92bad0a60a
Fix #5822
224 lines
6.7 KiB
Python
224 lines
6.7 KiB
Python
#!/usr/bin/env python3
|
|
# CheckMentioned.py
|
|
# Find all the symbols in scintilla/include/Scintilla.h and check if they
|
|
# are mentioned in scintilla/doc/ScintillaDoc.html.
|
|
# Requires Python 2.7 or later
|
|
|
|
import re, string, sys
|
|
|
|
srcRoot = "../.."
|
|
|
|
sys.path.append(srcRoot + "/scintilla/scripts")
|
|
|
|
import Face
|
|
import ScintillaData
|
|
|
|
uninteresting = {
|
|
"SCINTILLA_H", "SCI_START", "SCI_LEXER_START", "SCI_OPTIONAL_START",
|
|
# These archaic names are #defined to the Sci_ prefixed modern equivalents.
|
|
# They are not documented so they are not used in new code.
|
|
"CharacterRange", "TextRange", "TextToFind", "RangeToFormat", "NotifyHeader",
|
|
}
|
|
|
|
incFileName = srcRoot + "/scintilla/include/Scintilla.h"
|
|
docFileName = srcRoot + "/scintilla/doc/ScintillaDoc.html"
|
|
try: # Old Python
|
|
identCharacters = "_" + string.letters + string.digits
|
|
except AttributeError: # Python 3.x
|
|
identCharacters = "_" + string.ascii_letters + string.digits
|
|
|
|
# Convert all punctuation characters except '_' into spaces.
|
|
def depunctuate(s):
|
|
d = ""
|
|
for ch in s:
|
|
if ch in identCharacters:
|
|
d = d + ch
|
|
else:
|
|
d = d + " "
|
|
return d
|
|
|
|
symbols = {}
|
|
with open(incFileName, "rt") as incFile:
|
|
for line in incFile.readlines():
|
|
if line.startswith("#define"):
|
|
identifier = line.split()[1]
|
|
symbols[identifier] = 0
|
|
|
|
with open(docFileName, "rt") as docFile:
|
|
for line in docFile.readlines():
|
|
for word in depunctuate(line).split():
|
|
if word in symbols.keys():
|
|
symbols[word] = 1
|
|
|
|
def convertIFaceTypeToC(t):
|
|
if t == "keymod":
|
|
return "int "
|
|
elif t == "string":
|
|
return "const char *"
|
|
elif t == "stringresult":
|
|
return "char *"
|
|
elif t == "cells":
|
|
return "cell *"
|
|
elif t == "textrange":
|
|
return "Sci_TextRange *"
|
|
elif t == "findtext":
|
|
return "Sci_TextToFind *"
|
|
elif t == "formatrange":
|
|
return "Sci_RangeToFormat *"
|
|
elif Face.IsEnumeration(t):
|
|
return "int "
|
|
return t + " "
|
|
|
|
def makeParm(t, n, v):
|
|
return (convertIFaceTypeToC(t) + n).rstrip()
|
|
|
|
def makeRet(params):
|
|
retType = params["ReturnType"]
|
|
if retType in ["void", "string", "stringresult"]:
|
|
retType = ""
|
|
if retType:
|
|
retType = " → " + retType
|
|
|
|
return retType
|
|
|
|
def makeSig(params):
|
|
p1 = makeParm(params["Param1Type"], params["Param1Name"], params["Param1Value"])
|
|
p2 = makeParm(params["Param2Type"], params["Param2Name"], params["Param2Value"])
|
|
|
|
retType = params["ReturnType"]
|
|
if retType in ["void", "string", "stringresult"]:
|
|
retType = ""
|
|
elif Face.IsEnumeration(retType):
|
|
retType = "int"
|
|
if retType:
|
|
retType = " → " + retType
|
|
|
|
if p1 == "" and p2 == "":
|
|
return retType
|
|
|
|
ret = ""
|
|
if p1 == "":
|
|
p1 = "<unused>"
|
|
joiner = ""
|
|
if p2 != "":
|
|
joiner = ", "
|
|
return "(" + p1 + joiner + p2 + ")" + retType
|
|
|
|
pathIface = srcRoot + "/scintilla/include/Scintilla.iface"
|
|
|
|
def retrieveFeatures():
|
|
sci = ScintillaData.ScintillaData(srcRoot + "/scintilla/")
|
|
face = Face.Face()
|
|
face.ReadFromFile(pathIface)
|
|
sciToFeature = {}
|
|
sccToValue = { "true":"1", "false":"0", "EN_SETFOCUS":"256", "EN_KILLFOCUS":"512"}
|
|
for name in face.order:
|
|
v = face.features[name]
|
|
if v["FeatureType"] in ["fun", "get", "set"]:
|
|
featureDefineName = "SCI_" + name.upper()
|
|
sciToFeature[featureDefineName] = name
|
|
elif v["FeatureType"] in ["val"]:
|
|
featureDefineName = name.upper()
|
|
sccToValue[featureDefineName] = v["Value"]
|
|
elif v["FeatureType"] in ["evt"]:
|
|
featureDefineName = "SCN_" + name.upper()
|
|
sccToValue[featureDefineName] = v["Value"]
|
|
return (face, sciToFeature, sccToValue)
|
|
|
|
def flattenSpaces(s):
|
|
return s.replace("\n", " ").replace(" ", " ").replace(" ", " ").replace(" ", " ").strip()
|
|
|
|
def printCtag(ident, path):
|
|
print(ident.strip() + "\t" + path + "\t" + "/^" + ident + "$/")
|
|
|
|
showCTags = True
|
|
|
|
def checkDocumentation():
|
|
with open(docFileName, "rt") as docFile:
|
|
docs = docFile.read()
|
|
|
|
face, sciToFeature, sccToValue = retrieveFeatures()
|
|
|
|
headers = {}
|
|
definitions = {}
|
|
|
|
# Examine header sections which point to definitions
|
|
#<a class="message" href="#SCI_SETLAYOUTCACHE">SCI_SETLAYOUTCACHE(int cacheMode)</a><br />
|
|
dirPattern = re.compile(r'<a class="message" href="#([A-Z0-9_]+)">([A-Z][A-Za-z0-9_() *&;,\n]+)</a>')
|
|
firstWord = re.compile(r'[A-Z0-9_]+')
|
|
for api, sig in re.findall(dirPattern, docs):
|
|
sigApi = re.split('\W+', sig)[0]
|
|
sigFlat = flattenSpaces(sig)
|
|
sigFlat = sigFlat.replace('alpha ', 'int ')
|
|
sigFlat = sigFlat.replace("document *", "int ")
|
|
sigFlat = sigFlat.rstrip()
|
|
if '(' in sigFlat or api.startswith("SCI_"):
|
|
name = sciToFeature[api]
|
|
sigFromFace = api + makeSig(face.features[name])
|
|
if sigFlat != sigFromFace:
|
|
print(sigFlat, "|", sigFromFace)
|
|
if showCTags:
|
|
printCtag(api, docFileName)
|
|
#~ printCtag(" " + name, pathIface)
|
|
if api != sigApi:
|
|
print(sigApi, ";;", sig, ";;", api)
|
|
headers[api] = 1
|
|
# Warns for most keyboard commands so not enabled
|
|
#~ for api in sorted(sciToFeature.keys()):
|
|
#~ if api not in headers:
|
|
#~ print("No header for ", api)
|
|
|
|
# Examine definitions
|
|
#<b id="SCI_SETLAYOUTCACHE">SCI_SETLAYOUTCACHE(int cacheMode)</b>
|
|
defPattern = re.compile(r'<b id="([A-Z_0-9]+)">([A-Z][A-Za-z0-9_() *#\"=<>/&;,\n-]+?)</b>')
|
|
for api, sig in re.findall(defPattern, docs):
|
|
sigFlat = flattenSpaces(sig)
|
|
if '<a' in sigFlat : # Remove anchors
|
|
sigFlat = re.sub('<a.*>(.+)</a>', '\\1', sigFlat)
|
|
sigFlat = sigFlat.replace('alpha ', 'int ')
|
|
sigFlat = sigFlat.replace("document *", "int ")
|
|
sigFlat = sigFlat.replace(' NUL-terminated', '')
|
|
sigFlat = sigFlat.rstrip()
|
|
#~ sigFlat = sigFlat.replace(' NUL-terminated', '')
|
|
sigApi = re.split('\W+', sigFlat)[0]
|
|
#~ print(sigFlat, ";;", sig, ";;", api)
|
|
if '(' in sigFlat or api.startswith("SCI_"):
|
|
try:
|
|
name = sciToFeature[api]
|
|
sigFromFace = api + makeSig(face.features[name])
|
|
if sigFlat != sigFromFace:
|
|
print(sigFlat, "|", sigFromFace)
|
|
if showCTags:
|
|
printCtag('="' + api, docFileName)
|
|
#~ printCtag(" " + name, pathIface)
|
|
except KeyError:
|
|
pass # Feature removed but still has documentation
|
|
if api != sigApi:
|
|
print(sigApi, ";;", sig, ";;", api)
|
|
definitions[api] = 1
|
|
# Warns for most keyboard commands so not enabled
|
|
#~ for api in sorted(sciToFeature.keys()):
|
|
#~ if api not in definitions:
|
|
#~ print("No definition for ", api)
|
|
|
|
outName = docFileName.replace("Doc", "Dox")
|
|
with open(outName, "wt") as docFile:
|
|
docFile.write(docs)
|
|
|
|
# Examine constant definitions
|
|
#<code>SC_CARETSTICKY_WHITESPACE</code> (2)
|
|
constPattern = re.compile(r'<code>(\w+)</code> *\((\w+)\)')
|
|
for name, val in re.findall(constPattern, docs):
|
|
try:
|
|
valOfName = sccToValue[name]
|
|
if val != valOfName:
|
|
print(val, "<-", name, ";;", valOfName)
|
|
except KeyError:
|
|
print("***", val, "<-", name)
|
|
|
|
for identifier in sorted(symbols.keys()):
|
|
if not symbols[identifier] and identifier not in uninteresting:
|
|
print(identifier)
|
|
|
|
checkDocumentation()
|