wip lib9p bitfields

5 files changed, 184 insertions, 79 deletions

diff --git a/lib9p/9P2000.txt b/lib9p/9P2000.txt
index 5f93cdf..3f1f10a 100644
--- a/lib9p/9P2000.txt
+++ b/lib9p/9P2000.txt
@@ -30,8 +30,39 @@ d = "len[4] len*(dat[1])"
 # string (u16le `n`, then `n` bytes of UTF-8)
 s = "len[2] len*(utf8[1])"
 
-# qid (TODO)
-qid = "type[1] vers[4] path[8]"
+bitfield qid_type 8
+	7/DIR
+	6/APPEND
+	5/EXCL
+	# QTMOUNT has been around in Plan 9 forever, but is
+	# undocumented, and is explicitly excluded from the 9P2000
+	# draft RFC.  As I understand it, QTMOUNT indicates that the
+	# file is mounted by the kernel as a 9P transport; that the
+	# kernel has a lock on doing I/O on it, so userspace can't do
+	# I/O on it.
+	4/_PLAN9_MOUNT
+	3/AUTH
+	# Fun historical fact: QTTMP was a relatively late addition to
+	# Plan 9, in 2003-12.
+	2/TMP
+	#1/unused
+	FILE=0
+
+# uniQue IDentification - "two files on the same server hierarchy are
+# the same if and only if their qids are the same"
+#
+#  - "path" is a unique uint64_t that does most of the work in the
+#    above statement about files being the same if their QIDs are the
+#    same; " If a file is deleted and recreated with the same name in
+#    the same directory, the old and new path components of the qids
+#    should be different"
+#
+#  - "vers" "is a version number for a file; typically, it is
+#    incremented every time the file is modified.
+#
+#  - "type" indicates "whether the file is a directory, append-only
+#    file, etc."; is an instance of the qid_type bitfield.
+qid = "type[qid_type] vers[4] path[8]"
 
 # stat (TODO)
 stat = "stat_size[2]"
diff --git a/lib9p/9P2000.u.txt b/lib9p/9P2000.u.txt
index 1fa71fb..a1dbcfb 100644
--- a/lib9p/9P2000.u.txt
+++ b/lib9p/9P2000.u.txt
@@ -19,16 +19,8 @@ Tauth += "n_uname[4]"
 
 Rerror += "errno[4]"
 
-# # qid.types
-# QTDIR     = 1<<7
-# QTAPPEND  = 1<<6
-# QTEXCL    = 1<<5
-# QTMOUNT   = 1<<4 # been around forever, but undocumented?
-# QTAUTH    = 1<<3
-# QTTMP     = 1<<2 # added to Plan 9 2003-12
-# QTSYMLINK = 1<<1 # .u
-# QTFILE    = 1<<0
-#
+qid_type += 1/SYMLINK
+
 # DMDIR       = 1<<31
 # DMAPPEND    = 1<<30
 # DMEXCL      = 1<<29
diff --git a/lib9p/include/lib9p/_types.h b/lib9p/include/lib9p/_types.h
index 94accc0..cd1ded5 100644
--- a/lib9p/include/lib9p/_types.h
+++ b/lib9p/include/lib9p/_types.h
@@ -30,9 +30,9 @@ struct lib9p_s {
 };
 
 struct lib9p_qid {
-	uint8_t    type;
-	uint32_t   vers;
-	uint64_t   path;
+	lib9p_qid_type_t   type;
+	uint32_t           vers;
+	uint64_t           path;
 };
 
 struct lib9p_stat {
diff --git a/lib9p/types.c b/lib9p/types.c
index 19ed322..4854ed4 100644
--- a/lib9p/types.c
+++ b/lib9p/types.c
@@ -344,7 +344,7 @@ static ALWAYS_INLINE bool checksize_s(struct _checksize_ctx *ctx) {
 }
 
 static ALWAYS_INLINE bool checksize_qid(struct _checksize_ctx *ctx) {
-	return checksize_1(ctx)
+	return checksize_qid_type(ctx)
 	    || checksize_4(ctx)
 	    || checksize_8(ctx);
 }
@@ -571,7 +571,7 @@ static ALWAYS_INLINE void unmarshal_s(struct _unmarshal_ctx *ctx, struct lib9p_s
 
 static ALWAYS_INLINE void unmarshal_qid(struct _unmarshal_ctx *ctx, struct lib9p_qid *out) {
 	memset(out, 0, sizeof(*out));
-	unmarshal_1(ctx, &out->type);
+	unmarshal_qid_type(ctx, &out->type);
 	unmarshal_4(ctx, &out->vers);
 	unmarshal_8(ctx, &out->path);
 }
@@ -860,7 +860,7 @@ static ALWAYS_INLINE bool marshal_s(struct _marshal_ctx *ctx, struct lib9p_s *va
 }
 
 static ALWAYS_INLINE bool marshal_qid(struct _marshal_ctx *ctx, struct lib9p_qid *val) {
-	return marshal_1(ctx, &val->type)
+	return marshal_qid_type(ctx, &val->type)
 	    || marshal_4(ctx, &val->vers)
 	    || marshal_8(ctx, &val->path);
 }
diff --git a/lib9p/types.gen b/lib9p/types.gen
index 9946e35..1a2d45a 100755
--- a/lib9p/types.gen
+++ b/lib9p/types.gen
@@ -8,7 +8,7 @@
 import enum
 import os.path
 import re
-from typing import Callable
+from typing import Callable, Sequence
 
 # This strives to be "general-purpose" in that it just acts on the
 # *.txt inputs; but (unfortunately?) there are a few special-cases in
@@ -32,6 +32,16 @@ class Atom(enum.Enum):
         return self.value
 
 
+class Bitfield:
+    name: str
+    bits: list[str]
+    aliases: dict[str, str]
+
+    @property
+    def static_size(self) -> int | None:
+        return int((len(self.bits) + 7) / 8)
+
+
 # `msgid/structname = "member1 member2..."`
 # `structname = "member1 member2..."`
 # `structname += "member1 member2..."`
@@ -60,7 +70,7 @@ class Struct:
 class Member:
     cnt: str | None = None
     name: str
-    typ: Atom | Struct
+    typ: Atom | Bitfield | Struct
     ver: set[str]
 
     @property
@@ -77,7 +87,10 @@ re_memberspec = (
 
 
 def parse_members(
-    ver: str, env: dict[str, Atom | Struct], existing: list[Member], specs: str
+    ver: str,
+    env: dict[str, Atom | Bitfield | Struct],
+    existing: list[Member],
+    specs: str,
 ) -> list[Member]:
     ret = existing
     for spec in specs.split():
@@ -112,23 +125,26 @@ re_import = r"from\s+(?P<file>\S+)\s+import\s+(?P<syms>\S+(?:\s*,\s*\S+)*)\s*"
 re_structspec = (
     r'(?:(?P<msgid>[0-9]+)/)?(?P<name>\S+)\s*(?P<op>\+?=)\s*"(?P<members>[^"]*)"'
 )
-re_structspec_cont = r'"(?P<members>[^"]*)"'
+re_structspec_cont = r'\s+"(?P<members>[^"]*)"'
+re_bitfieldspec = r"bitfield\s+(?P<name>\S+)\s+(?P<size>[0-9]+)"
+re_bitfieldspec_bit = r"(?:\s+|(?P<bitfield>\S+)\s*\+=\s*)(?P<bit>[0-9]+)/(?P<name>\S+)"
+re_bitfieldspec_alias = r"(?:\s+|(?P<bitfield>\S+)\s*\+=\s*)(?P<name>\S+)=(?P<val>.*)"
 
 
 def parse_file(
-    filename: str, get_include: Callable[[str], tuple[str, list[Struct]]]
-) -> tuple[str, list[Struct]]:
+    filename: str, get_include: Callable[[str], tuple[str, list[Bitfield | Struct]]]
+) -> tuple[str, list[Bitfield | Struct]]:
     version: str | None = None
-    env: dict[str, Atom | Struct] = {
+    env: dict[str, Atom | Bitfield | Struct] = {
         "1": Atom.u8,
         "2": Atom.u16,
         "4": Atom.u32,
         "8": Atom.u64,
     }
     with open(filename, "r") as fh:
-        prev: Struct | None = None
+        prev: Struct | Bitfield | None = None
         for line in fh:
-            line = line.split("#", 1)[0].strip()
+            line = line.split("#", 1)[0].rstrip()
             if not line:
                 continue
             if m := re.fullmatch(re_version, line):
@@ -138,17 +154,18 @@ def parse_file(
             elif m := re.fullmatch(re_import, line):
                 if not version:
                     raise SyntaxError("must have exactly 1 version line")
-                other_version, other_structs = get_include(m.group("file"))
+                other_version, other_typs = get_include(m.group("file"))
                 for symname in m.group("syms").split(sep=","):
                     symname = symname.strip()
-                    for struct in other_structs:
-                        if struct.name == symname or symname == "*":
-                            if struct.msgid:
-                                struct.msgver.add(version)
-                            for member in struct.members:
-                                if other_version in member.ver:
-                                    member.ver.add(version)
-                            env[struct.name] = struct
+                    for typ in other_typs:
+                        if typ.name == symname or symname == "*":
+                            if isinstance(typ, Struct):
+                                if typ.msgid:
+                                    typ.msgver.add(version)
+                                for member in typ.members:
+                                    if other_version in member.ver:
+                                        member.ver.add(version)
+                            env[typ.name] = typ
             elif m := re.fullmatch(re_structspec, line):
                 if not version:
                     raise SyntaxError("must have exactly 1 version line")
@@ -172,7 +189,7 @@ def parse_file(
                         _struct = env[m.group("name")]
                         if not isinstance(_struct, Struct):
                             raise NameError(
-                                f"Type {repr(m.group('name'))} is not a struct"
+                                f"Type {repr(_struct.name)} is not a struct"
                             )
                         struct = _struct
                         struct.members = parse_members(
@@ -180,27 +197,82 @@ def parse_file(
                         )
                         prev = struct
             elif m := re.fullmatch(re_structspec_cont, line):
-                if not prev:
-                    raise SyntaxError("continuation line must come after a struct line")
+                if not isinstance(prev, Struct):
+                    raise SyntaxError(
+                        "struct-continuation line must come after a struct line"
+                    )
                 assert version
                 prev.members = parse_members(
                     version, env, prev.members, m.group("members")
                 )
+            elif m := re.fullmatch(re_bitfieldspec, line):
+                bf = Bitfield()
+                bf.name = m.group("name")
+                bf.bits = int(m.group("size")) * [""]
+                bf.aliases = {}
+                env[bf.name] = bf
+                prev = bf
+            elif m := re.fullmatch(re_bitfieldspec_bit, line):
+                if m.group("bitfield"):
+                    if m.group("bitfield") not in env:
+                        raise NameError(f"Unknown bitfield {repr(m.group('bitfield'))}")
+                    _bf = env[m.group("bitfield")]
+                    if not isinstance(_bf, Bitfield):
+                        raise NameError(f"Type {repr(_bf.name)} is not a bitfield")
+                    bf = _bf
+                else:
+                    if not isinstance(prev, Bitfield):
+                        raise SyntaxError(
+                            "bitfield-continuation line must come after a bitfield line"
+                        )
+                    bf = prev
+                bit = int(m.group("bit"))
+                name = m.group("name")
+                if bit < 0 or bit >= len(bf.bits):
+                    raise ValueError(f"{bf.name}: bit {bit} is out-of-bounds")
+                if bf.bits[bit]:
+                    raise ValueError(f"{bf.name}: bit {bit} already assigned")
+                if name in bf.aliases:
+                    raise ValueError(f"{bf.name}: name {name} already assigned")
+                bf.bits[bit] = name
+                bf.aliases[name] = ""
+            elif m := re.fullmatch(re_bitfieldspec_alias, line):
+                if m.group("bitfield"):
+                    if m.group("bitfield") not in env:
+                        raise NameError(f"Unknown bitfield {repr(m.group('bitfield'))}")
+                    _bf = env[m.group("bitfield")]
+                    if not isinstance(_bf, Bitfield):
+                        raise NameError(f"Type {repr(_bf.name)} is not a bitfield")
+                    bf = _bf
+                else:
+                    if not isinstance(prev, Bitfield):
+                        raise SyntaxError(
+                            "bitfield-continuation line must come after a bitfield line"
+                        )
+                    bf = prev
+                name = m.group("name")
+                val = m.group("val")
+                if name in bf.aliases:
+                    raise ValueError(f"{bf.name}: name {name} already assigned")
+                bf.aliases[name] = val
             else:
                 raise SyntaxError(f"invalid line {repr(line)}")
     if not version:
         raise SyntaxError("must have exactly 1 version line")
-    structs = [x for x in env.values() if isinstance(x, Struct)]
-    return version, structs
+    typs = [x for x in env.values() if not isinstance(x, Atom)]
+
+    return version, typs
 
 
 # Generate C ###################################################################
 
 
-def c_typename(idprefix: str, typ: Atom | Struct) -> str:
+def c_typename(idprefix: str, typ: Atom | Bitfield | Struct) -> str:
     match typ:
         case Atom():
             return f"uint{typ.value*8}_t"
+        case Bitfield():
+            return f"{idprefix}{typ.name}_t"
         case Struct():
             if typ.msgid is not None:
                 return f"struct {idprefix}msg_{typ.name}"
@@ -227,7 +299,25 @@ def c_vercond(idprefix: str, versions: set[str]) -> str:
     )
 
 
-def gen_h(idprefix: str, versions: set[str], structs: list[Struct]) -> str:
+def just_structs_all(typs: list[Bitfield | Struct]) -> Sequence[Struct]:
+    return list(typ for typ in typs if isinstance(typ, Struct))
+
+
+def just_structs_nonmsg(typs: list[Bitfield | Struct]) -> Sequence[Struct]:
+    return list(typ for typ in typs if isinstance(typ, Struct) and typ.msgid is None)
+
+
+def just_structs_msg(typs: list[Bitfield | Struct]) -> Sequence[Struct]:
+    return list(
+        typ for typ in typs if isinstance(typ, Struct) and typ.msgid is not None
+    )
+
+
+def just_bitfields(typs: list[Bitfield | Struct]) -> Sequence[Bitfield]:
+    return list(typ for typ in typs if isinstance(typ, Bitfield))
+
+
+def gen_h(idprefix: str, versions: set[str], typs: list[Bitfield | Struct]) -> str:
     guard = "_LIB9P__TYPES_H_"
     ret = f"""/* Generated by `{' '.join(sys.argv)}`.  DO NOT EDIT!  */
 
@@ -255,10 +345,7 @@ enum {idprefix}version {{
     ret += """
 /* non-message structs ********************************************************/
 """
-    for struct in structs:
-        if struct.msgid is not None:
-            continue
-
+    for struct in just_structs_nonmsg(typs):
         all_the_same = len(struct.members) == 0 or all(
             m.ver == struct.members[0].ver for m in struct.members
         )
@@ -280,10 +367,8 @@ enum {idprefix}version {{
 
 """
     ret += f"enum {idprefix}msg_type {{ /* uint8_t */\n"
-    namewidth = max(len(msg.name) for msg in structs if msg.msgid is not None)
-    for msg in structs:
-        if msg.msgid is None:
-            continue
+    namewidth = max(len(msg.name) for msg in just_structs_msg(typs))
+    for msg in just_structs_msg(typs):
         ret += f"\t{idprefix.upper()}TYP_{msg.name.ljust(namewidth)} = {msg.msgid},"
         if comment := c_vercomment(msg.msgver):
             ret += " " + comment
@@ -292,10 +377,7 @@ enum {idprefix}version {{
     ret += "\n"
     ret += f"const char *{idprefix}msg_type_str(enum {idprefix}msg_type);\n"
 
-    for msg in structs:
-        if msg.msgid is None:
-            continue
-
+    for msg in just_structs_msg(typs):
         ret += "\n"
         if comment := c_vercomment(msg.msgver):
             ret += comment + "\n"
@@ -324,7 +406,7 @@ enum {idprefix}version {{
     return ret
 
 
-def gen_c(idprefix: str, versions: set[str], structs: list[Struct]) -> str:
+def gen_c(idprefix: str, versions: set[str], typs: list[Bitfield | Struct]) -> str:
     ret = f"""/* Generated by `{' '.join(sys.argv)}`.  DO NOT EDIT!  */
 
 #include <assert.h>
@@ -362,9 +444,9 @@ const char *{idprefix}version_str(enum {idprefix}version ver) {{
 static const char *msg_type_strs[0x100] = {{
 """
     id2name: dict[int, str] = {}
-    for msg in structs:
-        if msg.msgid is not None:
-            id2name[msg.msgid] = msg.name
+    for msg in just_structs_msg(typs):
+        assert msg.msgid
+        id2name[msg.msgid] = msg.name
     for n in range(0, 0x100):
         ret += '\t[0x{:02X}] = "{}",\n'.format(n, id2name.get(n, "0x{:02X}".format(n)))
     ret += "};\n"
@@ -412,7 +494,7 @@ static ALWAYS_INLINE bool _checksize_list(struct _checksize_ctx *ctx,
 #define checksize_4(ctx) _checksize_net(ctx, 4)
 #define checksize_8(ctx) _checksize_net(ctx, 8)
 """
-    for struct in structs:
+    for struct in just_structs_all(typs):
         inline = " ALWAYS_INLINE" if struct.msgid is None else " FLATTEN"
         argfn = used if struct.members else unused
         ret += "\n"
@@ -495,7 +577,7 @@ static ALWAYS_INLINE void unmarshal_8(struct _unmarshal_ctx *ctx, uint64_t *out)
 	ctx->net_offset += 8;
 }
 """
-    for struct in structs:
+    for struct in just_structs_all(typs):
         inline = " ALWAYS_INLINE" if struct.msgid is None else " FLATTEN"
         argfn = used if struct.members else unused
         ret += "\n"
@@ -567,7 +649,7 @@ static ALWAYS_INLINE bool marshal_8(struct _marshal_ctx *ctx, uint64_t *val) {
 	return false;
 }
 """
-    for struct in structs:
+    for struct in just_structs_all(typs):
         inline = " ALWAYS_INLINE" if struct.msgid is None else " FLATTEN"
         argfn = used if struct.members else unused
         ret += "\n"
@@ -616,14 +698,14 @@ struct _vtable_version _{idprefix}vtables[{c_verenum(idprefix, 'NUM')}] = {{
 """
 
     ret += f"\t[{c_verenum(idprefix, 'unknown')}] = {{ .msgs = {{\n"
-    for msg in structs:
+    for msg in just_structs_msg(typs):
         if msg.name in ["Tversion", "Rversion", "Rerror"]:  # SPECIAL
             ret += f"\t\t_MSG({msg.name}),\n"
     ret += "\t}},\n"
 
     for ver in sorted(versions):
         ret += f"\t[{c_verenum(idprefix, ver)}] = {{ .msgs = {{\n"
-        for msg in structs:
+        for msg in just_structs_msg(typs):
             if ver not in msg.msgver:
                 continue
             ret += f"\t\t_MSG({msg.name}),\n"
@@ -638,32 +720,32 @@ struct _vtable_version _{idprefix}vtables[{c_verenum(idprefix, 'NUM')}] = {{
 
 
 class Parser:
-    cache: dict[str, tuple[str, list[Struct]]] = {}
+    cache: dict[str, tuple[str, list[Bitfield | Struct]]] = {}
 
-    def parse_file(self, filename: str) -> tuple[str, list[Struct]]:
+    def parse_file(self, filename: str) -> tuple[str, list[Bitfield | Struct]]:
         filename = os.path.normpath(filename)
         if filename not in self.cache:
 
-            def get_include(other_filename: str) -> tuple[str, list[Struct]]:
+            def get_include(other_filename: str) -> tuple[str, list[Bitfield | Struct]]:
                 return self.parse_file(os.path.join(filename, "..", other_filename))
 
             self.cache[filename] = parse_file(filename, get_include)
         return self.cache[filename]
 
-    def all(self) -> tuple[set[str], list[Struct]]:
+    def all(self) -> tuple[set[str], list[Bitfield | Struct]]:
         ret_versions: set[str] = set()
-        ret_structs: dict[str, Struct] = {}
-        for version, structs in self.cache.values():
+        ret_typs: dict[str, Bitfield | Struct] = {}
+        for version, typs in self.cache.values():
             if version in ret_versions:
                 raise ValueError(f"duplicate protocol version {repr(version)}")
             ret_versions.add(version)
-            for struct in structs:
-                if struct.name in ret_structs:
-                    if struct != ret_structs[struct.name]:
-                        raise ValueError(f"duplicate struct name {repr(struct.name)}")
+            for typ in typs:
+                if typ.name in ret_typs:
+                    if typ != ret_typs[typ.name]:
+                        raise ValueError(f"duplicate type name {repr(typ.name)}")
                 else:
-                    ret_structs[struct.name] = struct
-        return ret_versions, list(ret_structs.values())
+                    ret_typs[typ.name] = typ
+        return ret_versions, list(ret_typs.values())
 
 
 if __name__ == "__main__":
@@ -674,9 +756,9 @@ if __name__ == "__main__":
     parser = Parser()
     for txtname in sys.argv[1:]:
         parser.parse_file(txtname)
-    versions, structs = parser.all()
+    versions, typs = parser.all()
     outdir = os.path.normpath(os.path.join(sys.argv[0], ".."))
     with open(os.path.join(outdir, "include/lib9p/_types.h"), "w") as fh:
-        fh.write(gen_h("lib9p_", versions, structs))
+        fh.write(gen_h("lib9p_", versions, typs))
     with open(os.path.join(outdir, "types.c"), "w") as fh:
-        fh.write(gen_c("lib9p_", versions, structs))
+        fh.write(gen_c("lib9p_", versions, typs))