aboutsummaryrefslogtreecommitdiff
path: root/libxml2-2.9.10/doc/examples/index.py
diff options
context:
space:
mode:
Diffstat (limited to 'libxml2-2.9.10/doc/examples/index.py')
-rwxr-xr-xlibxml2-2.9.10/doc/examples/index.py315
1 files changed, 315 insertions, 0 deletions
diff --git a/libxml2-2.9.10/doc/examples/index.py b/libxml2-2.9.10/doc/examples/index.py
new file mode 100755
index 0000000..ac44ec5
--- /dev/null
+++ b/libxml2-2.9.10/doc/examples/index.py
@@ -0,0 +1,315 @@
+#!/usr/bin/python -u
+#
+# Indexes the examples and build an XML description
+#
+import string
+import glob
+import sys
+try:
+ import libxml2
+except:
+ sys.exit(1)
+sys.path.insert(0, "..")
+from apibuild import CParser, escape
+
+examples = []
+extras = ['examples.xsl', 'index.html', 'index.py']
+tests = []
+sections = {}
+symbols = {}
+api_dict = None
+api_doc = None
+
+def load_api():
+ global api_dict
+ global api_doc
+
+ if api_dict != None:
+ return
+ api_dict = {}
+ try:
+ print "loading ../libxml2-api.xml"
+ api_doc = libxml2.parseFile("../libxml2-api.xml")
+ except:
+ print "failed to parse ../libxml2-api.xml"
+ sys.exit(1)
+
+def find_symbol(name):
+ global api_dict
+ global api_doc
+
+ if api_doc == None:
+ load_api()
+
+ if name == None:
+ return
+ if api_dict.has_key(name):
+ return api_dict[name]
+ ctxt = api_doc.xpathNewContext()
+ res = ctxt.xpathEval("/api/symbols/*[@name = '%s']" % (name))
+ if type(res) == type([]) and len(res) >= 1:
+ if len(res) > 1:
+ print "Found %d references to %s in the API" % (len(res), name)
+ node = res[0]
+ typ = node.name
+ file = node.xpathEval("string(@file)")
+ info = node.xpathEval("string(info)")
+ else:
+ print "Reference %s not found in the API" % (name)
+ return None
+ ret = (typ, file, info)
+ api_dict[name] = ret
+ return ret
+
+def parse_top_comment(filename, comment):
+ res = {}
+ lines = string.split(comment, "\n")
+ item = None
+ for line in lines:
+ while line != "" and (line[0] == ' ' or line[0] == '\t'):
+ line = line[1:]
+ while line != "" and line[0] == '*':
+ line = line[1:]
+ while line != "" and (line[0] == ' ' or line[0] == '\t'):
+ line = line[1:]
+ try:
+ (it, line) = string.split(line, ":", 1)
+ item = it
+ while line != "" and (line[0] == ' ' or line[0] == '\t'):
+ line = line[1:]
+ if res.has_key(item):
+ res[item] = res[item] + " " + line
+ else:
+ res[item] = line
+ except:
+ if item != None:
+ if res.has_key(item):
+ res[item] = res[item] + " " + line
+ else:
+ res[item] = line
+ return res
+
+def parse(filename, output):
+ global symbols
+ global sections
+
+ parser = CParser(filename)
+ parser.collect_references()
+ idx = parser.parse()
+ info = parse_top_comment(filename, parser.top_comment)
+ output.write(" <example filename='%s'>\n" % filename)
+ try:
+ synopsis = info['synopsis']
+ output.write(" <synopsis>%s</synopsis>\n" % escape(synopsis));
+ except:
+ print "Example %s lacks a synopsis description" % (filename)
+ try:
+ purpose = info['purpose']
+ output.write(" <purpose>%s</purpose>\n" % escape(purpose));
+ except:
+ print "Example %s lacks a purpose description" % (filename)
+ try:
+ usage = info['usage']
+ output.write(" <usage>%s</usage>\n" % escape(usage));
+ except:
+ print "Example %s lacks an usage description" % (filename)
+ try:
+ test = info['test']
+ output.write(" <test>%s</test>\n" % escape(test));
+ progname=filename[0:-2]
+ command=string.replace(test, progname, './' + progname, 1)
+ tests.append(command)
+ except:
+ pass
+ try:
+ author = info['author']
+ output.write(" <author>%s</author>\n" % escape(author));
+ except:
+ print "Example %s lacks an author description" % (filename)
+ try:
+ copy = info['copy']
+ output.write(" <copy>%s</copy>\n" % escape(copy));
+ except:
+ print "Example %s lacks a copyright description" % (filename)
+ try:
+ section = info['section']
+ output.write(" <section>%s</section>\n" % escape(section));
+ if sections.has_key(section):
+ sections[section].append(filename)
+ else:
+ sections[section] = [filename]
+ except:
+ print "Example %s lacks a section description" % (filename)
+ for topic in info.keys():
+ if topic != "purpose" and topic != "usage" and \
+ topic != "author" and topic != "copy" and \
+ topic != "section" and topic != "synopsis" and topic != "test":
+ str = info[topic]
+ output.write(" <extra topic='%s'>%s</extra>\n" % (
+ escape(topic), escape(str)))
+ output.write(" <includes>\n")
+ for include in idx.includes.keys():
+ if include.find("libxml") != -1:
+ output.write(" <include>%s</include>\n" % (escape(include)))
+ output.write(" </includes>\n")
+ output.write(" <uses>\n")
+ for ref in idx.references.keys():
+ id = idx.references[ref]
+ name = id.get_name()
+ line = id.get_lineno()
+ if symbols.has_key(name):
+ sinfo = symbols[name]
+ refs = sinfo[0]
+ # gather at most 5 references per symbols
+ if refs > 5:
+ continue
+ sinfo.append(filename)
+ sinfo[0] = refs + 1
+ else:
+ symbols[name] = [1, filename]
+ info = find_symbol(name)
+ if info != None:
+ type = info[0]
+ file = info[1]
+ output.write(" <%s line='%d' file='%s' name='%s'/>\n" % (type,
+ line, file, name))
+ else:
+ type = id.get_type()
+ output.write(" <%s line='%d' name='%s'/>\n" % (type,
+ line, name))
+
+ output.write(" </uses>\n")
+ output.write(" </example>\n")
+
+ return idx
+
+def dump_symbols(output):
+ global symbols
+
+ output.write(" <symbols>\n")
+ keys = symbols.keys()
+ keys.sort()
+ for symbol in keys:
+ output.write(" <symbol name='%s'>\n" % (symbol))
+ info = symbols[symbol]
+ i = 1
+ while i < len(info):
+ output.write(" <ref filename='%s'/>\n" % (info[i]))
+ i = i + 1
+ output.write(" </symbol>\n")
+ output.write(" </symbols>\n")
+
+def dump_sections(output):
+ global sections
+
+ output.write(" <sections>\n")
+ keys = sections.keys()
+ keys.sort()
+ for section in keys:
+ output.write(" <section name='%s'>\n" % (section))
+ info = sections[section]
+ i = 0
+ while i < len(info):
+ output.write(" <example filename='%s'/>\n" % (info[i]))
+ i = i + 1
+ output.write(" </section>\n")
+ output.write(" </sections>\n")
+
+def dump_Makefile():
+ for file in glob.glob('*.xml'):
+ extras.append(file)
+ for file in glob.glob('*.res'):
+ extras.append(file)
+ Makefile="""##
+## This file is auto-generated by index.py
+## DO NOT EDIT !!!
+##
+
+AM_CPPFLAGS = -I$(top_builddir)/include -I$(top_srcdir)/include -I$(srcdir)/include
+AM_CFLAGS = $(THREAD_CFLAGS) $(Z_CFLAGS)
+LDADD = $(RDL_LIBS) $(STATIC_BINARIES) $(top_builddir)/libxml2.la $(THREAD_LIBS) $(Z_LIBS) $(ICONV_LIBS) -lm $(WIN32_EXTRA_LIBADD)
+
+CLEANFILES = *.tmp
+
+if REBUILD_DOCS
+rebuild: examples.xml index.html
+.PHONY: rebuild
+
+examples.xml: index.py $(check_PROGRAMS:=.c)
+ cd $(srcdir) && $(PYTHON) index.py
+ $(MAKE) Makefile
+
+index.html: examples.xml examples.xsl
+ cd $(srcdir) && xsltproc examples.xsl examples.xml && echo "Rebuilt web page"
+ -cd $(srcdir) && xmllint --valid --noout index.html
+endif
+
+install-data-local:
+ $(MKDIR_P) $(DESTDIR)$(HTML_DIR)
+ -$(INSTALL) -m 0644 $(srcdir)/*.html $(srcdir)/*.c $(srcdir)/*.xml $(srcdir)/*.xsl $(srcdir)/*.res $(DESTDIR)$(HTML_DIR)
+
+clean-local:
+ test -f Makefile.am || rm -f test?.xml
+
+"""
+ examples.sort()
+ extras.sort()
+ tests.sort()
+ EXTRA_DIST=""
+ for extra in extras:
+ EXTRA_DIST = EXTRA_DIST + " \\\n\t" + extra
+ Makefile = Makefile + "EXTRA_DIST =%s\n\n" % (EXTRA_DIST)
+ check_PROGRAMS=""
+ for example in examples:
+ check_PROGRAMS = check_PROGRAMS + " \\\n\t" + example
+ Makefile = Makefile + "check_PROGRAMS =%s\n\n" % (check_PROGRAMS)
+ for example in examples:
+ Makefile = Makefile + "%s_SOURCES = %s.c\n\n" % (example, example)
+ Makefile = Makefile + "valgrind: \n\t$(MAKE) CHECKER='valgrind' tests\n\n"
+ Makefile = Makefile + "tests: $(check_PROGRAMS)\n"
+ Makefile = Makefile + "\t@test -f Makefile.am || test -f test1.xml || $(LN_S) $(srcdir)/test?.xml .\n"
+ Makefile = Makefile + "\t@(echo '## examples regression tests')\n"
+ Makefile = Makefile + "\t@(echo > .memdump)\n"
+ for test in tests:
+ Makefile = Makefile + "\t@$(CHECKER) %s\n" % (test)
+ Makefile = Makefile + '\t@grep "MORY ALLO" .memdump | grep -v "MEMORY ALLOCATED : 0" ; exit 0\n'
+ Makefile = Makefile + "\t@rm *.tmp\n"
+ try:
+ old = open("Makefile.am", "r").read()
+ if old != Makefile:
+ n = open("Makefile.am", "w").write(Makefile)
+ print "Updated Makefile.am"
+ except:
+ print "Failed to read or save Makefile.am"
+# #
+# # Autogenerate the .cvsignore too ... DEPRECATED
+# #
+# ignore = """.memdump
+#Makefile.in
+#Makefile
+#"""
+# for example in examples:
+# ignore = ignore + "%s\n" % (example)
+# try:
+# old = open(".cvsignore", "r").read()
+# if old != ignore:
+# n = open(".cvsignore", "w").write(ignore)
+# print "Updated .cvsignore"
+# except:
+# print "Failed to read or save .cvsignore"
+
+if __name__ == "__main__":
+ load_api()
+ output = open("examples.xml", "w")
+ output.write("<examples>\n")
+
+ for file in sorted(glob.glob('*.c')):
+ parse(file, output)
+ examples.append(file[:-2])
+
+ dump_symbols(output)
+ dump_sections(output)
+ output.write("</examples>\n")
+ output.close()
+ dump_Makefile()
+