From: Junio C Hamano <gitster@pobox.com>
Date: Mon, 13 Dec 2010 07:21:37 GMT
Subject: Re: [PATCH jn/fast-import-blob-access] t9300: avoid short reads from dd
Message-ID: <7vfwu25e9q.fsf@alter.siamese.dyndns.org>
In-Reply-To: <20101213063151.GB20812@burratino>

Jonathan Nieder <jrnieder@gmail.com> writes:

> dd is a thin wrapper around read(2).  As open group Issue 7 explains:
>
> 	It shall read the input one block at a time, using the specified
> 	input block size; it shall then process the block of data
> 	actually returned, which could be smaller than the requested
> 	block size.
>
> Any short read --- for example from a pipe whose capacity cannot fill
> a block --- results in that block being truncated.  As a result, the
> first cat-blob test (9300.114) fails on Mac OS X, where the pipe
> capacity is around 8 KiB.

I saw a similar breakage on my FBSD 8 bochs.  It is unfortunate and feels
yucky that we have to issue 8k+ read(2) of one byte, but I don't think of
a better way.  I thought it might be possible to specify cbs and/or conv
to have the input buffered to a size to defeat the short read issue, but
count specifies in terms of input blocks, so there doesn't seem to be a
way to do so...  Oh well...

Thanks.