Просмотр исходного кода

Rewrote explode_asserts.py to be more efficient

Normally I wouldn't consider optimizing this sort of script, but
explode_asserts.py proved to be terribly inefficient and dominated
the build time for running tests. It was slow enough to be distracting
when attempting to test patches while debugging. Just running
explode_asserts.py was ~10x slower than the rest of the compilation
process.

After implementing a proper tokenizer and switching to a handwritten
recursive descent parser, I was able to speed up explode_asserts.py
by ~5x and make test compilation much more tolerable.

I don't think this was a limitaiton of parsy, but rather switching
to a recursive descent parser made it much easier to find the hotspots
where parsing was wasting cycles (string slicing for one).

It's interesting to note that while the assert patterns can be parsed
with a LL(1) parser (by dumping seen tokens if a pattern fails),
I didn't bother as it's much easier to write the patterns with LL(k)
and parsing asserts is predicated by the "assert" string.

A few other tweaks:
- allowed combining different test modes in one run
- added a --no-internal option
- changed test_.py to start counting cases from 1
- added assert(memcmp(a, b) == 0) matching
- added better handling of string escapes in assert messages

time to run tests:
before: 1m31.122s
after:  0m41.447s
Christopher Haster 5 лет назад
Родитель
Сommit
b9d0695e0a
2 измененных файлов с 404 добавлено и 237 удалено
  1. 352 184
      scripts/explode_asserts.py
  2. 52 53
      scripts/test_.py

+ 352 - 184
scripts/explode_asserts.py

@@ -1,208 +1,372 @@
 #!/usr/bin/env python3
 
-import parsy as p
 import re
-import io
 import sys
 
-ASSERT_PATTERN = p.string('LFS_ASSERT') | p.string('assert')
-ASSERT_CHARS = 'La'
-ASSERT_TARGET = '__LFS_ASSERT_{TYPE}_{COMP}'
-ASSERT_TESTS = {
-    'int': """
-        __typeof__({lh}) _lh = {lh};
-        __typeof__({lh}) _rh = (__typeof__({lh})){rh};
-        if (!(_lh {op} _rh)) {{
-            printf("%s:%d:assert: "
-                "assert failed with %"PRIiMAX", expected {comp} %"PRIiMAX"\\n",
-                {file}, {line}, (intmax_t)_lh, (intmax_t)_rh);
-            fflush(NULL);
-            raise(SIGABRT);
+PATTERN = ['LFS_ASSERT', 'assert']
+PREFIX = 'LFS'
+MAXWIDTH = 16
+
+ASSERT = "__{PREFIX}_ASSERT_{TYPE}_{COMP}"
+FAIL = """
+__attribute__((unused))
+static void __{prefix}_assert_fail_{type}(
+        const char *file, int line, const char *comp,
+        {ctype} lh, size_t lsize,
+        {ctype} rh, size_t rsize) {{
+    printf("%s:%d:assert: assert failed with ", file, line);
+    __{prefix}_assert_print_{type}(lh, lsize);
+    printf(", expected %s ", comp);
+    __{prefix}_assert_print_{type}(rh, rsize);
+    printf("\\n");
+    fflush(NULL);
+    raise(SIGABRT);
+}}
+"""
+
+COMP = {
+    '==': 'eq',
+    '!=': 'ne',
+    '<=': 'le',
+    '>=': 'ge',
+    '<':  'lt',
+    '>':  'gt',
+}
+
+TYPE = {
+    'int': {
+        'ctype': 'intmax_t',
+        'fail': FAIL,
+        'print': """
+        __attribute__((unused))
+        static void __{prefix}_assert_print_{type}({ctype} v, size_t size) {{
+            (void)size;
+            printf("%"PRIiMAX, v);
+        }}
+        """,
+        'assert': """
+        #define __{PREFIX}_ASSERT_{TYPE}_{COMP}(file, line, lh, rh)
+        do {{
+            __typeof__(lh) _lh = lh;
+            __typeof__(lh) _rh = (__typeof__(lh))rh;
+            if (!(_lh {op} _rh)) {{
+                __{prefix}_assert_fail_{type}(file, line, "{comp}",
+                        (intmax_t)_lh, 0, (intmax_t)_rh, 0);
+            }}
+        }} while (0)
+        """
+    },
+    'bool': {
+        'ctype': 'bool',
+        'fail': FAIL,
+        'print': """
+        __attribute__((unused))
+        static void __{prefix}_assert_print_{type}({ctype} v, size_t size) {{
+            (void)size;
+            printf("%s", v ? "true" : "false");
         }}
-    """,
-    'str': """
-        const char *_lh = {lh};
-        const char *_rh = {rh};
-        if (!(strcmp(_lh, _rh) {op} 0)) {{
-            printf("%s:%d:assert: "
-                "assert failed with \\\"%s\\\", expected {comp} \\\"%s\\\"\\n",
-                {file}, {line}, _lh, _rh);
-            fflush(NULL);
-            raise(SIGABRT);
+        """,
+        'assert': """
+        #define __{PREFIX}_ASSERT_{TYPE}_{COMP}(file, line, lh, rh)
+        do {{
+            bool _lh = !!(lh);
+            bool _rh = !!(rh);
+            if (!(_lh {op} _rh)) {{
+                __{prefix}_assert_fail_{type}(file, line, "{comp}",
+                        _lh, 0, _rh, 0);
+            }}
+        }} while (0)
+        """
+    },
+    'mem': {
+        'ctype': 'const void *',
+        'fail': FAIL,
+        'print': """
+        __attribute__((unused))
+        static void __{prefix}_assert_print_{type}({ctype} v, size_t size) {{
+            const uint8_t *s = v;
+            printf("\\\"");
+            for (size_t i = 0; i < size && i < {maxwidth}; i++) {{
+                if (s[i] >= ' ' && s[i] <= '~') {{
+                    printf("%c", s[i]);
+                }} else {{
+                    printf("\\\\x%02x", s[i]);
+                }}
+            }}
+            if (size > {maxwidth}) {{
+                printf("...");
+            }}
+            printf("\\\"");
         }}
-    """,
-    'bool': """
-        bool _lh = !!({lh});
-        bool _rh = !!({rh});
-        if (!(_lh {op} _rh)) {{
-            printf("%s:%d:assert: "
-                "assert failed with %s, expected {comp} %s\\n",
-                {file}, {line}, _lh ? "true" : "false", _rh ? "true" : "false");
-            fflush(NULL);
-            raise(SIGABRT);
+        """,
+        'assert': """
+        #define __{PREFIX}_ASSERT_{TYPE}_{COMP}(file, line, lh, rh, size)
+        do {{
+            const void *_lh = lh;
+            const void *_rh = rh;
+            if (!(memcmp(_lh, _rh, size) {op} 0)) {{
+                __{prefix}_assert_fail_{type}(file, line, "{comp}",
+                        _lh, size, _rh, size);
+            }}
+        }} while (0)
+        """
+    },
+    'str': {
+        'ctype': 'const char *',
+        'fail': FAIL,
+        'print': """
+        __attribute__((unused))
+        static void __{prefix}_assert_print_{type}({ctype} v, size_t size) {{
+            __{prefix}_assert_print_mem(v, size);
         }}
-    """,
+        """,
+        'assert': """
+        #define __{PREFIX}_ASSERT_{TYPE}_{COMP}(file, line, lh, rh)
+        do {{
+            const char *_lh = lh;
+            const char *_rh = rh;
+            if (!(strcmp(_lh, _rh) {op} 0)) {{
+                __{prefix}_assert_fail_{type}(file, line, "{comp}",
+                        _lh, strlen(_lh), _rh, strlen(_rh));
+            }}
+        }} while (0)
+        """
+    }
 }
 
-def mkassert(lh, rh='true', type='bool', comp='eq'):
-    return ((ASSERT_TARGET + "({lh}, {rh}, __FILE__, __LINE__, __func__)")
-        .format(
-            type=type, TYPE=type.upper(),
-            comp=comp, COMP=comp.upper(),
-            lh=lh.strip(' '),
-            rh=rh.strip(' ')))
-
-def mkdecl(type, comp, op):
-    return ((
-        "#define "+ASSERT_TARGET+"(lh, rh, file, line, func)"
-        "   do {{"+re.sub('\s+', ' ', ASSERT_TESTS[type])+"}} while (0)\n")
-        .format(
-            type=type, TYPE=type.upper(),
-            comp=comp, COMP=comp.upper(),
-            lh='lh', rh='rh', op=op,
-            file='file', line='line', func='func'))
-
-# add custom until combinator
-def until(self, end):
-    return end.should_fail('should fail').then(self).many()
-p.Parser.until = until
-
-pcomp = (
-    p.string('==').tag('eq') |
-    p.string('!=').tag('ne') |
-    p.string('<=').tag('le') |
-    p.string('>=').tag('ge') |
-    p.string('<').tag('lt') |
-    p.string('>').tag('gt'));
-
-plogic = p.string('&&') | p.string('||')
-
-@p.generate
-def pstrassert():
-    yield ASSERT_PATTERN + p.regex('\s*') + p.string('(') + p.regex('\s*')
-    yield p.string('strcmp') + p.regex('\s*') + p.string('(') + p.regex('\s*')
-    lh = yield pexpr.until(p.string(',') | p.string(')') | plogic)
-    yield p.string(',') + p.regex('\s*')
-    rh = yield pexpr.until(p.string(')') | plogic)
-    yield p.string(')') + p.regex('\s*')
-    op = yield pcomp
-    yield p.regex('\s*') + p.string('0') + p.regex('\s*') + p.string(')')
-    return mkassert(''.join(lh), ''.join(rh), 'str', op[0])
-
-@p.generate
-def pintassert():
-    yield ASSERT_PATTERN + p.regex('\s*') + p.string('(') + p.regex('\s*')
-    lh = yield pexpr.until(pcomp | p.string(')') | plogic)
-    op = yield pcomp
-    rh = yield pexpr.until(p.string(')') | plogic)
-    yield p.string(')')
-    return mkassert(''.join(lh), ''.join(rh), 'int', op[0])
-
-@p.generate
-def pboolassert():
-    yield ASSERT_PATTERN + p.regex('\s*') + p.string('(') + p.regex('\s*')
-    expr = yield pexpr.until(p.string(')'))
-    yield p.string(')')
-    return mkassert(''.join(expr), 'true', 'bool', 'eq')
-
-passert = p.peek(ASSERT_PATTERN) >> (pstrassert | pintassert | pboolassert)
-
-@p.generate
-def pcomment1():
-    yield p.string('//')
-    s = yield p.regex('[^\\n]*')
-    yield p.string('\n')
-    return '//' + s + '\n'
-
-@p.generate
-def pcomment2():
-    yield p.string('/*')
-    s = yield p.regex('((?!\*/).)*')
-    yield p.string('*/')
-    return '/*' + ''.join(s) + '*/'
-
-@p.generate
-def pcomment3():
-    yield p.string('#')
-    s = yield p.regex('[^\\n]*')
-    yield p.string('\n')
-    return '#' + s + '\n'
-
-pws = p.regex('\s+') | pcomment1 | pcomment2 | pcomment3
-
-@p.generate
-def pstring():
-    q = yield p.regex('["\']')
-    s = yield (p.string('\\%s' % q) | p.regex('[^%s]' % q)).many()
-    yield p.string(q)
-    return q + ''.join(s) + q
-
-@p.generate
-def pnested():
-    l = yield p.string('(')
-    n = yield pexpr.until(p.string(')'))
-    r = yield p.string(')')
-    return l + ''.join(n) + r
-
-pexpr = (
-    # shortcut for a bit better performance
-    p.regex('[^%s/#\'"():;{}=><,&|-]+' % ASSERT_CHARS) |
-    pws |
-    passert |
-    pstring |
-    pnested |
-    p.string('->') |
-    p.regex('.', re.DOTALL))
-
-@p.generate
-def pstmt():
-    ws = yield pws.many()
-    lh = yield pexpr.until(p.string('=>') | p.regex('[:;{}]'))
-    op = yield p.string('=>').optional()
-    if op == '=>':
-        rh = yield pstmt
-        return ''.join(ws) + mkassert(''.join(lh), rh, 'int', 'eq')
+def mkdecls(outf, maxwidth=16):
+    outf.write("#include <stdio.h>\n")
+    outf.write("#include <stdbool.h>\n")
+    outf.write("#include <stdint.h>\n")
+    outf.write("#include <inttypes.h>\n")
+    outf.write("#include <signal.h>\n")
+
+    for type, desc in sorted(TYPE.items()):
+        format = {
+            'type': type.lower(), 'TYPE': type.upper(),
+            'ctype': desc['ctype'],
+            'prefix': PREFIX.lower(), 'PREFIX': PREFIX.upper(),
+            'maxwidth': maxwidth,
+        }
+        outf.write(re.sub('\s+', ' ',
+            desc['print'].strip().format(**format))+'\n')
+        outf.write(re.sub('\s+', ' ',
+            desc['fail'].strip().format(**format))+'\n')
+
+        for op, comp in sorted(COMP.items()):
+            format.update({
+                'comp': comp.lower(), 'COMP': comp.upper(),
+                'op': op,
+            })
+            outf.write(re.sub('\s+', ' ',
+                desc['assert'].strip().format(**format))+'\n')
+
+def mkassert(type, comp, lh, rh, size=None):
+    format = {
+        'type': type.lower(), 'TYPE': type.upper(),
+        'comp': comp.lower(), 'COMP': comp.upper(),
+        'prefix': PREFIX.lower(), 'PREFIX': PREFIX.upper(),
+        'lh': lh.strip(),
+        'rh': rh.strip(),
+        'size': size,
+    }
+    if size:
+        return ((ASSERT + '(__FILE__, __LINE__, {lh}, {rh}, {size})')
+            .format(**format))
     else:
-        return ''.join(ws) + ''.join(lh)
+        return ((ASSERT + '(__FILE__, __LINE__, {lh}, {rh})')
+            .format(**format))
+
+
+# simple recursive descent parser
+LEX = {
+    'ws':       [r'(?:\s|\n|#.*?\n|//.*?\n|/\*.*?\*/)+'],
+    'assert':   PATTERN,
+    'string':   [r'"(?:\\.|[^"])*"', r"'(?:\\.|[^'])\'"],
+    'arrow':    ['=>'],
+    'paren':    ['\(', '\)'],
+    'op':       ['strcmp', 'memcmp', '->'],
+    'comp':     ['==', '!=', '<=', '>=', '<', '>'],
+    'logic':    ['\&\&', '\|\|'],
+    'sep':      [':', ';', '\{', '\}', ','],
+}
+
+class ParseFailure(Exception):
+    def __init__(self, expected, found):
+        self.expected = expected
+        self.found = found
+
+    def __str__(self):
+        return "expected %r, found %s..." % (
+            self.expected, repr(self.found)[:70])
+
+class Parse:
+    def __init__(self, inf, lexemes):
+        p = '|'.join('(?P<%s>%s)' % (n, '|'.join(l))
+            for n, l in lexemes.items())
+        p = re.compile(p, re.DOTALL)
+        data = inf.read()
+        tokens = []
+        while True:
+            m = p.search(data)
+            if m:
+                if m.start() > 0:
+                    tokens.append((None, data[:m.start()]))
+                tokens.append((m.lastgroup, m.group()))
+                data = data[m.end():]
+            else:
+                tokens.append((None, data))
+                break
+        self.tokens = tokens
+        self.off = 0
+
+    def lookahead(self, *pattern):
+        if self.off < len(self.tokens):
+            token = self.tokens[self.off]
+            if token[0] in pattern or token[1] in pattern:
+                self.m = token[1]
+                return self.m
+        self.m = None
+        return self.m
+
+    def accept(self, *patterns):
+        m = self.lookahead(*patterns)
+        if m is not None:
+            self.off += 1
+        return m
+
+    def expect(self, *patterns):
+        m = self.accept(*patterns)
+        if not m:
+            raise ParseFailure(patterns, self.tokens[self.off:])
+        return m
+
+    def push(self):
+        return self.off
+
+    def pop(self, state):
+        self.off = state
+
+def passert(p):
+    def pastr(p):
+        p.expect('assert') ; p.accept('ws') ; p.expect('(') ; p.accept('ws')
+        p.expect('strcmp') ; p.accept('ws') ; p.expect('(') ; p.accept('ws')
+        lh = pexpr(p) ; p.accept('ws')
+        p.expect(',') ; p.accept('ws')
+        rh = pexpr(p) ; p.accept('ws')
+        p.expect(')') ; p.accept('ws')
+        comp = p.expect('comp') ; p.accept('ws')
+        p.expect('0') ; p.accept('ws')
+        p.expect(')')
+        return mkassert('str', COMP[comp], lh, rh)
+
+    def pamem(p):
+        p.expect('assert') ; p.accept('ws') ; p.expect('(') ; p.accept('ws')
+        p.expect('memcmp') ; p.accept('ws') ; p.expect('(') ; p.accept('ws')
+        lh = pexpr(p) ; p.accept('ws')
+        p.expect(',') ; p.accept('ws')
+        rh = pexpr(p) ; p.accept('ws')
+        p.expect(',') ; p.accept('ws')
+        size = pexpr(p) ; p.accept('ws')
+        p.expect(')') ; p.accept('ws')
+        comp = p.expect('comp') ; p.accept('ws')
+        p.expect('0') ; p.accept('ws')
+        p.expect(')')
+        return mkassert('mem', COMP[comp], lh, rh, size)
+
+    def paint(p):
+        p.expect('assert') ; p.accept('ws') ; p.expect('(') ; p.accept('ws')
+        lh = pexpr(p) ; p.accept('ws')
+        comp = p.expect('comp') ; p.accept('ws')
+        rh = pexpr(p) ; p.accept('ws')
+        p.expect(')')
+        return mkassert('int', COMP[comp], lh, rh)
+
+    def pabool(p):
+        p.expect('assert') ; p.accept('ws') ; p.expect('(') ; p.accept('ws')
+        lh = pexprs(p) ; p.accept('ws')
+        p.expect(')')
+        return mkassert('bool', 'eq', lh, 'true')
+
+    def pa(p):
+        return p.expect('assert')
+
+    state = p.push()
+    lastf = None
+    for pa in [pastr, pamem, paint, pabool, pa]:
+        try:
+            return pa(p)
+        except ParseFailure as f:
+            p.pop(state)
+            lastf = f
+    else:
+        raise lastf
+
+def pexpr(p):
+    res = []
+    while True:
+        if p.accept('('):
+            res.append(p.m)
+            while True:
+                res.append(pexprs(p))
+                if p.accept('sep'):
+                    res.append(p.m)
+                else:
+                    break
+            res.append(p.expect(')'))
+        elif p.lookahead('assert'):
+            res.append(passert(p))
+        elif p.accept('assert', 'ws', 'string', 'op', None):
+            res.append(p.m)
+        else:
+            return ''.join(res)
+
+def pexprs(p):
+    res = []
+    while True:
+        res.append(pexpr(p))
+        if p.accept('comp', 'logic', ','):
+            res.append(p.m)
+        else:
+            return ''.join(res)
+
+def pstmt(p):
+    ws = p.accept('ws') or ''
+    lh = pexprs(p)
+    if p.accept('=>'):
+        rh = pexprs(p)
+        return ws + mkassert('int', 'eq', lh, rh)
+    else:
+        return ws + lh
 
-@p.generate
-def pstmts():
-    a = yield pstmt
-    b = yield (p.regex('[:;{}]') + pstmt).many()
-    return [a] + b
 
 def main(args):
     inf = open(args.input, 'r') if args.input else sys.stdin
     outf = open(args.output, 'w') if args.output else sys.stdout
 
-    # parse C code
-    input = inf.read()
-    stmts = pstmts.parse(input)
+    lexemes = LEX.copy()
+    if args.pattern:
+        lexemes['assert'] = args.pattern
+    p = Parse(inf, lexemes)
 
     # write extra verbose asserts
-    outf.write("#include <stdbool.h>\n")
-    outf.write("#include <stdint.h>\n")
-    outf.write("#include <inttypes.h>\n")
-    outf.write("#include <signal.h>\n")
-    outf.write(mkdecl('int',  'eq', '=='))
-    outf.write(mkdecl('int',  'ne', '!='))
-    outf.write(mkdecl('int',  'lt', '<'))
-    outf.write(mkdecl('int',  'gt', '>'))
-    outf.write(mkdecl('int',  'le', '<='))
-    outf.write(mkdecl('int',  'ge', '>='))
-    outf.write(mkdecl('str',  'eq', '=='))
-    outf.write(mkdecl('str',  'ne', '!='))
-    outf.write(mkdecl('str',  'lt', '<'))
-    outf.write(mkdecl('str',  'gt', '>'))
-    outf.write(mkdecl('str',  'le', '<='))
-    outf.write(mkdecl('str',  'ge', '>='))
-    outf.write(mkdecl('bool', 'eq', '=='))
+    mkdecls(outf, maxwidth=args.maxwidth)
     if args.input:
         outf.write("#line %d \"%s\"\n" % (1, args.input))
 
-    # write parsed statements
-    for stmt in stmts:
-        outf.write(stmt)
+    # parse and write out stmt at a time
+    try:
+        while True:
+            outf.write(pstmt(p))
+            if p.accept('sep'):
+                outf.write(p.m)
+            else:
+                break
+    except ParseFailure as f:
+        pass
+
+    for i in range(p.off, len(p.tokens)):
+        outf.write(p.tokens[i][1])
 
 if __name__ == "__main__":
     import argparse
@@ -210,6 +374,10 @@ if __name__ == "__main__":
         description="Cpp step that increases assert verbosity")
     parser.add_argument('input', nargs='?',
         help="Input C file after cpp.")
-    parser.add_argument('-o', '--output',
+    parser.add_argument('-o', '--output', required=True,
         help="Output C file.")
+    parser.add_argument('-p', '--pattern', action='append',
+        help="Patterns to search for starting an assert statement.")
+    parser.add_argument('--maxwidth', default=MAXWIDTH, type=int,
+        help="Maximum number of characters to display for strcmp and memcmp.")
     main(parser.parse_args())

+ 52 - 53
scripts/test_.py

@@ -4,24 +4,6 @@
 # .toml files stored in the tests directory.
 #
 
-# TODO
-# x nargs > 1?
-# x show perm config on failure
-# x filtering
-# n show perm config on verbose?
-# x better lineno tracking for cases?
-# n non-int perms?
-# x different path format?
-# - suite.prologue, suite.epilogue
-# x in
-# x change BLOCK_CYCLES to -1 by default
-# x change persist behaviour
-# x config chaining correct
-# - why can't gdb see my defines?
-# - say no to internal?
-# x buffering stdout issues?
-# - debug fast
-
 import toml
 import glob
 import re
@@ -125,6 +107,7 @@ class TestFailure(Exception):
 class TestCase:
     def __init__(self, config, filter=filter,
             suite=None, caseno=None, lineno=None, **_):
+        self.config = config
         self.filter = filter
         self.suite = suite
         self.caseno = caseno
@@ -150,8 +133,10 @@ class TestCase:
             return '%s#%d' % (
                 self.suite.name, self.caseno)
 
-    def permute(self, defines, permno=None, **_):
-        ncase = copy.copy(self)
+    def permute(self, class_=None, defines={}, permno=None, **_):
+        ncase = (class_ or type(self))(self.config)
+        for k, v in self.__dict__.items():
+            setattr(ncase, k, v)
         ncase.case = self
         ncase.perms = [ncase]
         ncase.permno = permno
@@ -194,6 +179,8 @@ class TestCase:
                 len(self.filter) >= 2 and
                 self.filter[1] != self.permno):
             return False
+        elif args.get('no_internal', False) and self.in_ is not None:
+            return False
         elif self.if_ is not None:
             return eval(self.if_, None, self.defines.copy())
         else:
@@ -210,6 +197,8 @@ class TestCase:
             if persist != 'noerase':
                 try:
                     os.remove(self.suite.path + '.disk')
+                    if args.get('verbose', False):
+                        print('rm', self.suite.path + '.disk')
                 except FileNotFoundError:
                     pass
 
@@ -225,7 +214,7 @@ class TestCase:
             if gdb == 'assert':
                 ncmd.extend(['-ex', 'r'])
                 if failure.assert_:
-                    ncmd.extend(['-ex', 'up'])
+                    ncmd.extend(['-ex', 'up 2'])
             elif gdb == 'start':
                 ncmd.extend([
                     '-ex', 'b %s:%d' % (self.suite.path, self.code_lineno),
@@ -331,13 +320,15 @@ class ReentrantTestCase(TestCase):
                     raise
 
 class TestSuite:
-    def __init__(self, path, filter=None, TestCase=TestCase, **args):
+    def __init__(self, path, classes=[TestCase], defines={},
+            filter=None, **args):
         self.name = os.path.basename(path)
         if self.name.endswith('.toml'):
             self.name = self.name[:-len('.toml')]
         self.path = path
+        self.classes = classes
+        self.defines = defines
         self.filter = filter
-        self.TestCase = TestCase
 
         with open(path) as f:
             # load tests
@@ -356,7 +347,9 @@ class TestSuite:
             code_linenos.reverse()
 
         # grab global config
-        self.defines = config.get('define', {})
+        for k, v in config.get('define', {}).items():
+            if k not in self.defines:
+                self.defines[k] = v
         self.code = config.get('code', None)
         if self.code is not None:
             self.code_lineno = code_linenos.pop()
@@ -372,8 +365,8 @@ class TestSuite:
                 if k not in case:
                     case[k] = v
             # initialize test case
-            self.cases.append(self.TestCase(case, filter=filter,
-                suite=self, caseno=i, lineno=lineno, **args))
+            self.cases.append(TestCase(case, filter=filter,
+                suite=self, caseno=i+1, lineno=lineno, **args))
 
     def __str__(self):
         return self.name
@@ -381,14 +374,14 @@ class TestSuite:
     def __lt__(self, other):
         return self.name < other.name
 
-    def permute(self, defines={}, **args):
+    def permute(self, **args):
         for case in self.cases:
             # lets find all parameterized definitions, in one of [args.D,
             # suite.defines, case.defines, DEFINES]. Note that each of these
             # can be either a dict of defines, or a list of dicts, expressing
             # an initial set of permutations.
             pending = [{}]
-            for inits in [defines, self.defines, case.defines, DEFINES]:
+            for inits in [self.defines, case.defines, DEFINES]:
                 if not isinstance(inits, list):
                     inits = [inits]
 
@@ -422,8 +415,10 @@ class TestSuite:
 
             # generate permutations
             case.perms = []
-            for i, perm in enumerate(expanded):
-                case.perms.append(case.permute(perm, permno=i, **args))
+            for i, (class_, defines) in enumerate(
+                    it.product(self.classes, expanded)):
+                case.perms.append(case.permute(
+                    class_, defines, permno=i+1, **args))
 
             # also track non-unique defines
             case.defines = {}
@@ -519,16 +514,12 @@ class TestSuite:
         self.target = self.path + '.test'
         return self.makefile, self.target
 
-    def test(self, caseno=None, permno=None, **args):
+    def test(self, **args):
         # run test suite!
         if not args.get('verbose', True):
             sys.stdout.write(self.name + ' ')
             sys.stdout.flush()
         for perm in self.perms:
-            if caseno is not None and perm.caseno != caseno:
-                continue
-            if permno is not None and perm.permno != permno:
-                continue
             if not perm.shouldtest(**args):
                 continue
 
@@ -553,6 +544,23 @@ class TestSuite:
             sys.stdout.write('\n')
 
 def main(**args):
+    # figure out explicit defines
+    defines = {}
+    for define in args['D']:
+        k, v, *_ = define.split('=', 2) + ['']
+        defines[k] = v
+
+    # and what class of TestCase to run
+    classes = []
+    if args.get('normal', False):
+        classes.append(TestCase)
+    if args.get('reentrant', False):
+        classes.append(ReentrantTestCase)
+    if args.get('valgrind', False):
+        classes.append(ValgrindTestCase)
+    if not classes:
+        classes = [TestCase]
+
     suites = []
     for testpath in args['testpaths']:
         # optionally specified test case/perm
@@ -571,27 +579,14 @@ def main(**args):
 
         # find tests
         for path in glob.glob(testpath):
-            if args.get('valgrind', False):
-                TestCase_ = ValgrindTestCase
-            elif args.get('reentrant', False):
-                TestCase_ = ReentrantTestCase
-            else:
-                TestCase_ = TestCase
-
-            suites.append(TestSuite(path,
-                filter=filter, TestCase=TestCase_, **args))
+            suites.append(TestSuite(path, classes, defines, filter, **args))
 
     # sort for reproducability
     suites = sorted(suites)
 
     # generate permutations
-    defines = {}
-    for define in args['D']:
-        k, v, *_ = define.split('=', 2) + ['']
-        defines[k] = v
-
     for suite in suites:
-        suite.permute(defines, **args)
+        suite.permute(**args)
 
     # build tests in parallel
     print('====== building ======')
@@ -736,10 +731,14 @@ if __name__ == "__main__":
     parser.add_argument('-g', '--gdb', choices=['init', 'start', 'assert'],
         nargs='?', const='assert',
         help="Drop into gdb on test failure.")
-    parser.add_argument('--valgrind', action='store_true',
-        help="Run non-leaky tests under valgrind to check for memory leaks.")
-    parser.add_argument('--reentrant', action='store_true',
+    parser.add_argument('--no-internal', action='store_true',
+        help="Don't run tests that require internal knowledge.")
+    parser.add_argument('-n', '--normal', action='store_true',
+        help="Run tests normally.")
+    parser.add_argument('-r', '--reentrant', action='store_true',
         help="Run reentrant tests with simulated power-loss.")
+    parser.add_argument('-V', '--valgrind', action='store_true',
+        help="Run non-leaky tests under valgrind to check for memory leaks.")
     parser.add_argument('-e', '--exec', default=[], type=lambda e: e.split(' '),
         help="Run tests with another executable prefixed on the command line.")
     sys.exit(main(**vars(parser.parse_args())))