Mercurial > hg > cc > cirrus_work
view bin/warc.py @ 43:69be1131bcc5
get EOF right, finally
author | Henry S. Thompson <ht@inf.ed.ac.uk> |
---|---|
date | Wed, 05 Jul 2023 17:49:24 +0100 |
parents | 689a0e311cd2 |
children | 44d3a4f4ea51 |
line wrap: on
line source
#!/usr/bin/env python3 '''Stream a gzipped warc format file, invoking a callback on each record. Callback can be limited by WARC-Type, record part''' import sys,io from isal import igzip def warc(filename,callback,types=['response'],whole=False,parts=7,debug=False): types=[(t if isinstance(t,bytes) else bytes(t,'utf8')) for t in types] nb=0 stream=open(filename,'rb',0) bufsize=128*1024*1024 buf=bytearray(128*1024*1024) l=b'\r\n' while not stream.closed: bp=0 while l==b'\r\n': l=stream.readline() nb+=(ln:=len(l)) if ln==0: break if l!=b'WARC/1.0\r\n': raise ValueError("Not a WARC file? At %s: %s[%s]"%(nb-len(l), l.decode('latin-1'),len(l))) wtype=None length=None state=1 tr=None # Was this record truncated? while l!=b'\r\n': # WARC header if parts & 1: buf[bp:(bp:=bp+ln)]=l l=stream.readline() nb+=(ln:=len(l)) if l.startswith(b"Content-Length: "): length=wl=int(l[16:].rstrip()) elif l.startswith(b"WARC-Truncated: "): tr=l[16:].rstrip() tr="EMPTY" if tr=="" else tr elif l.startswith(b'WARC-Type: '): wtype = l[11:-2] start_2=bp if (wtype in types): if whole: buf[bp:(bp:=bp+ln)]=l elif (parts & 1): callback(wtype,buf[:start_2],1) if parts==1: start_2=0 else: start_2=bp else: start_2=0 bv=memoryview(buf)[start_2:start_2+length] ii=0 while True and not stream.closed: if (i:=stream.readinto(bv))==0: break ii+=i if ii>=length: break bv=memoryview(buf)[start_2+ii:start_2+length] if ii!=length: raise ValueError("Chunk read losing, from %s got %s expected %s"%(nb,ii,length)) nb+=length if wtype in types: if whole: callback(wtype,buf[0:start_2+length],7) continue # Only output parts (1 = WARC header, 2 = HTTP header, 4 = body) that are wanted bl=None # for HTTP Content-Length for the length of the body? L_start=start_2 state=2 bv=memoryview(buf)[start_2:start_2+length] with io.BytesIO(bv) as rec_text: for L in rec_text: if state==2: # HTTP header wl -= len(L) if not (L==b"" or L.startswith(b"\r")): # Non-empty, it's (a continuation of) a header if bl is None and L.startswith(b"Content-Length: "): bl=int(L[16:].rstrip()) else: # Blank line, HTTP header is finished if parts & 2: callback(wtype,buf[start_2:start_2+L_start],2) state=4 # The above is just for sanity, because we do _not_ # continue with the outer loop, # since we can now block-output the entire rest of the # input buffer. if bl is not None: if bl!=wl: print("length mismatch: %s %s %s here: %s given: %s trunc: %s"%\ (length,offset,filename,wl,bl,tr),file=sys.stderr) # HTTP body balance=start_2+rec_text.tell() #print(balance,bl,wl,ll,ll-balance,file=sys.stderr) # Output whatever is left if parts & 4: callback(wtype,buf[balance:balance+wl],4) state=1 L_start=rec_text.tell()