10
0
mirror of https://gitlab.com/scemama/irpf90.git synced 2024-11-07 22:53:45 +01:00
irpf90/src/preprocessed_text.py
Anthony Scemama 3521e4b926 Allowed do\n enddo
Version:1.1.60
2010-11-26 09:28:12 +01:00

850 lines
26 KiB
Python

#!/usr/bin/python
# IRPF90 is a Fortran90 preprocessor written in Python for programming using
# the Implicit Reference to Parameters (IRP) method.
# Copyright (C) 2009 Anthony SCEMAMA
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License along
# with this program; if not, write to the Free Software Foundation, Inc.,
# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
#
# Anthony Scemama
# LCPQ - IRSAMC - CNRS
# Universite Paul Sabatier
# 118, route de Narbonne
# 31062 Toulouse Cedex 4
# scemama@irsamc.ups-tlse.fr
from irpf90_t import *
from regexps import *
import error
from command_line import command_line
from util import *
# Local regular expressions
re_endif = re.compile("end\s+if")
re_elseif = re.compile("else\s+if")
re_enddo = re.compile("end\s+do")
re_endselect = re.compile("end\s+select")
# Local variables
Free_form = 0
Fixed_form = 1
######################################################################
# Dictionary of simple statements
simple_dict = {
"program": Program ,
"subroutine": Subroutine ,
"begin_shell": Begin_shell ,
"end_shell": End_shell ,
"begin_template": Begin_template ,
"end_template": End_template ,
"subst": Subst ,
"end_doc": End_doc ,
"begin_provider": Begin_provider ,
"&begin_provider": Cont_provider ,
"end_provider": End_provider ,
"assert": Assert ,
"touch": Touch ,
"soft_touch": SoftTouch ,
"provide": Provide ,
"free": Free ,
"irp_if": Irp_If ,
"irp_else": Irp_Else ,
"irp_endif": Irp_Endif ,
"irp_read": Irp_read ,
"irp_write": Irp_write ,
"use": Use ,
"do": Do ,
"if": If ,
"case": Case ,
"elseif": Elseif ,
"else": Else ,
"enddo": Enddo ,
"endif": Endif ,
"endselect": End_select ,
"end": End ,
"include": Include ,
"call": Call ,
"continue": Continue ,
"return": Return ,
"implicit": Implicit ,
"save": Declaration ,
"function": Function ,
"recursive": Function ,
}
simple_dict_keys = simple_dict.keys()
def get_type (i, filename, line, is_doc):
'''Find the type of a text line'''
assert type(i) == int
assert type(filename) == str
assert type(line) == str
assert type(is_doc) == bool
line = line.rstrip()
lower_line0 = line.lstrip().lower()
lower_line = lower_line0.replace("!"," ! ")
# Replacements
lower_line = re_elseif.sub("elseif",lower_line)
lower_line = re_enddo.sub("enddo",lower_line)
lower_line = re_endif.sub("endif",lower_line)
lower_line = re_endselect.sub("endselect",lower_line)
for c in """()'"[]""":
lower_line = lower_line.replace(c," ")
buffer = lower_line.split()
if len(buffer) == 0:
return [ Empty_line(i,line,filename) ], is_doc
firstword = buffer[0]
if firstword.isdigit():
assert len(buffer) > 1
buffer = buffer[1:]
firstword = buffer[0]
# Identify line
if firstword == "end_doc":
return [ End_doc (i,line,filename) ], False
if firstword == "begin_doc":
return [ Begin_doc (i,line,filename) ], True
if is_doc:
return [ Doc (i,line,filename) ], is_doc
if firstword in simple_dict_keys:
return [ simple_dict[firstword](i,line,filename) ], is_doc
if firstword in [ "select", "selectcase" ]:
return [ Select (i,"",filename) ,
Simple_line (i,line,filename) ] , is_doc
if len(lower_line0) > 4:
if firstword[0] == '#':
result = [ Simple_line(i,line,filename) ]
error.warn ( result[0] ,
"""irpf90 may not work with preprocessor directives. You can use
Irp_if ... Irp_else ... Irp_endif
instead of
#ifdef ... #else ... #endif""" )
return result, is_doc
if firstword.startswith("case("):
return [ Case(i,line,filename) ], is_doc
if lower_line0[1:5] == "$omp":
return [ Openmp(i,line,filename) ], is_doc
if re_decl.match(lower_line) is not None:
if "function" in buffer[1:3]:
return [ Function (i,line,filename) ], is_doc
else:
return [ Declaration (i,line,filename) ], is_doc
# Detect errors
if firstword == "dowhile":
error.fail( Do(i,line,filename) , "'do while' should be in 2 words." )
return [ Simple_line(i,line,filename) ], is_doc
######################################################################
def get_text(lines,filename):
'''Read the input file and transform it to labeled lines'''
assert type(filename) == str
assert type(lines) == list
result = []
is_doc = False
for i,line in enumerate(lines):
line, is_doc = get_type(i+1,filename,line,is_doc)
result += line
return result
######################################################################
def execute_shell(text):
'''Execute the embedded shell scripts'''
def fail(l,a,b): error.fail(l,"In Begin_Shell, %s '%s'"%(a,b))
inside = False
result = []
for line in text:
if inside:
if type(line) == Begin_shell:
error.fail(line,"Nested Begin_shell")
elif type(line) == End_shell:
inside = False
# Write script file
scriptname = "%s%s_shell_%d"%(irpdir,line.filename,line.i)
file = open(scriptname,'w')
file.writelines(script)
file.close()
scriptname = "%s_shell_%d"%(line.filename,line.i)
file = open(scriptname,'w')
file.writelines(script)
file.close()
# Execute shell
import os
pipe = os.popen("%s < %s"%(shell,scriptname),'r')
lines = pipe.readlines()
pipe.close()
result += get_text(lines,scriptname)
os.remove(scriptname)
else:
script.append(line.text+'\n')
else:
if type(line) == Begin_shell:
inside = True
begin = line.i
script = []
# Find shell executable
buffer = line.text.split('[')
if len(buffer) > 2:
fail(line,"Too many",'[')
elif len(buffer) < 2:
fail(line,"Missing",'[')
buffer = buffer[1]
buffer = buffer.split(']')
if len(buffer) > 2:
fail(line,"Too many",']')
elif len(buffer) < 2:
fail(line,"Missing",']')
shell = buffer[0].strip()
elif type(line) == End_shell:
error.fail(line,"Begin_shell missing")
else:
result.append(line)
return result
######################################################################
def execute_templates(text):
'''Execute the templates'''
def fail(l,a,b): error.fail(l,"In %s, %s"%(a,b))
def get_variables(line):
buffer = line.text.split('[',1)
if len(buffer)<2:
fail(line,"Subst","Syntax error")
buffer = buffer[1].replace(']','')
buffer = buffer.split(',')
return map(lambda x: '$%s'%(x.strip()), buffer)
TEMPLATE = 1
SUBST = 2
inside = 0
result = []
for line in text:
if inside == 0:
if type(line) == Begin_template:
script = []
inside = TEMPLATE
script = "template = \"\"\"\n"
else:
result.append(line)
elif inside == TEMPLATE:
if type(line) == Begin_template:
fail(line,"template", "Nested Begin_Template")
elif type(line) == End_template:
fail(line,"template","Missing Subst")
elif type(line) == Subst:
inside = SUBST
script += "\"\"\"\n"
variables = get_variables(line)
script += "v = []\n"
subst = ""
else:
script += line.text+"\n"
else: # inside == SUBST
if type(line) == Begin_template:
fail(line,"subst","Nested Begin_template")
elif type(line) == Subst:
fail(line,"subst","Subst already defined")
elif type(line) == End_template:
inside = 0
subst = subst.rstrip()
if subst[-2:] == ';;':
subst = subst[:-2]
for s in subst.split(';;'):
buffer = map(lambda x: x.strip(), s.split(';'))
if len(buffer) != len(variables):
fail(line,"subst","%d variables defined, and %d substitutions"%(len(variables),len(buffer)))
script += "v.append( { \\\n"
for t,v in zip(variables,buffer):
script += ' "%s": """%s""" ,\n'%(t,v)
script += "} )\n"
script += "for d in v:\n t0 = str(template)\n"
for v in variables:
script += " t0 = t0.replace('%s',d['%s'])\n"%(v,v)
script += " print t0\n"
# Write script file
scriptname = "%s%s_template_%d"%(irpdir,line.filename,line.i)
file = open(scriptname,'w')
file.writelines(script)
file.close()
scriptname = "%s_template_%d"%(line.filename,line.i)
file = open(scriptname,'w')
file.writelines(script)
file.close()
# Execute shell
import os
pipe = os.popen("python < %s"%(scriptname),'r')
lines = pipe.readlines()
pipe.close()
result += get_text(lines,scriptname)
os.remove(scriptname)
else:
subst += line.text+'\n'
return result
######################################################################
def form(text):
'''Find if the text is in fixed form or in free form'''
assert type(text) == list
if len(text) == 0:
return Free_form
assert isinstance(text[0],Line)
re2 = re.compile(r"^\s*[!#]")
re3 = re.compile(r"^\s*[^ 0-9]+")
for line in text:
if type(line) in [ Empty_line, Doc, Openmp ]:
pass
else:
if len(line.text) > 5:
test = line.text[0:5]
if test[0] in "Cc#!*":
pass
else:
if re2.match(test) is None and \
re3.match(test) is not None:
return Free_form
if line.text.rstrip()[-1] == '&':
return Free_form
return Fixed_form
######################################################################
def add_operators(text):
re_incr = re.compile(r"(\s*)(.*)(\+=)(.*$)",re.S)
re_decr = re.compile(r"(\s*)(.*)(-=)(.*$)",re.S)
re_mult = re.compile(r"(\s*)(.*)(\*=)(.*$)",re.S)
'''Change additional operators'''
result = []
for line in text:
buffer = line.text
if "+=" in buffer:
line.text = re.sub(re_incr,r'\1\2=\2+(\4)', buffer)
elif "-=" in buffer:
line.text = re.sub(re_decr,r'\1\2=\2-(\4)', buffer)
elif "*=" in buffer:
line.text = re.sub(re_mult,r'\1\2=\2*(\4)', buffer)
result.append(line)
return result
######################################################################
def remove_comments(text,form):
'''Remove all comments'''
result = []
def remove_after_bang(line):
match = re_comment.match(line)
if match is None:
return line
else:
return re_comment.split(line)[1].rstrip()
if form == Free_form:
for line in text:
if type(line) in [ Openmp, Doc] :
result.append(line)
elif type(line) == Empty_line:
pass
else:
newline = line.text.lstrip()
if newline == "" or newline[0] == "!":
pass
else:
line.text = remove_after_bang(line.text)
result.append(line)
return result
else:
for line in text:
if type(line) in [ Openmp, Doc ]:
result.append(line)
elif type(line) == Empty_line:
pass
else:
newline = line.text.lstrip()
if newline == "" or newline[0] == "!":
pass
else:
line.text = remove_after_bang(line.text)
if line.text[0] in "#123456789 ":
result.append(line)
return result
######################################################################
def remove_continuation(text,form):
'''Removes continuation lines'''
result = []
buffer = ""
number = 0
t = None
if form == Free_form:
for line in text:
if line.text[-1] == '&':
buffer = "%s%s\n"%(buffer,line.text)
if number == 0:
t = type(line)
number = line.i
else:
if number != 0:
newline = t(number, \
"%s%s"%(buffer,line.text), \
line.filename)
line = newline
number = 0
buffer = ""
result.append(line)
else:
rev_text = list(text)
rev_text.reverse()
for line in rev_text:
is_continuation = False
if type(line) == Simple_line:
if len(line.text) >= 6:
if line.text[5] != ' ':
is_continuation = True
if is_continuation:
buffer = "&\n%s %s %s"%(line.text[:5],line.text[6:],buffer)
else:
line.text = line.text+buffer
result.insert(0,line)
buffer = ""
return result
######################################################################
def irp_simple_statements(text):
'''Processes simple statements'''
def process_irp_rw(line,rw,t):
assert type(line) == t
buffer = line.text.split()
if len(buffer) == 2:
dummy, variable = buffer
num = "0"
elif len(buffer) == 3:
dummy, variable, num = buffer
else:
error.fail(line,"Error in IRP_%s statement"%(rw,))
variable = variable.lower()
i = line.i
f = line.filename
txt = line.text.lstrip()
result = [
Empty_line(i,"!",f),
t(i,"! >>> %s"%(txt,),variable ),
Provide_all(i," call %ser_%s('%s')"%(rw,variable,num),f),
Empty_line(i,"! >>> END %s "%(txt,),f ),
Empty_line(line.i,"!",f),
]
return result
def process_irp_read (line):
assert type(line) == Irp_read
return process_irp_rw(line,'read' ,Irp_read )
def process_irp_write(line):
assert type(line) == Irp_write
return process_irp_rw(line,'writ' ,Irp_write)
def process_return(line):
assert type(line) == Return
if command_line.do_assert or command_line.do_debug:
newline = Simple_line(line.i," call irp_leave(irp_here)",line.filename)
result = [newline, line]
else:
result = [ line ]
return result
def debug_conditions(line):
'''Find condition in assert statement for debug'''
assert type(line) == Assert
match = re_test.search(line.text)
result = []
if match is not None:
matches = [ match.group(1).strip(), match.group(3).strip() ]
for m in matches:
if not(m.isdigit() or ("'" in m) or (m == "")):
result.append ( Simple_line (line.i, " print *, '%s = ', %s"%(m,m), line.filename) )
result.append ( Simple_line (line.i, " print *, ''", line.filename) )
return result
def process_assert(line):
assert type(line) == Assert
if command_line.do_assert:
condition = "(%s"%(line.text.split('(',1)[1])
if condition == "":
error.fail(line,"Error in Assert statement")
condition_str = condition.replace("'","''")
i = line.i
f = line.filename
txt = line.text.strip()
result = [
Empty_line(i, "!", f),
Empty_line(i, "! >>> %s"%(txt,), f),
If (i, " if (.not.%s) then"%(condition,), f),
Simple_line(i, " call irp_trace", f),
Simple_line(i, " print *, irp_here//': Assert failed:'", f),
Simple_line(i, " print *, ' file: %s, line: %d'"%(f,i), f),
Simple_line(i, " print *, '%s'"%(condition_str,), f),
] + debug_conditions(line) + [
Simple_line(i, " stop 1", f),
Endif (i, " endif", f),
Empty_line(i, "! <<< END %s"%(txt,), f),
Empty_line(i, "!", f)
]
else:
result = []
return result
def process_end(line):
'''Set irp_here variable in provider block'''
line.text = "end"
if command_line.do_assert or command_line.do_debug:
i = line.i
f = line.filename
result = [
Simple_line(i," call irp_leave(irp_here)", f),
line
]
else:
result = [ line ]
return result
def process_begin_provider(line):
assert type(line) == Begin_provider
buffer = line.lower.replace('['," ")
buffer = buffer.replace(']',"")
buffer = buffer.split(',')
if len(buffer) < 2:
error.fail(line,"Error in Begin_provider statement")
varname = buffer[1].strip()
length = len(varname)
i = line.i
f = line.filename
result = [ Begin_provider(i,line.text, (f,varname)),
Declaration(i," character*(%d) :: irp_here = '%s'"%(length,varname), f) ]
if command_line.do_assert or command_line.do_debug:
result += [
Simple_line(i," call irp_enter(irp_here)", f),
]
return result
def process_cont_provider(line):
assert type(line) == Cont_provider
buffer = line.lower.replace('['," ")
buffer = buffer.replace(']',"")
buffer = buffer.split(',')
if len(buffer) < 2:
error.fail(line,"Error in Cont_provider statement")
varname = buffer[1].strip()
i = line.i
f = line.filename
return [ Cont_provider(i,line.text,(f,varname)) ]
def process_subroutine(line):
assert type(line) == Subroutine
subname = find_subname(line)
length = len(subname)
i = line.i
f = line.filename
result = [ line,
Declaration(i," character*(%d) :: irp_here = '%s'"%(length,subname), f) ]
if command_line.do_assert or command_line.do_debug:
result += [
Simple_line(i," call irp_enter(irp_here)", f),
]
return result
def process_function(line):
assert type(line) == Function
buffer = line.text.split('(')
subname = find_subname(line)
length = len(subname)
i = line.i
f = line.filename
result = [ line,
Declaration(i," character*(%d) :: irp_here = '%s'"%(length,subname), f) ]
if command_line.do_assert or command_line.do_debug:
result += [
Simple_line(i," call irp_enter(irp_here)", f),
]
return result
def process_program(line):
assert type(line) == Program
program_name = line.lower.split()[1]
result = [ Program(0,"",program_name) ] + \
process_subroutine( Subroutine(line.i,"subroutine %s"%(program_name,),line.filename) )
return result
d = { Irp_read : process_irp_read,
Irp_write : process_irp_write,
Return : process_return,
Assert : process_assert,
End : process_end,
Begin_provider : process_begin_provider,
Cont_provider : process_cont_provider,
End_provider : process_end,
Subroutine : process_subroutine,
Function : process_function,
Program : process_program,
}
result = []
for line in text:
buffer = [ line ]
for t in d.keys():
if type(line) == t:
buffer = d[t](line)
break
result += buffer
return result
######################################################################
def change_includes(text):
'''Deals with include files'''
result = []
for line in text:
if type(line) == Include:
txt = line.text.replace('"',"'").split("'")
if len(txt) != 3:
print txt
error.fail(line,"Error in include statement")
filename = txt[1].strip()
try:
file = open(filename,'r')
file.close()
result.append(Include(line.i,"! include '%s'"%filename,filename))
result += create_preprocessed_text(filename)
except IOError:
result.append(Declaration(line.i,line.text,line.filename))
else:
result.append(line)
return result
######################################################################
def process_old_style_do(text):
'''Changes old-style do loops to new style'''
assert type(text) == list
def change_matching_enddo(begin,number):
for i in range(begin+1,len(text)):
line = text[i]
if type(line) in [Continue,Enddo]:
buffer = line.text.split()
if buffer[0] == number:
text[i] = Enddo(line.i," enddo",line.filename)
return
error.fail(text[begin],"Old-style do loops should end with 'continue' or 'end do'")
result = []
for i in range(len(text)):
line = text[i]
if type(line) == Do:
buffer = line.text.split()
try:
if buffer[1].isdigit():
number = buffer.pop(1)
change_matching_enddo(i,number)
line.text = " ".join(buffer)
except IndexError:
pass
result.append(line)
return result
######################################################################
def change_single_line_ifs(text):
'''Changes:
if (test) result
to
if (test) then
result
endif'''
assert type(text) == list
result = []
for line in text:
if type(line) == If:
if line.lower.endswith("then"):
result.append(line)
else:
buffer = line.text
begin = buffer.find('(')
if begin < 0:
error.fail(line,"Error in if statement")
level = 0
instring = False
for i,c in enumerate(buffer[begin:]):
if c == "'":
instring = not instring
if instring:
pass
elif c == '(':
level +=1
elif c == ')':
level -= 1
if level == 0:
end = begin+i+1
break
if level != 0:
error.fail(line,"Error in if statement")
test = buffer[:end]
code = buffer[end:]
i = line.i
f = line.filename
result.append( If(i,"%s then"%(test,),f) )
result += get_type(i,f,code,False)[0]
result.append( Endif(i," endif",f) )
else:
result.append(line)
return result
######################################################################
def check_begin_end(text):
'''Checks x...endx consistence'''
filter_line = lambda line: type(line) in [ Do, Enddo, If, Endif, \
Begin_provider, End_provider, \
Subroutine, Function, End, Begin_doc, End_doc ]
text = filter(filter_line, text)
d = { 'do' : Do, 'enddo': Enddo,
'if' : If, 'endif': Endif,
'_doc': Begin_doc, 'end_doc': End_doc}
assert type(text) == list
def find_matching_end_ifdo(begin,x):
level = 1
for i in range(begin+1,len(text)):
line = text[i]
if type(line) == d[x]:
level += 1
elif type(line) == d["end%s"%(x,)]:
level -= 1
if level == 0:
return True
elif type(line) in [End, End_provider]:
break
error.fail(text[begin],"Missing 'end%s'"%(x,))
def find_matching_end_subfunpro(begin,x):
for i in range(begin+1,len(text)):
line = text[i]
if type(line) == x:
return
if type(line) in [ Subroutine, Function, Begin_provider ]:
error.fail(text[begin],"Subroutine/Function/Provider is not closed")
error.fail(text[begin],"Subroutine/Function/Provider is not closed")
level = 0
for i,line in enumerate(text):
if type(line) == Begin_doc:
find_matching_end_ifdo(i,'_doc')
for i,line in enumerate(text):
if type(line) == Do:
find_matching_end_ifdo(i,'do')
elif type(line) == If:
find_matching_end_ifdo(i,'if')
elif type(line) == Subroutine:
level += 1
find_matching_end_subfunpro(i,End)
elif type(line) == Function:
level += 1
find_matching_end_subfunpro(i,End)
elif type(line) == Begin_provider:
level += 1
find_matching_end_subfunpro(i,End_provider)
elif type(line) == End:
level -= 1
elif type(line) == End_provider:
level -= 1
if level < 0:
error.fail(line,"Beginning of block not matched")
return True
######################################################################
def remove_ifdefs(text):
assert type(text) == list
result = []
do_print = True
for line in text:
if type(line) == Irp_If:
var = line.text.split()[1]
do_print = var in command_line.defined
elif type(line) == Irp_Else:
do_print = not do_print
elif type(line) == Irp_Endif:
do_print = True
else:
if do_print:
result.append(line)
return result
######################################################################
def create_preprocessed_text(filename):
file = open(filename,"r")
lines = file.readlines()
file.close()
result = get_text(lines,filename)
result = execute_templates(result)
result = execute_shell(result)
fortran_form = form(result)
result = remove_ifdefs(result)
result = remove_comments(result,fortran_form)
result = remove_continuation(result,fortran_form)
result = add_operators(result)
result = change_includes(result)
result = change_single_line_ifs(result)
result = process_old_style_do(result)
result = irp_simple_statements(result)
check_begin_end(result)
return result
######################################################################
preprocessed_text = parallel_loop( lambda x,y: create_preprocessed_text(x), \
map(lambda x: (x,None), irpf90_files ) )
######################################################################
def debug():
for filename, txt in preprocessed_text:
print "=== "+filename+" ==="
for line in txt:
print line
print irpf90_files
if __name__ == '__main__':
debug()