Go 1.17 introduced an unnecessary change to when a zip's data descriptor
is read for file entries, how it is parsed and how the crc32 field is
used.
Before Go 1.17, the data descriptor was read immediately after a file
entry's content. This continuous read is a pattern existing applications
have come to rely upon (for example, where reads at specific offsets
might be translated to HTTP range requests).
In Go 1.17, all data descriptors are immediately read upon opening the
file. This results in scattered and non-continuous reads of the archive,
and depending on the underlying reader, might have severe performance
implications. In addition, an additional object is now initialized for
each entry, but is mostly redundant.
Previously, the crc32 field in the data descriptor would return an error
if it did not match the central directory's entry. This check has
seemingly been unintentionally removed. If the central directory crc32
is invalid and a data descriptor is present, no error is returned.
This change reverts to the previous handling of data descriptors, before
CL 312310.
Fixes #48374
Fixes #49089
Change-Id: I5df2878c4fcc9e500064e7175f3ab9727c82f100
Reviewed-on: https://go-review.googlesource.com/c/go/+/357489
Run-TryBot: Ian Lance Taylor <iant@golang.org> Reviewed-by: Ian Lance Taylor <iant@golang.org>
Trust: Dmitri Shuralyov <dmitshur@golang.org>