unix - Reason for disparate results with mdfind using Python subprocess -

i'm trying write python wrapper unix mdfind utility. in simplest form, works well; however, cannot figure out 1 instance of odd behavior. things bit odd when running more complex queries (two or more fields). take following example:

import subprocess import itertools  def test1():     cmd = "mdfind 'kmditemfsname=pandoc&&kmditemcontenttype=public.unix-executable'"     shell_res = subprocess.check_output(cmd, shell=true)     find_res = mdfind(content_type='public.unix-executable',                       name='pandoc')     if shell_res == find_res:         print('passed!')  def mdfind(**kwargs):     cmd = ['mdfind']     key, arg in kwargs.iteritems():         if key in mdattributes().keys():             md_name = mdattributes()[key]['id']             query = '='.join([md_name, arg])             cmd.append(query)     if 'only_in' in kwargs:         cmd.append('-onlyin')         cmd.append(kwargs['only_in'])     return subprocess.check_output(cmd)  def mdattributes():     attributes_str = subprocess.check_output(['mdimport', '-a'])     # prepare key names 4 columns     keys = ('id', 'name', 'description', 'aliases')     # create list of dicts, mapping ``keys`` item's columns     data = [dict(itertools.izip(keys,                                 [item.replace("'", "")                                  item in attribute.split('\t\t')]))             attribute in attributes_str.splitlines()]     # coerce list of dicts large dict nested dicts     metadata = {}     md_dict in data:         # clean key         key = md_dict['id'].replace('kmditemfs', '')\                            .replace('kmditem', '')\                            .replace('kmd', '')\                            .replace('com_', '')         metadata[key] = md_dict     return metadata  test1() 

this code pass, both straight shell command , wrapper created command output same result.

now, take example, seems me of same kind, yet doesn't work:

def test2():     cmd = """mdfind 'kmditemkind=pdf&&kmditemfsname="*epistem*"c'"""     shell_res = run_shell(cmd)     find_res = mdfind(kind='pdf',                       name='"*epistem*"c') 

the straight shell command return single pdf on machine has "epistemology" in title, while wrapper made command return 13 pdfs (i have 1,000+ pdfs on machine in total). so, wrapper script filtering thousands pdfs somehow, apparently not whether *epistem* in title.

even more oddly, command return 144 results:

subprocess.check_output(['mdfind',                           """kmditemkind=pdf&&kmditemfsname="*epistemolog*"c"""]) 

so, in short these 3 different subprocess calls give radically different numbers of results:

"""mdfind 'kmditemkind=pdf&&kmditemfsname="*epistem*"c'""" ['mdfind', 'kmditemkind=pdf', u'kmditemfsname="*epistem*"c'] ['mdfind', """kmditemkind=pdf&&kmditemfsname="*epistemolog*"c"""] 

so, question: why? why subprocess.check_output() return 1 result straight shell command (i mean command string , shell=true set), 13 results 3 item list command, , 144 results 2 item list command? going on under covers? how can 3 item list return 1 item straight shell command does?

i sure has subtle important differences in command line argument processing pipeline. pipeline complex , when invoking command within programming language environment quite difficult obtain equivalent behavior typing command in favorite shell.

the bad thing is: depending on method target executable uses parsing command line arguments (there unfortunately -- in many cases -- no definite standard) outcome may vary depending among invocation methods. is, observation has processing , interpretation of whitespace, null characters, dashes, , quotes.

your question "why?". if want bottom of this, need @ source code of python's subprocess module , @ source code of target commands command line argument parsing code. also, might want have these reads:

• http://www.daviddeley.com/autohotkey/parameters/parameters.htm
• http://gehrcke.de/2014/02/command-line-argument-binary-data/

in order obtain conceptually equivalent behavior typing in shell, there non-ovious dead-simple workaround: create temporary shell script , invoke shell within python , provide 1 argument: path shell script. have used method in module creating systematic command line tool tests: https://github.com/jgehrcke/timegaps/blob/master/test/clitest.py