This is an experimental copy for testing Poikilos' issue mirroring system. Note that Gitea's migration tool can import issues, but the "Issues" checkbox is disabled when "This repository will be a mirror" is enabled (it is for this repo).
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

411 lines
21 KiB

import os
import sys
import keyword
#import types
import inspect
import traceback
module_list = list()
module_list.append("os")
module_list.append("sys")
module_list.append("keyword")
module_list.append("inspect")
module_list.append("traceback")
def view_traceback():
ex_type, ex, tb = sys.exc_info()
print(str(ex_type))
print(str(ex))
traceback.print_tb(tb)
del tb
class RegressionMismatch:
side_a_string = None
side_b_string = None
endswith_enable = None
startswith_enable = None
def __init__(self, side_a_string, side_b_string, startswith_enable, endswith_enable):
self.side_a_string = side_a_string
self.side_b_string = side_b_string
self.endswith_enable = endswith_enable
self.startswith_enable = startswith_enable
global_case_sensitive_enable = None
y_enable = False
print("Initializing...")
regression_mismatches = list()
independent_list = ["index","suffix","prefix","decachunk_x_path"]
print(" (Ignoring the following independent variables:")
print(','.join(independent_list)+")")
#NOTE: "for decachunk_z_name in os.listdir(decachunk_x_path):" is ok since z folders are in x folder
independent_endswith_list = list()
for word in independent_list:
independent_endswith_list.append("_"+word)
regression_mismatches.append(RegressionMismatch("_x","_z",False,True))
if y_enable:
regression_mismatches.append(RegressionMismatch("_x","_y",False,True))
regression_mismatches.append(RegressionMismatch("_y","_z",False,True))
regression_mismatches.append(RegressionMismatch("_x_","_z_",False,False))
if y_enable:
regression_mismatches.append(RegressionMismatch("_x_","_y_",False,False))
regression_mismatches.append(RegressionMismatch("_y_","_z_",False,False))
regression_mismatches.append(RegressionMismatch("x","z",False,True))
if y_enable:
regression_mismatches.append(RegressionMismatch("x","y",False,True))
regression_mismatches.append(RegressionMismatch("y","z",False,True))
regression_mismatches.append(RegressionMismatch("x_","z_",True,False))
if y_enable:
regression_mismatches.append(RegressionMismatch("x_","y_",True,False))
regression_mismatches.append(RegressionMismatch("y_","z_",True,False))
if global_case_sensitive_enable is True:
regression_mismatches.append(RegressionMismatch("_X_","_Z_",False,False))
if y_enable:
regression_mismatches.append(RegressionMismatch("_X_","_Y_",False,False))
regression_mismatches.append(RegressionMismatch("_Y_","_Z_",False,False))
regression_mismatches.append(RegressionMismatch("_X","_Z",False,True))
if y_enable:
regression_mismatches.append(RegressionMismatch("_X","_Y",False,True))
regression_mismatches.append(RegressionMismatch("_Y","_Z",False,True))
regression_mismatches.append(RegressionMismatch("X","Z",False,True))
if y_enable:
regression_mismatches.append(RegressionMismatch("X","Y",False,True))
regression_mismatches.append(RegressionMismatch("Y","Z",False,True))
regression_mismatches.append(RegressionMismatch("X_","Z_",True,False))
if y_enable:
regression_mismatches.append(RegressionMismatch("X_","Y_",True,False))
regression_mismatches.append(RegressionMismatch("Y_","Z_",True,False))
#splits by any non-alphanumeric characters
#print(keyword.kwlist)
# DOESN'T WORK (from linuxbochs on http://stackoverflow.com/questions/6315496/display-a-list-of-user-defined-functions-in-the-python-idle-session ):
#function_list = [f for f in globals().values() if type(f) == types.FunctionType]
# DOESN'T WORK
#def dummy(): pass
#function_list = [f.__name__ for f in globals().values() if type(f) == type(dummy)]
function_list = list()
for module_string in module_list:
exec ("function_list += inspect.getmembers("+module_string+", inspect.isroutine)")
# NOTE: isfunction does NOT include c-defined functions such as those in math
function_names = list()
for function_tuple in function_list:
function_names.append(function_tuple[0])
print(" (Ignoring known routines:")
print(','.join(function_names)+")")
print("")
print("")
def split_non_alnum(haystack, strip_enable=True, skip_keywords_enable=True):
results = list()
index = 0
start_index = 0
while index <= len(haystack):
if index==len(haystack) or not haystack[index].isalnum():
word = haystack[start_index:index]
if (not skip_keywords_enable) or (word not in keyword.kwlist and word not in function_names):
if strip_enable:
results.append(word.strip())
else:
results.append(word)
start_index = index+1
index += 1
return results
alpha_upper_chars = "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
alpha_lower_chars = alpha_upper_chars.lower()
alpha_chars = alpha_upper_chars+alpha_lower_chars
numeric_chars = "1234567890"
alnum_chars = alpha_chars+numeric_chars
identifier_chars = alnum_chars+"_"
issue_count = 0
file_list = list()
def is_identifier(needle):
result = True
for index in range(0,len(needle)):
if needle[index] not in identifier_chars:
result = False
break
return result
def split_non_identifier(haystack, strip_enable=True, skip_keywords_enable=True, skip_disoriented_enable=True):
results = list()
index = 0
start_index = 0
while index <= len(haystack):
if index==len(haystack) or not is_identifier(haystack[index]):
word = haystack[start_index:index]
if (not skip_keywords_enable) or word not in keyword.kwlist:
if (not skip_disoriented_enable) or (word not in independent_list and not endswith_any(word, independent_endswith_list)):
if strip_enable:
results.append(word.strip())
else:
results.append(word)
start_index = index+1
index += 1
return results
def endswith(haystack, needle, case_sensitive_enable=False):
if global_case_sensitive_enable is not None:
case_sensitive_enable = global_case_sensitive_enable
result = False
if haystack is not None and needle is not None and len(needle)>0 and len(haystack)>=len(needle):
if case_sensitive_enable:
if haystack[-len(needle):] == needle:
#print("haystack[-len(needle):] = "+haystack[-len(needle):]+" in "+haystack)
result = True
else:
if haystack[-len(needle):].lower() == needle.lower():
#print("haystack[-len(needle):].lower() = "+haystack[-len(needle):].lower()+" in "+haystack)
result = True
return result
def startswith(haystack, needle, case_sensitive_enable=False):
if global_case_sensitive_enable is not None:
case_sensitive_enable = global_case_sensitive_enable
result = False
if haystack is not None and needle is not None and len(needle)>0 and len(haystack)>=len(needle):
if case_sensitive_enable:
if haystack[:len(needle)] == needle:
result = True
else:
if haystack[:len(needle)].lower() == needle.lower():
result = True
return result
def any_endswith(haystacks, needle, case_sensitive_enable=False):
if global_case_sensitive_enable is not None:
case_sensitive_enable = global_case_sensitive_enable
result = False
for haystack in haystacks:
if endswith(haystack, needle, case_sensitive_enable):
result = True
break
return result
def endswith_any(haystack, needles, case_sensitive_enable=False):
if global_case_sensitive_enable is not None:
case_sensitive_enable = global_case_sensitive_enable
result = False
for needle in needles:
if endswith(haystack, needle, case_sensitive_enable):
result = True
break
return result
def any_startswith(haystacks, needle, case_sensitive_enable=False):
if global_case_sensitive_enable is not None:
case_sensitive_enable = global_case_sensitive_enable
result = False
for haystack in haystacks:
if startswith(haystack, needle, case_sensitive_enable):
result = True
break
return result
#only works on strings IF case_sensitive_enable, since then would do lower() on each element of haystacks
def in_any_string(needle, haystacks, case_sensitive_enable=False):
if global_case_sensitive_enable is not None:
case_sensitive_enable = global_case_sensitive_enable
result = False
if needle is not None and len(needle)>0:
for haystack in haystacks:
if haystack is not None and len(haystack)>0:
if case_sensitive_enable:
if needle in haystack:
result = True
break
else:
if needle.lower() in haystack.lower():
result = True
break
return result
min_indent = ""
def increase_indent():
global min_indent
min_indent+=" "
def decrease_indent():
global min_indent
if len(min_indent)>=2:
min_indent = min_indent[:-2]
def check_coord_mismatch(file_path):
global file_list
global issue_count
global module_list
print("Running check_coord_mismatch on "+file_path+"...")
global function_list
if (file_path not in file_list):
file_list.append(file_path)
line_counting_number = 1
ins = open(file_path, 'r')
line = True
global min_indent
while line:
line = ins.readline()
if line:
line_strip = line.strip()
if (len(line_strip)>0) and (line_strip[0]!="#"):
ao_index = line_strip.find("=")
import_string = "import "
from_string = "from "
import_index = -1
if line_strip[:len(import_string)]==import_string:
import_index = 0
elif line_strip[:len(from_string)]==from_string:
import_string = from_string
import_index = line_strip.find(import_string)
if import_index>=0:
module_string = line_strip[import_index+len(import_string):].strip()
space_index = module_string.find(" ")
if space_index>-1:
module_string = module_string[:space_index]
if module_string not in module_list:
module_list.append(module_string)
try:
#tmp_tuples = list()
import_string = "import "+module_string
exec_string = "tmp_tuples = inspect.getmembers("+module_string+", inspect.isroutine)"
#exec exec_string
try_enable = False
outs = open('expertmmregressiontmp.py','w')
outs.write("def get_module_contents():"+"\n")
outs_indent = " "
outs.write(outs_indent+"results = None"+"\n")
if try_enable:
outs.write(outs_indent+"try:"+"\n")
outs_indent = " "
outs.write(outs_indent+"import inspect"+"\n")
outs.write(outs_indent+import_string+"\n")
outs.write(outs_indent+exec_string+"\n")
outs.write(outs_indent+"results = list()"+"\n")
outs.write(outs_indent+"for function_tuple in tmp_tuples:"+"\n")
outs.write(outs_indent+" results.append(function_tuple[0])"+"\n")
outs_indent = " "
if try_enable:
outs.write(outs_indent+"except:"+"\n")
outs_indent = " "
outs.write(outs_indent+"print(\"Could not finish get_module_contents\")"+"\n")
outs_indent = " "
outs.write(outs_indent+"return results"+"\n")
outs.write("\n")
outs.close()
import expertmmregressiontmp
tmp_list = expertmmregressiontmp.get_module_contents()
new_list = None
if tmp_list is not None:
new_list = list()
for routine_string in tmp_list:
if routine_string not in function_list:
new_list.append(routine_string)
if new_list is not None:
function_list += new_list
print("Found "+str(len(new_list))+" new method(s) from '"+module_string+"' to ignore: "+','.join(new_list))
else:
print("unable to import module named '"+module_string+"', so some routines may not be successfully ignored:")
except:
print("Could not finish importing module named '"+module_string+"', so some routines may not be successfully ignored:")
view_traceback()
if ao_index<0:
ao_index = line_strip.find(">")
if ao_index<0:
ao_index = line_strip.find("<")
if ao_index<0:
ao_index = line_strip.find(" in ")
if ao_index>0: # intentionally >0 instead of =
increase_indent()
names_string = line_strip[:ao_index].strip()
values_string = line_strip[ao_index+1:].strip()
name_list = split_non_identifier(names_string)
value_list = split_non_identifier(values_string)
message = None
is_line_problematic = False
for regression_mismatch in regression_mismatches:
both_present = False
message_prefix = " WARNING: "
only_mismatched_coord_is_present_string = " ERROR (only has mismatch): "
if regression_mismatch.startswith_enable:
if (any_startswith(name_list,regression_mismatch.side_a_string) and not any_startswith(name_list,regression_mismatch.side_b_string)) and any_startswith(value_list,regression_mismatch.side_b_string):
name_index = line.find(regression_mismatch.side_a_string) + 1
if not both_present:
both_present = any_startswith(value_list,regression_mismatch.side_a_string)
message_prefix = only_mismatched_coord_is_present_string
if message is None:
message = (file_path+" ("+str(line_counting_number)+","+str(name_index)+")"+message_prefix+"name starts with "+regression_mismatch.side_a_string+", but "+regression_mismatch.side_b_string+" on right in check/assignment")
is_line_problematic = True
break
elif (any_startswith(name_list,regression_mismatch.side_b_string) and not any_startswith(name_list,regression_mismatch.side_a_string)) and any_startswith(value_list,regression_mismatch.side_a_string):
name_index = line.find(regression_mismatch.side_b_string) + 1
if not both_present:
both_present = any_startswith(value_list,regression_mismatch.side_b_string)
message_prefix = only_mismatched_coord_is_present_string
if message is None:
message = (file_path+" ("+str(line_counting_number)+","+str(name_index)+")"+message_prefix+"name starts with "+regression_mismatch.side_b_string+", but "+regression_mismatch.side_a_string+" on right in check/assignment")
is_line_problematic = True
break
elif regression_mismatch.endswith_enable:
if (any_endswith(name_list,regression_mismatch.side_a_string) and not any_endswith(name_list,regression_mismatch.side_b_string)) and any_endswith(value_list,regression_mismatch.side_b_string):
name_index = line.find(regression_mismatch.side_a_string) + 1
if not both_present:
both_present = any_endswith(value_list,regression_mismatch.side_a_string)
message_prefix = only_mismatched_coord_is_present_string
if message is None:
message = (file_path+" ("+str(line_counting_number)+","+str(name_index)+")"+message_prefix+"name ends with "+regression_mismatch.side_a_string+", but "+regression_mismatch.side_b_string+" on right in check/assignment")
is_line_problematic = True
break
elif (any_endswith(name_list,regression_mismatch.side_b_string) and not any_endswith(name_list,regression_mismatch.side_a_string)) and any_endswith(value_list,regression_mismatch.side_a_string):
name_index = line.find(regression_mismatch.side_b_string) + 1
if not both_present:
both_present = any_endswith(value_list,regression_mismatch.side_b_string)
message_prefix = only_mismatched_coord_is_present_string
if message is None:
message = (file_path+" ("+str(line_counting_number)+","+str(name_index)+")"+message_prefix+"name ends with "+regression_mismatch.side_b_string+", but "+regression_mismatch.side_a_string+" on right in check/assignment")
is_line_problematic = True
break
else:
if (in_any_string(regression_mismatch.side_a_string, name_list) and not in_any_string(regression_mismatch.side_b_string, name_list)) and in_any_string(regression_mismatch.side_b_string, value_list):
value_index = line.find(regression_mismatch.side_a_string) + 1
if not both_present:
both_present = in_any_string(regression_mismatch.side_a_string, value_list)
message_prefix = only_mismatched_coord_is_present_string
if message is None:
message = (file_path+" ("+str(line_counting_number)+","+str(value_index)+")"+message_prefix+"name contains "+regression_mismatch.side_a_string+", but "+regression_mismatch.side_b_string+" on right in check/assignment")
is_line_problematic = True
break
elif (in_any_string(regression_mismatch.side_b_string, name_list) and not in_any_string(regression_mismatch.side_a_string, name_list)) and in_any_string(regression_mismatch.side_a_string, value_list):
value_index = line.find(regression_mismatch.side_b_string) + 1
if not both_present:
both_present = in_any_string(regression_mismatch.side_b_string, value_list)
message_prefix = only_mismatched_coord_is_present_string
if message is None:
message = (file_path+" ("+str(line_counting_number)+","+str(value_index)+")"+message_prefix+"name contains "+regression_mismatch.side_b_string+", but "+regression_mismatch.side_a_string+" on right side of check/assignment")
is_line_problematic = True
break
if message is not None:
print("")
print(message)
print(line_strip)
if is_line_problematic:
issue_count += 1
line_counting_number += 1
ins.close()
check_coord_mismatch("chunkymap-regen.py")
check_coord_mismatch(os.path.join("web","chunkymap.php"))
print("Found "+str(issue_count)+" issue(s) in "+str(len(file_list))+" file(s)")
raw_input("Press enter to exit...")