tar.add(tdir,arcname='%s' % self.toolname)
tar.close()
shutil.copyfile(tarpath,self.opts.new_tool)
shutil.rmtree(tdir)
## TODO: replace with optional direct upload to local toolshed?
return retval
def compressPDF(self,inpdf=None,thumbformat='png'):
"""need absolute path to pdf
note that GS gets confoozled if no $TMP or $TEMP
so we set it
"""
assert os.path.isfile(inpdf), "## Input %s supplied to %s compressPDF not found" % (inpdf,self.myName)
hlog = os.path.join(self.opts.output_dir,"compress_%s.txt" % os.path.basename(inpdf))
sto = open(hlog,'a')
our_env = os.environ.copy()
our_tmp = our_env.get('TMP',None)
if not our_tmp:
our_tmp = our_env.get('TEMP',None)
if not (our_tmp and os.path.exists(our_tmp)):
newtmp = os.path.join(self.opts.output_dir,'tmp')
try:
os.mkdir(newtmp)
except:
sto.write('## WARNING - cannot make %s - it may exist or permissions need fixing\n' % newtmp)
our_env['TEMP'] = newtmp
if not self.temp_warned:
sto.write('## WARNING - no $TMP or $TEMP!!! Please fix - using %s temporarily\n' % newtmp)
self.temp_warned = True
outpdf = '%s_compressed' % inpdf
cl = ["gs", "-sDEVICE=pdfwrite", "-dNOPAUSE", "-dUseCIEColor", "-dBATCH","-dPDFSETTINGS=/printer", "-sOutputFile=%s" % outpdf,inpdf]
x = subprocess.Popen(cl,stdout=sto,stderr=sto,cwd=self.opts.output_dir,env=our_env)
retval1 = x.wait()
sto.close()
if retval1 == 0:
os.unlink(inpdf)
shutil.move(outpdf,inpdf)
os.unlink(hlog)
hlog = os.path.join(self.opts.output_dir,"thumbnail_%s.txt" % os.path.basename(inpdf))
sto = open(hlog,'w')
outpng = '%s.%s' % (os.path.splitext(inpdf)[0],thumbformat)
if self.useGM:
cl2 = ['gm', 'convert', inpdf, outpng]
else: # assume imagemagick
cl2 = ['convert', inpdf, outpng]
x = subprocess.Popen(cl2,stdout=sto,stderr=sto,cwd=self.opts.output_dir,env=our_env)
retval2 = x.wait()
sto.close()
if retval2 == 0:
os.unlink(hlog)
retval = retval1 or retval2
return retval
def getfSize(self,fpath,outpath):
"""
format a nice file size string
"""
size = ''
fp = os.path.join(outpath,fpath)
if os.path.isfile(fp):
size = '0 B'
n = float(os.path.getsize(fp))
if n > 2**20:
size = '%1.1f MB' % (n/2**20)
elif n > 2**10:
size = '%1.1f KB' % (n/2**10)
elif n > 0:
size = '%d B' % (int(n))
return size
def makeHtml(self):
""" Create an HTML file content to list all the artifacts found in the output_dir
"""
galhtmlprefix = """
\n"""
flist = os.listdir(self.opts.output_dir)
flist = [x for x in flist if x <> 'Rplots.pdf']
flist.sort()
html = []
html.append(galhtmlprefix % progname)
html.append('
Galaxy Tool "%s" run at %s
' % (self.toolname,timenow()))
fhtml = []
if len(flist) > 0:
logfiles = [x for x in flist if x.lower().endswith('.log')] # log file names determine sections
logfiles.sort()
logfiles = [x for x in logfiles if os.path.abspath(x) <> os.path.abspath(self.tlog)]
logfiles.append(os.path.abspath(self.tlog)) # make it the last one
pdflist = []
npdf = len([x for x in flist if os.path.splitext(x)[-1].lower() == '.pdf'])
for rownum,fname in enumerate(flist):
dname,e = os.path.splitext(fname)
sfsize = self.getfSize(fname,self.opts.output_dir)
if e.lower() == '.pdf' : # compress and make a thumbnail
thumb = '%s.%s' % (dname,self.thumbformat)
pdff = os.path.join(self.opts.output_dir,fname)
retval = self.compressPDF(inpdf=pdff,thumbformat=self.thumbformat)
if retval == 0:
pdflist.append((fname,thumb))
else:
pdflist.append((fname,fname))
if (rownum+1) % 2 == 0:
fhtml.append('
' % (fname,fname,sfsize))
for logfname in logfiles: # expect at least tlog - if more
if os.path.abspath(logfname) == os.path.abspath(self.tlog): # handled later
sectionname = 'All tool run'
if (len(logfiles) > 1):
sectionname = 'Other'
ourpdfs = pdflist
else:
realname = os.path.basename(logfname)
sectionname = os.path.splitext(realname)[0].split('_')[0] # break in case _ added to log
ourpdfs = [x for x in pdflist if os.path.basename(x[0]).split('_')[0] == sectionname]
pdflist = [x for x in pdflist if os.path.basename(x[0]).split('_')[0] <> sectionname] # remove
nacross = 1
npdf = len(ourpdfs)
if npdf > 0:
nacross = math.sqrt(npdf) ## int(round(math.log(npdf,2)))
if int(nacross)**2 != npdf:
nacross += 1
nacross = int(nacross)
width = min(400,int(1200/nacross))
html.append('
%s images and outputs
' % sectionname)
html.append('(Click on a thumbnail image to download the corresponding original PDF image) ')
ntogo = nacross # counter for table row padding with empty cells
html.append('
\n
')
for i,paths in enumerate(ourpdfs):
fname,thumb = paths
s= """
\n""" % (fname,thumb,fname,width,fname)
if ((i+1) % nacross == 0):
s += '
\n'
ntogo = 0
if i < (npdf - 1): # more to come
s += '
'*ntogo)
html.append('\n')
logt = open(logfname,'r').readlines()
logtext = [x for x in logt if x.strip() > '']
html.append('
%s log output
' % sectionname)
if len(logtext) > 1:
html.append('\n
\n')
html += logtext
html.append('\n
\n')
else:
html.append('%s is empty ' % logfname)
if len(fhtml) > 0:
fhtml.insert(0,'
Output File Name (click to view)
Size
\n')
fhtml.append('
')
html.append('
All output files available for downloading
\n')
html += fhtml # add all non-pdf files to the end of the display
else:
html.append('
### Error - %s returned no files - please confirm that parameters are sane
' % self.opts.interpreter)
html.append(galhtmlpostfix)
htmlf = file(self.opts.output_html,'w')
htmlf.write('\n'.join(html))
htmlf.write('\n')
htmlf.close()
self.html = html
def run(self):
"""
scripts must be small enough not to fill the pipe!
"""
if self.treatbashSpecial and self.opts.interpreter in ['bash','sh']:
retval = self.runBash()
else:
if self.opts.output_dir:
ste = open(self.elog,'w')
sto = open(self.tlog,'w')
sto.write('## Toolfactory generated command line = %s\n' % ' '.join(self.cl))
sto.flush()
p = subprocess.Popen(self.cl,shell=False,stdout=sto,stderr=ste,stdin=subprocess.PIPE,cwd=self.opts.output_dir)
else:
p = subprocess.Popen(self.cl,shell=False,stdin=subprocess.PIPE)
p.stdin.write(self.script)
p.stdin.close()
retval = p.wait()
if self.opts.output_dir:
sto.close()
ste.close()
err = open(self.elog,'r').readlines()
if retval <> 0 and err: # problem
print >> sys.stderr,err
if self.opts.make_HTML:
self.makeHtml()
return retval
def runBash(self):
"""
cannot use - for bash so use self.sfile
"""
if self.opts.output_dir:
s = '## Toolfactory generated command line = %s\n' % ' '.join(self.cl)
sto = open(self.tlog,'w')
sto.write(s)
sto.flush()
p = subprocess.Popen(self.cl,shell=False,stdout=sto,stderr=sto,cwd=self.opts.output_dir)
else:
p = subprocess.Popen(self.cl,shell=False)
retval = p.wait()
if self.opts.output_dir:
sto.close()
if self.opts.make_HTML:
self.makeHtml()
return retval
def main():
u = """
This is a Galaxy wrapper. It expects to be called by a special purpose tool.xml as:
rgBaseScriptWrapper.py --script_path "$scriptPath" --tool_name "foo" --interpreter "Rscript"
"""
op = optparse.OptionParser()
a = op.add_option
a('--script_path',default=None)
a('--tool_name',default=None)
a('--interpreter',default=None)
a('--output_dir',default='./')
a('--output_html',default=None)
a('--input_tab',default=[], action="append") # these are "galaxypath,metadataname" pairs
a("--input_formats",default="tabular")
a('--output_tab',default=None)
a('--output_format',default='tabular')
a('--user_email',default='Unknown')
a('--bad_user',default=None)
a('--make_Tool',default=None)
a('--make_HTML',default=None)
a('--help_text',default=None)
a('--tool_desc',default=None)
a('--new_tool',default=None)
a('--tool_version',default=None)
a('--include_dependencies',default="yes")
a('--citations',default=None)
a('--additional_parameters', dest='additional_parameters', action='append', default=[])
a('--edit_additional_parameters', action="store_true", default=False)
opts, args = op.parse_args()
assert not opts.bad_user,'UNAUTHORISED: %s is NOT authorized to use this tool until Galaxy admin adds %s to admin_users in universe_wsgi.ini' % (opts.bad_user,opts.bad_user)
assert opts.tool_name,'## Tool Factory expects a tool name - eg --tool_name=DESeq'
assert opts.interpreter,'## Tool Factory wrapper expects an interpreter - eg --interpreter=Rscript'
assert os.path.isfile(opts.script_path),'## Tool Factory wrapper expects a script path - eg --script_path=foo.R'
if opts.output_dir:
try:
os.makedirs(opts.output_dir)
except:
pass
opts.input_tab = [x.replace('"','').replace("'",'') for x in opts.input_tab]
for i,x in enumerate(opts.additional_parameters): # remove quotes we need to deal with spaces in CL params
opts.additional_parameters[i] = opts.additional_parameters[i].replace('"','')
r = ScriptRunner(opts)
if opts.make_Tool:
retcode = r.makeTooltar()
else:
retcode = r.run()
os.unlink(r.sfile)
if retcode:
sys.exit(retcode) # indicate failure to job runner
if __name__ == "__main__":
main()