summaryrefslogtreecommitdiffstats
path: root/Tools/bgen
diff options
context:
space:
mode:
authorJack Jansen <jack.jansen@cwi.nl>2001-07-13 22:28:36 (GMT)
committerJack Jansen <jack.jansen@cwi.nl>2001-07-13 22:28:36 (GMT)
commitb2a1de4679c9141b03781551356c51cd48f20519 (patch)
tree6d8a64b147840f2cbfbab28dcddb0cae0fbb79c4 /Tools/bgen
parentd4b83619102c1e0d51fac7c3773ed1943ed516fa (diff)
downloadcpython-b2a1de4679c9141b03781551356c51cd48f20519.zip
cpython-b2a1de4679c9141b03781551356c51cd48f20519.tar.gz
cpython-b2a1de4679c9141b03781551356c51cd48f20519.tar.bz2
Allow [] after a parameter name. We currently take this to be the same as * in front, which isn't 100% correct but good enough.
Diffstat (limited to 'Tools/bgen')
-rw-r--r--Tools/bgen/bgen/scantools.py20
1 files changed, 7 insertions, 13 deletions
diff --git a/Tools/bgen/bgen/scantools.py b/Tools/bgen/bgen/scantools.py
index 7dc46a9..23da86a 100644
--- a/Tools/bgen/bgen/scantools.py
+++ b/Tools/bgen/bgen/scantools.py
@@ -244,7 +244,7 @@ if missing: raise "Missing Types"
self.whole_pat = self.type_pat + self.name_pat + self.args_pat
self.sym_pat = "^[ \t]*\(<name>[a-zA-Z0-9_]+\)[ \t]*=" + \
"[ \t]*\(<defn>[-0-9_a-zA-Z'\"(][^\t\n,;}]*\),?"
- self.asplit_pat = "^\(<type>.*[^a-zA-Z0-9_]\)\(<name>[a-zA-Z0-9_]+\)$"
+ self.asplit_pat = "^\(<type>.*[^a-zA-Z0-9_]\)\(<name>[a-zA-Z0-9_]+\)\(<array>\[\]\)?$"
self.comment1_pat = "\(<rest>.*\)//.*"
# note that the next pattern only removes comments that are wholly within one line
self.comment2_pat = "\(<rest1>.*\)/\*.*\*/\(<rest2>.*\)"
@@ -470,8 +470,12 @@ if missing: raise "Missing Types"
mode = "InMode"
if self.asplit.match(part) < 0:
self.error("Indecipherable argument: %s", `part`)
+ import pdb ; pdb.set_trace()
return ("unknown", part, mode)
- type, name = self.asplit.group('type', 'name')
+ type, name, array = self.asplit.group('type', 'name', 'array')
+ if array:
+ # array matches an optional [] after the argument name
+ type = type + " ptr "
type = regsub.gsub("\*", " ptr ", type)
type = string.strip(type)
type = regsub.gsub("[ \t]+", "_", type)
@@ -573,10 +577,7 @@ class Scanner_PreUH3(Scanner):
def initpatterns(self):
Scanner.initpatterns(self)
self.head_pat = "^extern pascal[ \t]+" # XXX Mac specific!
- self.tail_pat = "[;={}]"
self.type_pat = "pascal[ \t\n]+\(<type>[a-zA-Z0-9_ \t]*[a-zA-Z0-9_]\)[ \t\n]+"
- self.name_pat = "\(<name>[a-zA-Z0-9_]+\)[ \t\n]*"
- self.args_pat = "(\(<args>\([^(;=)]+\|([^(;=)]*)\)*\))"
self.whole_pat = self.type_pat + self.name_pat + self.args_pat
self.sym_pat = "^[ \t]*\(<name>[a-zA-Z0-9_]+\)[ \t]*=" + \
"[ \t]*\(<defn>[-0-9'\"][^\t\n,;}]*\),?"
@@ -585,22 +586,15 @@ class Scanner_PreUH3(Scanner):
class Scanner_OSX(Scanner):
"""Scanner for modern (post UH3.3) Universal Headers """
def initpatterns(self):
+ Scanner.initpatterns(self)
self.head_pat = "^EXTERN_API_C"
- self.tail_pat = "[;={}]"
self.type_pat = "EXTERN_API_C" + \
"[ \t\n]*([ \t\n]*" + \
"\(<type>[a-zA-Z0-9_* \t]*[a-zA-Z0-9_*]\)" + \
"[ \t\n]*)[ \t\n]*"
- self.name_pat = "\(<name>[a-zA-Z0-9_]+\)[ \t\n]*"
- self.args_pat = "(\(<args>\([^(;=)]+\|([^(;=)]*)\)*\))"
self.whole_pat = self.type_pat + self.name_pat + self.args_pat
self.sym_pat = "^[ \t]*\(<name>[a-zA-Z0-9_]+\)[ \t]*=" + \
"[ \t]*\(<defn>[-0-9_a-zA-Z'\"(][^\t\n,;}]*\),?"
- self.asplit_pat = "^\(<type>.*[^a-zA-Z0-9_]\)\(<name>[a-zA-Z0-9_]+\)$"
- self.comment1_pat = "\(<rest>.*\)//.*"
- # note that the next pattern only removes comments that are wholly within one line
- self.comment2_pat = "\(<rest1>.*\)/\*.*\*/\(<rest2>.*\)"
-
def test():
input = "D:Development:THINK C:Mac #includes:Apple #includes:AppleEvents.h"