diff options
author | hajoscher <hajoscher@gmail.com> | 2018-07-04 10:13:18 +0200 |
---|---|---|
committer | INADA Naoki <methane@users.noreply.github.com> | 2018-07-04 17:13:18 +0900 |
commit | 12a08c47601cadea8e7d3808502cdbcca87b2ce2 (patch) | |
tree | 18813841cb73ced22a5e12d338f241e49a84e1f1 | |
parent | 97ae32c92ecc7b3c29f8829a2b79f0f8f8bbf2cc (diff) | |
download | cpython-git-12a08c47601cadea8e7d3808502cdbcca87b2ce2.tar.gz |
bpo-34010: Fix tarfile read performance regression (GH-8020)
During buffered read, use a list followed by join instead of extending a bytes object.
This is how it was done before but changed in commit b506dc32c1a.
-rwxr-xr-x | Lib/tarfile.py | 20 | ||||
-rw-r--r-- | Misc/NEWS.d/next/Library/2018-07-04-07-36-53.bpo-34010.VNDkde.rst | 2 |
2 files changed, 13 insertions, 9 deletions
diff --git a/Lib/tarfile.py b/Lib/tarfile.py index 7b4732d471..59f044cc5a 100755 --- a/Lib/tarfile.py +++ b/Lib/tarfile.py @@ -525,7 +525,7 @@ class _Stream: if not buf: break t.append(buf) - buf = "".join(t) + buf = b"".join(t) else: buf = self._read(size) self.pos += len(buf) @@ -538,6 +538,7 @@ class _Stream: return self.__read(size) c = len(self.dbuf) + t = [self.dbuf] while c < size: buf = self.__read(self.bufsize) if not buf: @@ -546,26 +547,27 @@ class _Stream: buf = self.cmp.decompress(buf) except self.exception: raise ReadError("invalid compressed data") - self.dbuf += buf + t.append(buf) c += len(buf) - buf = self.dbuf[:size] - self.dbuf = self.dbuf[size:] - return buf + t = b"".join(t) + self.dbuf = t[size:] + return t[:size] def __read(self, size): """Return size bytes from stream. If internal buffer is empty, read another block from the stream. """ c = len(self.buf) + t = [self.buf] while c < size: buf = self.fileobj.read(self.bufsize) if not buf: break - self.buf += buf + t.append(buf) c += len(buf) - buf = self.buf[:size] - self.buf = self.buf[size:] - return buf + t = b"".join(t) + self.buf = t[size:] + return t[:size] # class _Stream class _StreamProxy(object): diff --git a/Misc/NEWS.d/next/Library/2018-07-04-07-36-53.bpo-34010.VNDkde.rst b/Misc/NEWS.d/next/Library/2018-07-04-07-36-53.bpo-34010.VNDkde.rst new file mode 100644 index 0000000000..4cb7892ee8 --- /dev/null +++ b/Misc/NEWS.d/next/Library/2018-07-04-07-36-53.bpo-34010.VNDkde.rst @@ -0,0 +1,2 @@ +Fixed a performance regression for reading streams with tarfile. The +buffered read should use a list, instead of appending to a bytes object. |