Welcome to mirror list, hosted at ThFree Co, Russian Federation.

cygwin.com/git/newlib-cygwin.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorJon Turney <jon.turney@dronecode.org.uk>2019-12-16 22:28:59 +0300
committerJon Turney <jon.turney@dronecode.org.uk>2019-12-18 16:35:35 +0300
commit91f1eab9a987011d84681f0842c4cad7478cd334 (patch)
tree16a6d06ebb94d511c69f56e57fd7130ff2c5a4f3 /newlib/doc
parent7fc78f07edc119c91b57e5ad225a75f9ac8081cd (diff)
doc: Untabify python scripts used for making man pages
These scripts fail with a TabError exception if 'python' is python3, as mixing tabs and spaces in indentation is forbidden in python3.
Diffstat (limited to 'newlib/doc')
-rwxr-xr-xnewlib/doc/chapter-texi2docbook.py38
-rwxr-xr-xnewlib/doc/makedocbook.py400
2 files changed, 219 insertions, 219 deletions
diff --git a/newlib/doc/chapter-texi2docbook.py b/newlib/doc/chapter-texi2docbook.py
index eb606dc68..e21489b9a 100755
--- a/newlib/doc/chapter-texi2docbook.py
+++ b/newlib/doc/chapter-texi2docbook.py
@@ -19,25 +19,25 @@ def main():
print ('<!DOCTYPE chapter PUBLIC "-//OASIS//DTD DocBook V4.5//EN" "http://www.oasis-open.org/docbook/xml/4.5/docbookx.dtd">')
for l in sys.stdin.readlines():
- l = l.rstrip()
-
- # transform @file{foo} to <filename>foo</filename>
- l = re.sub("@file{(.*?)}", "<filename>\\1</filename>", l)
-
- if l.startswith("@node"):
- l = l.replace("@node", "", 1)
- l = l.strip()
- l = l.lower()
- if first_node:
- print ('<chapter id="%s" xmlns:xi="http://www.w3.org/2001/XInclude">' % l.replace(' ', '_'))
- first_node = False
- elif l.startswith("@chapter "):
- l = l.replace("@chapter ", "", 1)
- print ('<title>%s</title>' % l)
- elif l.startswith("@include "):
- l = l.replace("@include ", "", 1)
- l = l.replace(".def", ".xml", 1)
- print ('<xi:include href="%s"/>' % l.strip())
+ l = l.rstrip()
+
+ # transform @file{foo} to <filename>foo</filename>
+ l = re.sub("@file{(.*?)}", "<filename>\\1</filename>", l)
+
+ if l.startswith("@node"):
+ l = l.replace("@node", "", 1)
+ l = l.strip()
+ l = l.lower()
+ if first_node:
+ print ('<chapter id="%s" xmlns:xi="http://www.w3.org/2001/XInclude">' % l.replace(' ', '_'))
+ first_node = False
+ elif l.startswith("@chapter "):
+ l = l.replace("@chapter ", "", 1)
+ print ('<title>%s</title>' % l)
+ elif l.startswith("@include "):
+ l = l.replace("@include ", "", 1)
+ l = l.replace(".def", ".xml", 1)
+ print ('<xi:include href="%s"/>' % l.strip())
print ('</chapter>')
diff --git a/newlib/doc/makedocbook.py b/newlib/doc/makedocbook.py
index 97255345d..92d0f279d 100755
--- a/newlib/doc/makedocbook.py
+++ b/newlib/doc/makedocbook.py
@@ -35,11 +35,11 @@ verbose = 0
def dump(s, stage, threshold = 1):
if verbose > threshold:
- print('*' * 40, file=sys.stderr)
- print(stage, file=sys.stderr)
- print('*' * 40, file=sys.stderr)
- print('%s' % s, file=sys.stderr)
- print('*' * 40, file=sys.stderr)
+ print('*' * 40, file=sys.stderr)
+ print(stage, file=sys.stderr)
+ print('*' * 40, file=sys.stderr)
+ print('%s' % s, file=sys.stderr)
+ print('*' * 40, file=sys.stderr)
#
# Stage 1
@@ -48,7 +48,7 @@ def dump(s, stage, threshold = 1):
def skip_whitespace_and_stars(i, src):
while i < len(src) and (src[i].isspace() or (src[i] == '*' and src[i+1] != '/')):
- i += 1
+ i += 1
return i
@@ -60,37 +60,37 @@ def comment_contents_generator(src):
i = 0
while i < len(src) - 2:
- if src[i] == '\n' and src[i+1] == '/' and src[i+2] == '*':
- i = i + 3
-
- i = skip_whitespace_and_stars(i, src)
-
- if src[i] == '.':
- i += 1
-
- while i < len(src):
- if src[i] == '\n':
- yield '\n'
- i += 1
-
- # allow a single blank line
- if i < len(src) and src[i] == '\n':
- yield '\n'
- i += 1
-
- i = skip_whitespace_and_stars(i, src)
-
- elif src[i] == '*' and src[i+1] == '/':
- i = i + 2
- # If we have just output \n\n, this adds another blank line.
- # This is the only way a double blank line can occur.
- yield '\nEND\n'
- break
- else:
- yield src[i]
- i += 1
- else:
- i += 1
+ if src[i] == '\n' and src[i+1] == '/' and src[i+2] == '*':
+ i = i + 3
+
+ i = skip_whitespace_and_stars(i, src)
+
+ if src[i] == '.':
+ i += 1
+
+ while i < len(src):
+ if src[i] == '\n':
+ yield '\n'
+ i += 1
+
+ # allow a single blank line
+ if i < len(src) and src[i] == '\n':
+ yield '\n'
+ i += 1
+
+ i = skip_whitespace_and_stars(i, src)
+
+ elif src[i] == '*' and src[i+1] == '/':
+ i = i + 2
+ # If we have just output \n\n, this adds another blank line.
+ # This is the only way a double blank line can occur.
+ yield '\nEND\n'
+ break
+ else:
+ yield src[i]
+ i += 1
+ else:
+ i += 1
def remove_noncomments(src):
src = '\n' + src
@@ -107,7 +107,7 @@ def remove_noncomments(src):
def iscommand(l):
if re.match('^[A-Z_]{3,}\s*$', l):
- return True
+ return True
return False
def command_block_generator(content):
@@ -115,12 +115,12 @@ def command_block_generator(content):
text = ''
for l in content.splitlines():
- if iscommand(l):
- yield (command, text)
- command = l.rstrip()
- text = ''
- else:
- text = text + l + '\n'
+ if iscommand(l):
+ yield (command, text)
+ command = l.rstrip()
+ text = ''
+ else:
+ text = text + l + '\n'
yield (command, text)
# Look for commands, which give instructions how to process the following input
@@ -142,17 +142,17 @@ def process(content):
# invoke each command on it's text
def perform(processed):
for i in processed:
- c = i[0].rstrip()
- t = i[1].strip() + '\n'
+ c = i[0].rstrip()
+ t = i[1].strip() + '\n'
- if verbose:
- print("performing command '%s'" % c, file=sys.stderr)
+ if verbose:
+ print("performing command '%s'" % c, file=sys.stderr)
- if c in command_dispatch_dict:
- command_dispatch_dict[c](c, t)
- else:
- print("command '%s' is not recognized" % c, file=sys.stderr)
- # the text following an unrecognized command is discarded
+ if c in command_dispatch_dict:
+ command_dispatch_dict[c](c, t)
+ else:
+ print("command '%s' is not recognized" % c, file=sys.stderr)
+ # the text following an unrecognized command is discarded
# FUNCTION (aka TYPEDEF)
#
@@ -162,33 +162,33 @@ def function(c, l):
l = l.strip()
if verbose:
- print('FUNCTION %s' % l, file=sys.stderr)
+ print('FUNCTION %s' % l, file=sys.stderr)
separator = '---'
if ';' in l:
- # fpclassify has an unusual format we also need to handle
- spliton = ';'
- l = l.splitlines()[0]
+ # fpclassify has an unusual format we also need to handle
+ spliton = ';'
+ l = l.splitlines()[0]
elif len(l.splitlines()) > 1:
- # a few pages like mktemp have two '---' lines
- spliton = ';'
- o = ''
- for i in l.splitlines():
- if separator in i:
- o += i + ';'
- else:
- o += i
- l = o[:-1]
+ # a few pages like mktemp have two '---' lines
+ spliton = ';'
+ o = ''
+ for i in l.splitlines():
+ if separator in i:
+ o += i + ';'
+ else:
+ o += i
+ l = o[:-1]
else:
- spliton = '\n'
+ spliton = '\n'
namelist = []
descrlist = []
for a in l.split(spliton):
- (n, d) = a.split(separator, 1)
- namelist = namelist + n.split(',')
- descrlist = descrlist + [d]
+ (n, d) = a.split(separator, 1)
+ namelist = namelist + n.split(',')
+ descrlist = descrlist + [d]
# only copysign and log1p use <[ ]> markup in descr,
# only gets() uses << >> markup
@@ -201,17 +201,17 @@ def function(c, l):
namelist = map(lambda v: v.strip().lstrip('<').rstrip('>'), namelist)
if verbose:
- print(namelist, file=sys.stderr)
+ print(namelist, file=sys.stderr)
# additional alternate names may also appear in INDEX commands
# create the root element if needed
if rootelement is None:
- rootelement = lxml.etree.Element('refentrycontainer')
+ rootelement = lxml.etree.Element('refentrycontainer')
# FUNCTION implies starting a new refentry
if refentry is not None:
- print("multiple FUNCTIONs without NEWPAGE", file=sys.stderr)
- exit(1)
+ print("multiple FUNCTIONs without NEWPAGE", file=sys.stderr)
+ exit(1)
# create the refentry
refentry = lxml.etree.SubElement(rootelement, 'refentry')
@@ -232,8 +232,8 @@ def function(c, l):
refdescriptor.text = namelist[0]
# refname elements exist for all alternate names
for n in namelist:
- refname = lxml.etree.SubElement(refnamediv, 'refname')
- refname.text = n
+ refname = lxml.etree.SubElement(refnamediv, 'refname')
+ refname.text = n
refpurpose = lxml.etree.SubElement(refnamediv, 'refpurpose')
refnamediv.replace(refpurpose, lxml.etree.fromstring('<refpurpose>' + descr + '</refpurpose>'))
@@ -249,7 +249,7 @@ def index(c, l):
l = l.strip()
if verbose:
- print('INDEX %s' % l, file=sys.stderr)
+ print('INDEX %s' % l, file=sys.stderr)
# discard anything after the first word
l = l.split()[0]
@@ -269,13 +269,13 @@ def index(c, l):
# as long as it doesn't already exist
if not refnamediv.xpath(('refname[.="%s"]') % l):
- refname = lxml.etree.SubElement(refnamediv, 'refname')
- refname.text = l
- if verbose > 1:
- print('added refname %s' % l, file=sys.stderr)
+ refname = lxml.etree.SubElement(refnamediv, 'refname')
+ refname.text = l
+ if verbose > 1:
+ print('added refname %s' % l, file=sys.stderr)
else:
- if verbose > 1:
- print('duplicate refname %s discarded' % l, file=sys.stderr)
+ if verbose > 1:
+ print('duplicate refname %s discarded' % l, file=sys.stderr)
# to validate, it seems we need to maintain refnamediv elements in a certain order
refnamediv[:] = sorted(refnamediv, key = lambda x: x.tag)
@@ -293,28 +293,28 @@ def synopsis(c, t):
s = ''
for l in t.splitlines():
- if re.match('\s*(#|\[|struct)', l):
- # preprocessor # directives, structs, comments in square brackets
- funcsynopsisinfo = lxml.etree.SubElement(funcsynopsis, 'funcsynopsisinfo')
- funcsynopsisinfo.text = l.strip() + '\n'
- elif re.match('[Ll]ink with', l):
- pass
- else:
- s = s + l
-
- # a prototype without a terminating ';' is an error
- if s.endswith(')'):
- print("'%s' missing terminating semicolon" % l, file=sys.stderr)
- s = s + ';'
- exit(1)
-
- if ';' in s:
- synopsis_for_prototype(funcsynopsis, s)
- s = ''
+ if re.match('\s*(#|\[|struct)', l):
+ # preprocessor # directives, structs, comments in square brackets
+ funcsynopsisinfo = lxml.etree.SubElement(funcsynopsis, 'funcsynopsisinfo')
+ funcsynopsisinfo.text = l.strip() + '\n'
+ elif re.match('[Ll]ink with', l):
+ pass
+ else:
+ s = s + l
+
+ # a prototype without a terminating ';' is an error
+ if s.endswith(')'):
+ print("'%s' missing terminating semicolon" % l, file=sys.stderr)
+ s = s + ';'
+ exit(1)
+
+ if ';' in s:
+ synopsis_for_prototype(funcsynopsis, s)
+ s = ''
if s.strip():
- print("surplus synopsis '%s'" % s, file=sys.stderr)
- exit(1)
+ print("surplus synopsis '%s'" % s, file=sys.stderr)
+ exit(1)
def synopsis_for_prototype(funcsynopsis, s):
s = s.strip()
@@ -323,48 +323,48 @@ def synopsis_for_prototype(funcsynopsis, s):
# bare prototype into it. Fortunately, since the parameter names are marked
# up, we have enough information to do this.
for fp in s.split(';'):
- fp = fp.strip()
- if fp:
+ fp = fp.strip()
+ if fp:
- if verbose:
- print("'%s'" % fp, file=sys.stderr)
+ if verbose:
+ print("'%s'" % fp, file=sys.stderr)
- match = re.match(r'(.*?)([\w\d]*) ?\((.*)\)', fp)
+ match = re.match(r'(.*?)([\w\d]*) ?\((.*)\)', fp)
- if verbose:
- print(match.groups(), file=sys.stderr)
+ if verbose:
+ print(match.groups(), file=sys.stderr)
- funcprototype = lxml.etree.SubElement(funcsynopsis, 'funcprototype')
- funcdef = lxml.etree.SubElement(funcprototype, 'funcdef')
- funcdef.text = match.group(1)
- function = lxml.etree.SubElement(funcdef, 'function')
- function.text = match.group(2)
+ funcprototype = lxml.etree.SubElement(funcsynopsis, 'funcprototype')
+ funcdef = lxml.etree.SubElement(funcprototype, 'funcdef')
+ funcdef.text = match.group(1)
+ function = lxml.etree.SubElement(funcdef, 'function')
+ function.text = match.group(2)
- if match.group(3).strip() == 'void':
- void = lxml.etree.SubElement(funcprototype, 'void')
- else:
- # Split parameters on ',' except if it is inside ()
- for p in re.split(',(?![^()]*\))', match.group(3)):
- p = p.strip()
+ if match.group(3).strip() == 'void':
+ void = lxml.etree.SubElement(funcprototype, 'void')
+ else:
+ # Split parameters on ',' except if it is inside ()
+ for p in re.split(',(?![^()]*\))', match.group(3)):
+ p = p.strip()
- if verbose:
- print(p, file=sys.stderr)
+ if verbose:
+ print(p, file=sys.stderr)
- if p == '...':
- varargs = lxml.etree.SubElement(funcprototype, 'varargs')
- else:
- paramdef = lxml.etree.SubElement(funcprototype, 'paramdef')
- parameter = lxml.etree.SubElement(paramdef, 'parameter')
+ if p == '...':
+ varargs = lxml.etree.SubElement(funcprototype, 'varargs')
+ else:
+ paramdef = lxml.etree.SubElement(funcprototype, 'paramdef')
+ parameter = lxml.etree.SubElement(paramdef, 'parameter')
- # <[ ]> enclose the parameter name
- match2 = re.match('(.*)<\[(.*)\]>(.*)', p)
+ # <[ ]> enclose the parameter name
+ match2 = re.match('(.*)<\[(.*)\]>(.*)', p)
- if verbose:
- print(match2.groups(), file=sys.stderr)
+ if verbose:
+ print(match2.groups(), file=sys.stderr)
- paramdef.text = match2.group(1)
- parameter.text = match2.group(2)
- parameter.tail = match2.group(3)
+ paramdef.text = match2.group(1)
+ parameter.text = match2.group(2)
+ parameter.tail = match2.group(3)
# DESCRIPTION
@@ -384,15 +384,15 @@ def refsect(t, s):
title.text = t.title()
if verbose:
- print('%s has %d paragraphs' % (t, len(s.split('\n\n'))) , file=sys.stderr)
+ print('%s has %d paragraphs' % (t, len(s.split('\n\n'))) , file=sys.stderr)
if verbose > 1:
- dump(s, 'before lexing')
+ dump(s, 'before lexing')
- # dump out lexer token sequence
- lex.input(s)
- for tok in lexer:
- print(tok, file=sys.stderr)
+ # dump out lexer token sequence
+ lex.input(s)
+ for tok in lexer:
+ print(tok, file=sys.stderr)
# parse the section text for makedoc markup and the few pieces of texinfo
# markup we understand, and output an XML marked-up string
@@ -421,25 +421,25 @@ def discarded(c, t):
return
command_dispatch_dict = {
- 'FUNCTION' : function,
- 'TYPEDEF' : function, # TYPEDEF is not currently used, but described in doc.str
- 'INDEX' : index,
- 'TRAD_SYNOPSIS' : discarded, # K&R-style synopsis, obsolete and discarded
- 'ANSI_SYNOPSIS' : synopsis,
- 'SYNOPSIS' : synopsis,
- 'DESCRIPTION' : refsect,
- 'RETURNS' : refsect,
- 'ERRORS' : refsect,
- 'PORTABILITY' : refsect,
- 'BUGS' : refsect,
- 'WARNINGS' : refsect,
- 'SEEALSO' : seealso,
- 'NOTES' : refsect, # NOTES is not described in doc.str, so is currently discarded by makedoc, but that doesn't seem right
- 'QUICKREF' : discarded, # The intent of QUICKREF and MATHREF is not obvious, but they don't generate any output currently
- 'MATHREF' : discarded,
- 'START' : discarded, # a START command is inserted to contain the text before the first command
- 'END' : discarded, # an END command is inserted merely to terminate the text for the last command in a comment block
- 'NEWPAGE' : newpage,
+ 'FUNCTION' : function,
+ 'TYPEDEF' : function, # TYPEDEF is not currently used, but described in doc.str
+ 'INDEX' : index,
+ 'TRAD_SYNOPSIS' : discarded, # K&R-style synopsis, obsolete and discarded
+ 'ANSI_SYNOPSIS' : synopsis,
+ 'SYNOPSIS' : synopsis,
+ 'DESCRIPTION' : refsect,
+ 'RETURNS' : refsect,
+ 'ERRORS' : refsect,
+ 'PORTABILITY' : refsect,
+ 'BUGS' : refsect,
+ 'WARNINGS' : refsect,
+ 'SEEALSO' : seealso,
+ 'NOTES' : refsect, # NOTES is not described in doc.str, so is currently discarded by makedoc, but that doesn't seem right
+ 'QUICKREF' : discarded, # The intent of QUICKREF and MATHREF is not obvious, but they don't generate any output currently
+ 'MATHREF' : discarded,
+ 'START' : discarded, # a START command is inserted to contain the text before the first command
+ 'END' : discarded, # an END command is inserted merely to terminate the text for the last command in a comment block
+ 'NEWPAGE' : newpage,
}
#
@@ -483,7 +483,7 @@ def line_markup_convert(p):
s = s.replace('@*', '</para><para>')
if (verbose > 3) and (s != p):
- print('%s-> line_markup_convert ->\n%s' % (p, s), file=sys.stderr)
+ print('%s-> line_markup_convert ->\n%s' % (p, s), file=sys.stderr)
return s
@@ -529,9 +529,9 @@ def t_TEXINFO(t):
# if the line starts with a known texinfo command, change t.type to the
# token for that command
for k in texinfo_commands.keys():
- if t.value[1:].startswith(k):
- t.type = texinfo_commands[k]
- break
+ if t.value[1:].startswith(k):
+ t.type = texinfo_commands[k]
+ break
return t
@@ -575,10 +575,10 @@ def t_BLANKLINE(t):
def t_eof(t):
if hasattr(t.lexer,'at_eof'):
- # remove eof flag ready for lexing next input
- delattr(t.lexer,'at_eof')
- t.lexer.lineno = 0
- return None
+ # remove eof flag ready for lexing next input
+ delattr(t.lexer,'at_eof')
+ t.lexer.lineno = 0
+ return None
t.type = 'EOF'
t.lexer.at_eof = True;
@@ -598,15 +598,15 @@ lexer = lex.lex()
def parser_verbose(p):
if verbose > 2:
- print(p[0], file=sys.stderr)
+ print(p[0], file=sys.stderr)
def p_input(p):
'''input : paragraph
| input paragraph'''
if len(p) == 3:
- p[0] = p[1] + '\n' + p[2]
+ p[0] = p[1] + '\n' + p[2]
else:
- p[0] = p[1]
+ p[0] = p[1]
parser_verbose(p)
# Strictly, text at top level should be paragraphs (i.e terminated by a
@@ -621,9 +621,9 @@ def p_paragraph_content(p):
'''paragraph_content : paragraph_line
| paragraph_line paragraph_content'''
if len(p) == 3:
- p[0] = p[1] + p[2]
+ p[0] = p[1] + p[2]
else:
- p[0] = p[1]
+ p[0] = p[1]
parser_verbose(p)
def p_paragraph_line(p):
@@ -649,9 +649,9 @@ def p_maybe_lines(p):
'''maybe_lines : empty
| paragraph maybe_lines'''
if len(p) == 3:
- p[0] = p[1] + p[2]
+ p[0] = p[1] + p[2]
else:
- p[0] = p[1]
+ p[0] = p[1]
parser_verbose(p)
def p_maybe_blankline(p):
@@ -668,32 +668,32 @@ def p_courier(p):
'''courier : COURIER
| COURIER courier'''
if len(p) == 3:
- p[0] = p[1] + p[2]
+ p[0] = p[1] + p[2]
else:
- p[0] = p[1]
+ p[0] = p[1]
parser_verbose(p)
def p_bullet(p):
'''bullet : ITEM maybe_lines
| ITEM BLANKLINE maybe_lines'''
if len(p) == 3:
- # Glue any text in ITEM into the first para of maybe_lines
- # (This is an unfortunate consequence of the line-based tokenization we do)
- if p[2].startswith('<para>'):
- p[0] = '<listitem><para>' + p[1] + p[2][len('<para>'):] + '</listitem>'
- else:
- p[0] = '<listitem><para>' + p[1] + '</para>' + p[2] + '</listitem>'
+ # Glue any text in ITEM into the first para of maybe_lines
+ # (This is an unfortunate consequence of the line-based tokenization we do)
+ if p[2].startswith('<para>'):
+ p[0] = '<listitem><para>' + p[1] + p[2][len('<para>'):] + '</listitem>'
+ else:
+ p[0] = '<listitem><para>' + p[1] + '</para>' + p[2] + '</listitem>'
else:
- p[0] = '<listitem><para>' + p[1] + '</para>' + p[3] + '</listitem>'
+ p[0] = '<listitem><para>' + p[1] + '</para>' + p[3] + '</listitem>'
parser_verbose(p)
def p_bullets(p):
'''bullets : bullet
| bullet bullets'''
if len(p) == 3:
- p[0] = p[1] + '\n' + p[2]
+ p[0] = p[1] + '\n' + p[2]
else:
- p[0] = p[1]
+ p[0] = p[1]
parser_verbose(p)
def p_bulletlist(p):
@@ -705,18 +705,18 @@ def p_row(p):
'''row : ITEM maybe_lines
| ITEM BLANKLINE maybe_lines'''
if len(p) == 3:
- p[0] = '<row><entry><code>' + p[1] + '</code></entry><entry>' + p[2] + '</entry></row>'
+ p[0] = '<row><entry><code>' + p[1] + '</code></entry><entry>' + p[2] + '</entry></row>'
else:
- p[0] = '<row><entry><code>' + p[1] + '</code></entry><entry>' + p[3] + '</entry></row>'
+ p[0] = '<row><entry><code>' + p[1] + '</code></entry><entry>' + p[3] + '</entry></row>'
parser_verbose(p)
def p_rows(p):
'''rows : row
| row rows'''
if len(p) == 3:
- p[0] = p[1] + '\n' + p[2]
+ p[0] = p[1] + '\n' + p[2]
else:
- p[0] = p[1]
+ p[0] = p[1]
parser_verbose(p)
def p_table(p):
@@ -756,9 +756,9 @@ def p_mct_columns(p):
'''mct_columns : maybe_lines
| maybe_lines MCT_COLUMN_SEPARATOR mct_columns'''
if len(p) == 4:
- p[0] = '<entry>' + p[1] + '</entry>' + p[3]
+ p[0] = '<entry>' + p[1] + '</entry>' + p[3]
else:
- p[0] = '<entry>' + p[1] + '</entry>'
+ p[0] = '<entry>' + p[1] + '</entry>'
parser_verbose(p)
def p_mct_row(p):
@@ -770,9 +770,9 @@ def p_mct_rows(p):
'''mct_rows : mct_row
| mct_row mct_rows'''
if len(p) == 3:
- p[0] = p[1] + '\n' + p[2]
+ p[0] = p[1] + '\n' + p[2]
else:
- p[0] = p[1]
+ p[0] = p[1]
parser_verbose(p)
def p_mct_header(p):
@@ -810,15 +810,15 @@ def main(file):
s = lxml.etree.tostring(rootelement, pretty_print=True)
if not s:
- print('No output produced (perhaps the input has no makedoc markup?)', file=sys.stderr)
- exit(1)
+ print('No output produced (perhaps the input has no makedoc markup?)', file=sys.stderr)
+ exit(1)
print(s)
# warn about texinfo commands which didn't get processed
match = re.search('@[a-z*]+', s)
if match:
- print('texinfo command %s remains in output' % match.group(), file=sys.stderr)
+ print('texinfo command %s remains in output' % match.group(), file=sys.stderr)
#
#
@@ -831,11 +831,11 @@ if __name__ == '__main__' :
(opts, args) = options.parse_args()
if opts.cache:
- sys.exit()
+ sys.exit()
verbose = opts.verbose
if len(args) > 0:
- main(open(args[0], 'rb'))
+ main(open(args[0], 'rb'))
else:
- main(sys.stdin)
+ main(sys.stdin)