diff options
| author | Miss Islington (bot) <31488909+miss-islington@users.noreply.github.com> | 2018-07-04 08:43:42 (GMT) |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2018-07-04 08:43:42 (GMT) |
| commit | d7a0ad7dd7bd7dfbdbf6be2c89fde5a71813628a (patch) | |
| tree | 6a7e1a7251cfe7b9e536fdf19891f8d0174d1752 | |
| parent | de6a2dec9c2b1280d70a29396d4e141bd1614655 (diff) | |
| download | cpython-d7a0ad7dd7bd7dfbdbf6be2c89fde5a71813628a.zip cpython-d7a0ad7dd7bd7dfbdbf6be2c89fde5a71813628a.tar.gz cpython-d7a0ad7dd7bd7dfbdbf6be2c89fde5a71813628a.tar.bz2 | |
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.
(cherry picked from commit 12a08c47601cadea8e7d3808502cdbcca87b2ce2)
Co-authored-by: hajoscher <hajoscher@gmail.com>
| -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 395b846..62d2215 100755 --- a/Lib/tarfile.py +++ b/Lib/tarfile.py @@ -534,7 +534,7 @@ 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) @@ -547,6 +547,7 @@ class _Stream: return self.__read(size) c = len(self.dbuf) + t = [self.dbuf] while c < size: buf = self.__read(self.bufsize) if not buf: @@ -555,26 +556,27 @@ 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 0000000..4cb7892 --- /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. |