1 files changed, 125 insertions, 0 deletions
diff --git a/research/experiments/044-specs/http_requests.py b/research/experiments/044-specs/http_requests.py
new file mode 100644
index 0000000..20b0fb9
--- /dev/null
+++ b/research/experiments/044-specs/http_requests.py
@@ -0,0 +1,125 @@
+# HTTP/1.1 requests (RFC 9112) read a packet at a time in plain Python: the reference 140 counts aup's lines against.
+# Each packet's bytes are added to what the last ones left, and every request they complete is read: its line, its
+# fields (names lowercased), its body by Content-Length or chunked, a chunked one's trailers. The grammar is h11's.
+TOKEN = set(b"!#$%&'*+-.^_`|~0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ")
+HEX = set(b'0123456789abcdefABCDEF')
+SPACE = set(b'\0\t\n\x0b\x0c\r ')  # never in a field value's words
+
+
+def line(d, i):
+    """The line from i to the next CRLF, and where the next starts; None, None if it has not come yet."""
+    j = d.find(b'\r\n', i)
+    return (None, None) if j < 0 else (d[i:j], j + 2)
+
+
+def field(text):
+    name, colon, value = text.partition(b':')
+    value = value.strip(b' \t')
+    if not colon or not name or not set(name) <= TOKEN or any(b in SPACE and b not in b' \t' for b in value):
+        raise ValueError('a field')
+    return name.lower(), value
+
+
+def fields(d, i):
+    """The field lines from i to an empty line, and where it ends; None, None if they have not all come."""
+    out = []
+    while True:
+        text, i = line(d, i)
+        if text is None or not text:
+            return (None, None) if text is None else (out, i)
+        out.append(field(text))
+
+
+def request(d):
+    """The request at the start of d and its length; None if d holds only part of one."""
+    text, i = line(d, 0)
+    if text is None:
+        return None
+    parts = text.split(b' ')
+    if len(parts) != 3 or not parts[0] or not set(parts[0]) <= TOKEN or not parts[1] or \
+            any(b < 0x21 or b > 0x7e for b in parts[1]) or parts[2] not in (b'HTTP/1.0', b'HTTP/1.1'):
+        raise ValueError('a request line')
+    method, target, minor = parts[0], parts[1], parts[2][-1] - 0x30
+    found, i = fields(d, i)
+    if found is None:
+        return None
+    named = lambda n: [v for k, v in found if k == n]
+    hosts, coded, length = named(b'host'), named(b'transfer-encoding'), named(b'content-length')
+    if len(hosts) > 1 or (minor == 1 and not hosts):
+        raise ValueError('a host')
+    body, trailers = b'', []
+    if coded:
+        if length or len(coded) > 1 or coded[0].lower() != b'chunked':
+            raise ValueError('a coding')
+        while True:
+            text, i = line(d, i)
+            if text is None:
+                return None
+            size, semi, ext = text.partition(b';')
+            size = size if semi else size.rstrip(b' \t')
+            if not 1 <= len(size) <= 20 or not set(size) <= HEX or b'\n' in ext or int(size, 16) > 2 ** 31:
+                raise ValueError('a chunk size')  # 2^31 at most: ours, not h11's
+            n = int(size, 16)
+            if n == 0:
+                trailers, i = fields(d, i)
+                if trailers is None:
+                    return None
+                break
+            if len(d) < i + n + 2:
+                return None
+            if d[i + n:i + n + 2] != b'\r\n':
+                raise ValueError('a chunk end')
+            body += d[i:i + n]
+            i += n + 2
+    elif length:
+        if len(length) > 1 or not length[0].isdigit() or int(length[0]) > 2 ** 31:
+            raise ValueError('a length')
+        if len(d) < i + int(length[0]):
+            return None
+        body, i = d[i:i + int(length[0])], i + int(length[0])
+    return (method, target, minor, found, body, trailers), i
+
+
+def feed(pending, packet):
+    """The requests the packet completes, and the bytes left for the next."""
+    d, out = pending + packet, []
+    while d:
+        got = request(d)
+        if got is None:
+            break
+        out.append(got[0])
+        d = d[got[1]:]
+    return out, d
+
+
+if __name__ == '__main__':
+    import sys, time
+    import h11
+    with open(sys.argv[1], 'rb') as f:  # http_check.py's whole streams: each a u32 big-endian length, then its bytes
+        given = f.read()
+    streams, at = [], 0
+    while at < len(given):
+        streams.append(given[at + 4:at + 4 + int.from_bytes(given[at:at + 4], 'big')])
+        at += 4 + len(streams[-1])
+    size = sum(map(len, streams))
+
+    def by_h11(stream):  # h11 as a server reads each request, answers it, and goes on
+        conn = h11.Connection(h11.SERVER)
+        conn.receive_data(stream)
+        while (event := conn.next_event()) not in (h11.NEED_DATA, h11.PAUSED):
+            if isinstance(event, h11.EndOfMessage):
+                conn.send(h11.Response(status_code=200, headers=[(b'content-length', b'0')]))
+                conn.send(h11.EndOfMessage())
+                if conn.our_state is h11.DONE and conn.their_state is h11.DONE:
+                    conn.start_next_cycle()
+
+    times = {}
+    for name, read in (('python', lambda s: feed(b'', s)), ('h11', by_h11)):
+        runs = []
+        for _ in range(7):
+            start = time.perf_counter()
+            for s in streams:
+                read(s)
+            runs.append((time.perf_counter() - start) * 1e9 / size)
+        times[name] = sorted(runs)[3]
+    print('{"streams":%d,"bytes":%d,"python_ns_per_byte":%.1f,"h11_ns_per_byte":%.1f}' % (len(streams), size, times['python'], times['h11']))